Biol Psychol 2004

Biological Psychology 67 (2004) 15
Editorial
Frontal EEG asymmetry, emotion, and psychopathology: the rst, and the next 25 years
Over 25 years ago, at the annual meeting of the Society for Psychophysiological Research in Madison, Wisconsin, in September of 1978, Richie Davidson presented a paper suggesting that the experiences of positive and negative effect were associated with differential patterns asymmetrical frontal brain electrical activity. The abstract of this presentation represented the rst published account of the use of frontal electroencephalographic (EEG) asymmetry in emotional experience (Davidson et al., 1979), and there was a dearth of work on this topic in the ensuing decade. How things have changed! Research on frontal EEG asymmetry and emotion has enjoyed considerable popularity in recent years. The tabular review of this literature in Coan and Allen (2004) reveals that after this initial paper in 1979, there were ve empirical papers from 19801985, 10 more papers by 1990, another 19 by 1995, an additional 40 by 2000, and now 29 papers published or in press since 2000. Frontal EEG asymmetries have been examined as a trait individual difference that has been found to be related to other trait individual differences, that has been found to identify those with depressive and anxious psychopathology and possibly those at risk for psychopathology, and that has been found to predict subsequent emotional responses. Additionally, alterations of frontal EEG asymmetry have been found in response to tasks or emotion manipulations. In an era where functional neuroimaging methodologies, such as positron emission tomography (PET) and functional magnetic resonance imaging (fMRI), have seen an exponential increase in the study of emotion and psychopathology, one might question whether there remains a role for frontal EEG asymmetry in the study of emotion, motivation, and psychopathology. Aside from the obvious advantages of EEG as a measure that is less invasive, less expensive, and more widely available than many neuroimaging modalities, frontal EEG asymmetry has establishedby virtue of the nearly 100 studies using this measurea sizable literature that embeds the measure in a network of psychological and behavioral constructs, thus bestowing frontal EEG asymmetry with sizable construct validity as a measure of an underlying approach-related or withdrawal-related motivational style (e.g. Davidson et al., 2000; Harmon-Jones, 2004), or as an index of potential risk for emotion-related psychopathology (Coan and Allen, 2004). As such, frontal EEG asymmetry has greater construct validity as a measure of this motivational style than does any neuroimaging measure to date. On the other hand, the evidence linking frontal EEG asymmetry to the activity of underlying neural systems involved in the experience, expression, and regulation of emotion is
0301-0511/$ see front matter 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.biopsycho.2004.03.001
Editorial / Biological Psychology 67 (2004) 15
considerably lacking. When one reads the frontal EEG asymmetry literature, one might be struck by the irony that this psychophysiological measure seems much more psychological than physiological in terms of the inferences that can be drawn from the ndings to date. As one reads the papers in this special issue, it is apparent (Cacioppo, 2004) that they in large part continue in this tradition of linking frontal EEG asymmetry to important psychological constructs, including depression and related constructs (Jones et al., 2004; Minnix et al., 2004; Nitschke et al., 2004; Tomarken et al., 2004) as well as states of anger or cognitive dissonance (Harmon-Jones, 2004). As such, the psychological construct validity (see Cacioppo, 2004; Hagemann, 2004) of frontal EEG asymmetry is bolstered, suggesting it may have considerable utility in studies of motivation, emotion, and psychopathology, even in the absence of strong links to activity in underlying neural structures. If one wishes, however, to develop a theory of how neural systems give rise to emotional experience, expression, and regulation, then studies that solely link EEG asymmetry to psychological or behavioral measures will ultimately prove insufcient (Davidson, 2004). A complete account of the constructs assessed by frontal EEG asymmetry must link frontal EEG asymmetry to both psychological/behavioral constructs and also to neurophysiological constructs. Hagemann (2004) argues EEG asymmetry can be better linked to underlying neural activity using specic reference montages that highlight local activity, and Davidson (2004) suggests the merit of measuring EEG in the context of functional neuroimaging studies. Heeding such advice will bolster the neurophysiological construct validity of EEG asymmetry (Cacioppo, 2004; Hagemann, 2004), and, moreover, identify neural systems involved in emotion and psychopathology that may not generate electrical signatures which would be detectable using a rather global and spatially imprecise measure like frontal EEG asymmetry. Optimally, such work will inform and help rene the psychological theories of the constructs assessed by frontal EEG asymmetry. By identifying brain structures and systems involved in aspects of emotional experience, expression, or regulation, or those that may be related to individual differences or psychopathology, research can shape the psychological theory by drawing on what is known about such structures and systems (cf. Willingham and Dunn, 2003). But one must also be cautious in this approach, as localization per se is not necessarily informative, unless such localization efforts employ careful behavioral designs, and take advantage of extant knowledge of the functional signicance of neural systems involved in emotion. Stated more provocatively, although brain activation is indeed potentially highly informative, as an independent orespeciallymediating or intervening variable brain activation is not often informative when treated as a criterion variable (Coan, personal communication). Brain activations that differ between conditions, tasks, or individuals constitute weak evidence for a neural substrate (Cacioppo et al., 2003). Similarly, EEG asymmetry should not be mistaken for a substrate, a claim that has appeared occasionally in this literature (Sutton and Davidson, 1997; Wiedemann et al., 1999). Particularly challenging for EEG asymmetry research is the fact that a vast majority of the studies involve assessing individual differences in resting activity across a relatively long interval of several minutes, rather than tightly controlled experimental paradigms. Under such conditions, it may be especially challenging to use PET or fMRI to nd neural activity that is well localized in a consistent fashion across subjects, and inferring the meaning of such activity may be additionally vexing. It is possible that EEG asymmetry
may serve as a useful summary index of individual differences in the recruitment of underlying neuronal systems by virtue of it being a measure with such coarse spatial and temporal resolution, as it integrates information across systems and across time to provide a summary index of an underlying propensity or affective style. It may therefore be the case that establishing links between frontal EEG asymmetry and underlying neural systems may not easily be accomplished using data obtained under relatively uncontrolled resting conditions. The identication of such relationships may be enhanced, instead, by examining surface-recorded EEG and underlying neural activity in well-controlled state and task manipulations. As researchers continue to use measures of frontal EEG asymmetry, and as new researchers come to use these measures, several important methodological issues are highlighted in this special issue (Allen et al., 2004; Hagemann, 2004) that may help provide some greater standardization for this eld, thus reducing the variation across studies in terms of analysis and recording methods and procedures, which hopefully will promote a more systematic approach for the investigation of emotion, motivation, and psychopathology using measures of frontal EEG asymmetry. In this special issue readers will nd integrative reviews of the work in this area. In the rst paper, Coan and Allen (2004) selectively review the nearly 100 studies in this area, challenging researchers to think about the way in which frontal EEG asymmetry may index factors that moderate, or mediate relationships with emotion, motivation, and psychopathology. Harmon-Jones (2004) then addresses theoretical models of frontal EEG asymmetry, and with an extensive series of clever studies from his laboratory, provides evidence that favors a model that frontal EEG asymmetry taps motivational propensities along a motivational dimension of approach and withdrawal. Following these reviews are four empirical studies reporting new original research on depression. In common among these papers is the nding that the relationship between frontal EEG asymmetry and depression or risk for depression is not independent of the inuence of other important factors. Tomarken et al. (2004), with a highly valuable dataset comprised of adolescents of depressed mothers, examine whether resting frontal EEG asymmetry may serve as a marker of vulnerability for depression among adolescents, identifying also the important role of socioeconomic status as a relevant factor in this relationship. Jones et al. (2004) report that although infants of depressed mothers generally show relatively less left frontal activity and more reactive temperaments, these relationships are not apparent if the depressed mothers demonstrate a stable pattern of breastfeeding. Nitschke et al. (2004) identied that features of frontal EEG activity preceding and during encoding of sad material predict subsequent memory performance, but differently for subjects high or low in depression. Whereas overall frontal power prior to encoding predicted memory performance in low depressed controls, relatively greater right frontal activity during encoding of a sad narrative predicted better memory performance among those with higher depression scores. Finally, Minnix et al. (2004) report that the expected relationship between relatively less left frontal activity depression was moderated by the extent to which subjects report they engage in reassurance-seeking. As research in this eld moves from the rst to the next 25 years, attention to methodological and interpretive details will be paramount. Research in this eld had the luxury in the early years of using limited montages, employing single (and possibly non-optimal)
reference schemes, and analyzing data using rather simple univariate statistical models. Three papers in this special issue (Allen et al., 2004; Coan and Allen, 2004; Hagemann, 2004) challenge us to be more demanding researchers and discerning consumers of the literature on frontal EEG asymmetry, emotion, and psychopathology. The promise of frontal EEG asymmetry for informing the study of emotion and psychopathology is apparent as one reads this special issue. Yet the future of frontal EEG asymmetry research will benet from deliberately strategic use; such strategic use should further our understanding of psychological phenomenaincluding emotional experience, expression, and regulation as well as risk for psychopathologyor should serve as a bridge that can help identify the links between social, behavioral, experiential, psychological, neuroanatomical, and cellular levels of analysis (Anderson and Scott, 1999). There is sufcient work to be done using measures of frontal EEG asymmetry to occupy scientists for at least another 25 years. We wish to thank the authors whose work is featured in this issue, and also Bob Simons for providing the idea for this special issue and entrusting us with its construction. We are also indebted to the reviewers for their thoughtful critiques, especially to John Cacioppo and Richie Davidson, who provided highly thoughtful commentaries at miserably short notice. John J.B. Allen (Guest Editor) Department of Psychology University of Arizona, P.O. Box 210068 Tucson, AZ 85721-0068, USA E-mail address: jallen@u.arizona.edu (J.J.B. Allen) John P. Kline Department of Psychology University of South Alabama, USA
References
Allen, J.J.B., Coan, J.A., Nazarian, M., 2004. Issues and assumptions on the road from raw signals to metrics of frontal EEG asymmetry in emotion. Biological Psychology 67, 183218. Anderson, N.B., Scott, P.A., 1999. Making the case for psychophysiology during the era of molecular biology. Psychophysiology 36, 113. Cacioppo, J.T., Berntson, G.G., Lorig, T.S., Norris, C.J., Rickett, E., Nusbaum, H., 2003. Just because youre imaging the brain doesnt mean you can stop using your head: A primer and set of rst principles. Journal of Personality and Social Psychology 85, 650661. Cacioppo, J.T., 2004. Feelings and emotions: Roles for electrophysiological markers. Biological Psychology 67, 235243. Coan, J.A. Allen, J.J.B., 2004. EEG Asymmetry as a moderator and mediator of emotion. Biological Psychology 67, 749. Davidson, R.J., Schwartz, G.E., Saron, C., Bennett, J., Goleman, D.J., 1979. Frontal versus parietal EEG asymmetry during positive and negative affect. Psychophysiology 16, 202203. Davidson, R.J., Jackson, D.C., Kalin, N.H., 2000. Emotion, plasticity, context, and regulation: perspectives from affective neuroscience. Psychological Bulletin 126 (6), 890909. Davidson, R.J., 2004. What does the prefrontal cortex do in affect: Perspectives on frontal EEG asymmetry research. Biological Psychology 67, 219233.
Hagemann, D., 2004. Individual differences in anterior EEG-asymmetry: Methodological problems and solutions. Biological Psychology 67, 157182. Harmon-Jones, E., 2004. Contributions from research on anger and cognitive dissonance to understanding the motivational functions of asymmetrical frontal brain activity. Biological Psychology 67, 5176. Jones, N.A., McFall, B.A., Diego, M.A., 2004. Patterns of brain electrical activity in infants of depressed mothers who breastfeed and bottle feed: The mediating role of infant temperament. Biological Psychology 67, 103124. Minnix, J.A., Kline, J.P., Blackhart, G.C., Pettit, J.W., Perez, M., Joiner, T.E., 2004. Relative left frontal activity is associated with increased depression in high reassurance-seekers. Biological Psychology 67, 145155. Nitschke, J.B., Heller, W., Etienne, M.A., Miller, G.A., 2004. Prefrontal cortex activity differentiates processes affecting memory in depression. Biological Psychology 67, 125143. Sutton, S.K., Davidson, R.J., 1997. Prefrontal brain asymmetry: A biological substrate of the behavioral approach and inhibition systems. Psychological Science 8, 204210. Tomarken, A.J., Dichter, G.S., Garber, J., Simien, C., 2004. Resting frontal brain activity: Linkages to maternal depression and socioeconomic status among adolescents. Biological Psychology 67, 77102. Wiedemann, G., Pauli, P., Dengier, W., Lutzenberger, W., Birbaumer, N., Buchkremer, G., 1999. Frontal brain asymmetry as a biological substrate of emotions in patients with panic disorders. Archives of General Psychiatry 56, 7884. Willingham, D.T., Dunn, E.W., 2003. What neuroimaging and brain localization can do, cannot do, and should not do for social psychology. Journal of Personality and Social Psychology 85, 662671.
Review
Frontal EEG asymmetry as a moderator and mediator of emotion

James A. Coan , John J.B. Allen
University of Arizona, Tucson, AZ 85721-0068, USA
Abstract Frontal EEG asymmetry appears to serve as (1) an individual difference variable related to emotional responding and emotional disorders, and (2) a state-dependent concomitant of emotional responding. Such ndings, highlighted in this review, suggest that frontal EEG asymmetry may serve as both a moderator and a mediator of emotion- and motivation-related constructs. Unequivocal evidence supporting frontal EEG asymmetry as a moderator and/or mediator of emotion is lacking, as insufcient attention has been given to analyzing the frontal EEG asymmetries in terms of moderators and mediators. The present report reviews the frontal EEG asymmetry literature from the framework of moderators and mediators, and overviews data analytic strategies that would support claims of moderation and mediation. 2004 Elsevier B.V. All rights reserved.
Keywords: Frontal EEG Asymmetry; Emotion; Mediators; Moderators
1. Introduction Over 70 published studies have now examined the relationship between emotion or emotion-related constructs and asymmetries in electroencephalographic (EEG) activity over the frontal cortex. A review of these studies suggests asymmetries in frontal EEG activityincluding resting levels of activity as well as state-related activationare ubiquitous and involved in both trait predispositions to respond to emotional stimuli and changes in emotional state (Coan and Allen, 2003a).
Corresponding author. Present address: W.M. Keck Laboratory for Functional Brain Imaging and Behavior, Waisman Center, University of WisconsinMadison, Madison, WI 53705, USA. Corresponding author. Tel.: +1-608-265-6602. E-mail address: jacoan@wisc.edu (J.A. Coan).
J.A. Coan, J.J.B. Allen / Biological Psychology 67 (2004) 749
Studies of asymmetry can be classied as one of four types: (1) studies examining frontal EEG asymmetry as an individual difference that is related to other traits or trait-like measures; (2) Studies examining frontal EEG asymmetry as an individual difference that can predict state-related emotional changes and responses; (3) Studies examining frontal EEG asymmetry as an individual difference that is related to psychopathology or risk for psychopathology, especially depression and anxiety; and (4) Studies examining state-related change in asymmetry as a function of state changes in emotion. The rst three types of studies explicitly assume that frontal EEG asymmetry has trait-like properties, whereas the fourth type of study assumes that state-related changes in EEG asymmetry can be elicited and observed. Such ndings suggest the possibility that the brain systems tapped by frontal EEG asymmetries may moderate, in the case of activity, and mediate, in the case of activation, emotional responding. In particular, Davidson (e.g., Davidson, 1993) has proposed that frontal EEG asymmetries reect the activity of brain systems that moderate trait tendencies to approach, and withdraw from, novel stimuli, and mediate approach and withdrawal-motivational tendencies that underlie state emotional responding. Nevertheless, denitive analyses of frontal EEG asymmetry as a moderator and mediator of emotion1 are rare, possibly due to insufcient attention paid to the precise differences between moderators and mediators, as well as to the data analytic demands that the differentiation of moderators and mediators entail. This article reviews the large and growing literature on frontal EEG asymmetry, emphasizing its potential as an important moderator and mediator of emotional responding. Additionally, conceptual and statistical considerations in assessing the extent to which frontal EEG asymmetry functions as both a moderator and mediator of emotional responding are reviewed, and recommendations are made for future investigations.
2. Moderators, mediators and frontal EEG asymmetry Conceptually, moderators and mediators have been confused in the literature at large (Baron and Kenny, 1986). This may be particularly true of the literature on frontal EEG asymmetry, which frequently implicates moderators and mediatorssometimes both conceptually, but making use of neither term, nor the special statistical considerations that optimize the specication of moderators and mediators. Though important statistical considerations will be discussed in detail later, it will be useful to review moderators and mediators conceptually before continuing. Moderators: Moderators are essentially third variables that represent conditions under which some independent variable becomes maximally potent or effective. For example, in many people, images of poisonous snakes (indeed, any snakes) elicit fear responses, while, by contrast, more neutral images (such as, for example, images of elephants), do
1 To be precise, we would seldom be interested in whether frontal asymmetries in electrical activity per se represent moderators or mediators of emotional responding, but rather would like to know whether activity in the system(s) tapped by frontal EEG asymmetries serve as a moderators or mediators of emotional responding (see Kline et al., 2003). For brevity and ease of reading, in the present discussion, we will talk about frontal EEG asymmetry as a moderator and/or a mediator.
not. If one were to argue that resting frontal EEG activity functions as a moderator of such responses, one might draw from the literature to formulate the following argument: Since resting asymmetries characterized by greater right-than-left frontal activity seem to be associated with traits and behaviors indicative of withdrawal behaviors (e.g., Wheeler et al., 1993), and since the emotion of fear is considered by many to be an emotion with withdrawal-related motivational properties (e.g., Coan et al., 2001), one might predict that individuals who possess a greater trait tendency toward relatively greater right frontal resting EEG activity would be more sensitive to fear cues, or, at the very least, would react more strongly to fear-related cues. Using the emerging example of snakes versus elephants, such a prediction would be borne out by evidence that (1) individuals normatively respond with more fear to images of snakes than to images of elephants and (2) that individuals possessing relatively greater right frontal activity at rest would respond to images of snakes with still more fear than is otherwise normative. This type of sensible prediction has indeed been made by Davidson (1998a,b), who has referred to trait capacities to respond affectively in characteristic ways as affective style. Davidson (1998a,b) has argued that an individuals affective style is in part moderated by asymmetries in frontal cortical activity, and his model enjoys substantial, though not unequivocal, empirical support.2 Mediators: Mediators, by contrast, are third variables that represent the mechanism through which (or partially through which) the effect of a given independent variable is made manifest. For example, if one of the components of an ordinary fear experience is a motivational tendency to withdraw, then eliciting that component of fear might require activity in the brain systems tapped by frontal EEG asymmetries. This neednt necessarily mean that cortical activation asymmetries are always involved as a third variable mediator of fear. It might mean, however, that whenever fear is characterized by a withdrawal-oriented motivational component (and some would no doubt argue that this is always the case), that component is dependent upon systems that are tapped by frontal EEG asymmetries. To the extent that this is true, frontal EEG asymmetry would then function as a mediator of emotional responses.
3. The measurement of frontal EEG asymmetry Various issues surrounding the measurement and analysis of cortical EEG asymmetries can make apprehending results in this area challenging. Thus, some discussion of measurement and data analytic issues is warranted up front, though many of these issues are addressed in greater detail elsewhere in this volume (see Allen et al., this volume). First, evidence suggests that activity within the alpha range (typically 813 Hz) may be inversely related to underlying cortical processing, since decreases in alpha tend to be observed when underlying cortical systems engage in active processing. With this in mind, all results reviewed here will be reported in terms of theoretically assumed cortical activity rather than alpha power per se.
2 An equally plausible prediction, though one not as frequently or explicitly encountered in the research literature, is that the effects of frontal EEG asymmetry on, for example, depression, are themselves moderated by other psychological traits (for a discussion of this possibility, see Minnix et al., this volume).
10
Second, in reviewing the literature, one will variously encounter references to frontal EEG activity and frontal EEG activation. At rst blush, insistence upon this distinction may appear pedantic, but in fact, strict attention here can signicantly enhance understanding of research in this area, and is relevant to making inferences concerning the distinction between frontal EEG asymmetry as a moderator or a mediator. For the purposes of this article, and as a general recommendation, activity will refer to a tonic recording of cortical processes as measured by EEG, while activation will refer to the change in EEG activity in response to a provocation, such as the presentation of an emotional stimulus. For example, one may be interested in measuring an individuals asymmetry in frontal EEG activity at rest as well as that same individuals asymmetrical frontal EEG activation in response to an experimental manipulation, such as the presentation of an image of a venomous snake. While this implies that activity refers only to resting or baseline measures, this is not necessarily the case. Indeed, one could measure an individuals resting activity both at baseline and following a stimulus presentation. The difference between those post-stimulus and baseline activity measures, however, would represent that individuals inferred activation in response to the stimulus. This distinction is important for maintaining conceptual clarity when reviewing the literature on frontal EEG asymmetry. Thus, in reviewing the literature, one of the functions of this article will be to make clear this distinction in past literature as well as in the arguments and data presented here. Finally, an occasional difculty in reviewing the literature on frontal EEG asymmetry involves discerning precisely what specic hemispheric effects are responsible for observed asymmetries. This difculty frequently results from the widespread practice of computing and analyzing an asymmetry index, typically a difference score, rather than analyzing hemisphere as a two-level factor. The most commonly reported of these indexes is computed by subtracting the natural log of left hemisphere alpha power from the natural log of right hemisphere alpha power (ln[right alpha] ln[left alpha]). This approach results in a unidimensional scale representing the relative activity of the right and left hemispheres, with the middle point of the scale equaling zero or symmetrical activity. In interpreting this scale, higher scores indicate relatively greater left frontal activity whereas lower scores indicate relatively greater right frontal activity (again keeping in mind that higher scores result from relatively greater right frontal alpha powerthe putative inverse of activity). A limitation of this particular metricand other similar metricsis that the metric provides no information regarding the extent to which each hemisphere is contributing to the observed difference score (e.g., Jones et al., 1998). It is not necessarily the case, however, that the use of asymmetry metrics is either unwarranted or particularly problematic. Asymmetry scores have some distinct advantages, not least of which is their capacity for controlling individual differences in skull thickness that might produce artifactual and non-neurogenic individual differences in recorded power values. Asymmetry metrics can also make statistical tests more sensitive by reducing the number of contrasts in a particular model and increasing statistical power. When such difference score based asymmetry effects are followed up by specic hemisphere analyses, such an approach is not only warranted but may be the most efcient data analytic approach to these kinds of data (cf. Coan and Allen, 2003b). Asymmetry scores also conceptually simplify certain analyses, such as those involving correlations between frontal asymmetries (as a difference score) and other individual difference measures (e.g., behavioral activation;
11
Coan and Allen, 2003b; Harmon-Jones and Allen, 1997; Sutton and Davidson, 1997). Additionally, difference scores based on alpha power asymmetries tend to show high internal consistency and acceptable test-retest reliability, dispelling fears about reduced reliability attributable to difference scores per se (Allen et al., this issue; Jones et al., 1997; Tomarken et al., 1992b). Nevertheless, reports of specic hemispheric effects will be important for the ultimate understanding of the precise nature of cortical asymmetries in emotional responding. Thus, tables presented in this article include a column labeled Hem that species whether a given report investigated specic hemispheric effects (yes) or used an asymmetry metric only (no).
4. What frontal EEG asymmetry measures: evidence from its associations with other traits Resting measures of asymmetrical frontal EEG activity have been associated with other trait measures. Such relationships suggest the functional properties of the underlying cortical systems that give rise to frontal EEG asymmetry. Findings from numerous studies reect an emerging consensus that relatively greater trait left frontal activity is associated with trait tendencies toward a general appetitive, approach, or behavioral activation motivational system, and that relatively greater trait right frontal activity is associated with trait tendencies toward a general avoidance or withdrawal system (Coan and Allen, 2003a; Davidson, 1993). This has resulted in Davidsons highly inuential approach/withdrawal motivational model of emotion (Davidson, 1993; Davidson, 1998a,b). According to this model, left frontal activity, either as a state or a trait, indicates a propensity to approach or engage a stimulus, while relatively greater right frontal activity indicates a propensity to withdraw or disengage from a stimulus. A comprehensive tabular summary of the literature associating trait frontal EEG asymmetry with other trait measures can be found in Table 1. Though no attempt to review every report in this literature will be made within the body of this article, a number of particularly noteworthy studies will be highlighted in the remainder of this section. An obvious paper and pencil measure for assessing Davidsons approach-withdrawal model of frontal EEG asymmetry in emotion might be Carver and Whites (1994) BIS/BAS scales, which purport to measure Grays (1972, 1987) behavioral inhibition and activation systems (BIS and BAS, respectively) as human traits. According to Gray, the BIS initially inhibits action, increases arousal and attention and subsequently guides behavior toward removing or avoiding an undesirable stimulus. The BAS essentially functions in the opposite manner, responding to incentives, and guiding organisms toward attaining a desirable stimulus, including negative reinforcers. Researchers have identied relationships between these systems and frontal EEG asymmetry (Coan and Allen, 2003b; Harmon-Jones and Allen, 1997; Sutton and Davidson, 1997). In particular, Sutton and Davidson (1997) proposed that the BIS and BAS should map closely onto withdrawal and approach tendencies, respectively, arguing that these constructs are functionally identical. They found that relatively greater left frontal activity was associated with both higher BAS scores and greater BAS-BIS difference scores. Further, they found that relatively greater right frontal activity was associated with higher BIS scores (Sutton and Davidson, 1997). Work by Harmon-Jones and Allen (1997) and Coan and Allen (2003b), however, suggest that the relationship is
Table 1 Trait frontal EEG asymmetry and other trait-like measures

Citation N Age Handedness Reference scheme AR, Cz Independent variable Dependent variable Cortisol level (C), fear (F) and sadness (S) ratings Hem Results summary ER, C ER and Str., F ER and Str., S
12
Buss et al., 2003
From 85 infants samples sizes for specic analyses vary
6 months
NA
Coan, 2003
250 (132 female)
Mean = 19 years
No information
Cz, LM
Extreme right (ER), intermediate (I) and extreme Left (EL) groups from EEG (59 Hz) at FP1/2, F3/4, F7/8; baseline (B) vs. stranger (Str.) groups Sex (M vs. F); positive emotionality (PEM) and negative emotionality (NEM) and others from the multidimensional personality questionnaire (MPQ); monozygotic twins (MT) and dizygotic twins (DT); twins used to assess heritability (h2 )
Yes
EEG at F3/4;
No
In M, no signicant ndings In F, RFA, PEM In F, RFA, NEM In F, h2 of F3/4 0.22 In F, h2 of bivariate phenotypic correlation between F3/4 and NEM 0.42 LFA, BAS
Coan and Allen, 2003 Davidson et al., 1999
32 (26 female)
1724 years
AR, LM
24 (9 female)
1721 years
LM
EEG at FP1/2, F3/4, F7/8, FTC1/2, C3/4, T3/4, TCP1/2, T5/6, P3/4, O12 EEG at F3/4, F7/8 and T3/4
BIS/BAS
Yes
Natural killer (NK) cell activity at rest, before exam, and following pos and neg lm clips EEG at F3/4, C3/4, P3/4, O1/2
No
RFA, NK (rest) RFA, NK (exam) LFA, NK (pos lm clip)
Ehlers et al., 2001
134 (68 female)
713 years
Both
No information
Handedness (L vs. R); strongly native American (SNA) vs. native American (NA); sex (M vs. F); positive family history of alcohol abuse (FHP) vs. no such history (FHN)
No
SNA, LFA
Field et al., 2002
48 infants (29 female)
newborn
NA
Cz
Infant EEG at F3/4, P3/4
Fox et al., 1995
48 (28 female)
4962 months Mean = 24.7
No information R
Cz
EEG at F3/4, P3/4 and O1/2
Maternal serotonin (MS), maternal postnatal cortisol (MPC), maternal EEG at F3/4, P3/4, maternal vagal tone (VT), infant cortisol (IC), infant sleep state changes (SS), infant Brazelton assessment (BA) Social competence (SC) EEG at F3/4, T3/4, C3/4, P3/4 A1/2 BIS/BAS Dispositional anger (A) PANAS (PA and NA); BDI EEG at FPF1/2, FP1/2, F3/4, F7/8, FC3/4, FC/8, C3/4, CP3/4, CP5/6, T3/4, T5/6, P3/4, PO3/4
No
infant RFA, MS infant RFA, MPC infant RFA, maternal RFA infant RFA, maternal VT infant RFA, IC infant RFA, BA
Yes
RFA, SC LFA, SC NA, LATA
Hagemann et al., 1999
36 (24 female)
Cz
Positive affectivity (PA), negative affectivity (NA), extroversion (E), and neuroticism (N) EEG at F3/4 and P3/4 EEG at FP1/2, F3/4, C3/4, P3/4, T5/6, T3/4, O1/2 EEG at F3/4, P3/4 Unpleasant (U), pleasant (P) and neutral (N) pictures; startle probe during early viewing (A), later during viewing (B), and after viewing (C)
Yes
Harmon-Jones and Allen, 1997 Harmon-Jones and Allen, 1998 Jacobs and Snyder, 1996 Jackson et al., 2003
37 females 26 (11 female) 40 males 47 (30 female)
No information Mean = 13 years 1853 years 5760 years
R R R R
Cz Cz LM LE
No No No No
LFA, BAS LFA, A LFA, NA score LFA, BDI score RFA, C negativity
13
14
Table 1 (Continued )
Citation Jones et al., 1997 N 87 (infants) Age No information Handedness No information Reference scheme Cz Independent variable Baby groups: overstimulating (O) vs. understimulating (U) mothers Dependent variable EEG at F3/4 and P3/4, various physio and beh. measures Natural killer (NK) cell, lymphocyte and T-cell activity Cerebrospinal uid CRH Hem Yes Results summary O babies, LFA U babies, RFA (Mothers showed the same pattern as infants) RFA group, NK activity
Kang et al., 1991
20 females
1720 years
Cz, LM
Extreme LFA and RFA groups
No
Kalin et al., 2000
49 (23 female; rhesus monkeys)
Longitudinal data at 4, 8, 14, 40 and 52 months 1733 years
No information
No information
Extreme LFA and RFA groups in monkeys
No
RFA, CRH
Kline et al., 1998
85 (60 females)
Linked ears (LE)
Defensive coping (EPQ-L scale)
EEG at F3/4, FP1/2, F7/8, C3/4, T3/4, T5/6, P3/4, O1/2 EEG at FP1/2, T3/4, F7/8, and F3/4 EEG at F3/4, FP1/2, F7/8, C3/4, T3/4, T5/6, P3/4, O1/2 EEG at F3/4, P3/4, A1/2
Yes
For women, LFA defensiveness For men, LFA defensiveness defensiveness, LFA HD, LFA in presence of opposite sex. HF and HR, RFA
Kline et al., in press-a, in press-b Kline et al., in press-a, in press-b McManis et al., 2002
72 (42 female)
Mean = 20.4 Mean = 20.4
No information No information
Defensiveness (D), parental caring (PC) High (HD) vs. low (LD) defensiveness groups; experimenter gender: same vs. opposite Low fear (LF), moderate fear (MF) and high fear (HF) at 2 years old; low reactivity (LR) vs. high reactivity (HR) at 2 years old Cultural group (J vs. W)
Yes
235 (141 females)
Yes
166 children
1012 years
LM
No
Moss et al., 1985
12 (Japanese), 12 (Western), all Female
J, mean = 32.6 years; W, mean = 29.1 years
Cz
EEG at T3/4 and P3/4
Yes
W = LPA
Merckelbach et al., 1996 Schmidt, 1999
29 females 40 females (extreme scorers selected from among 271) 40 females (extreme scorers selected from among 282) 16 (11 females)
2238 years M = 20.97 years
No information R
A1 Cz
L vs. R hemisphere preference (questionnaire) Low shy vs. high shy; low soc vs. high soc
EEG at F3/4 and P3/4 EEG at F3/4, P3/4 and O1/2
No Yes
LHP, LFA shyness, RFA soc, LFA High shy, high soc had LFA than high shy, low soc soc, RFA Low shy, high soc = RPA Low shy, low soc = LPA No effects in alpha band.
Schmidt and Fox, 1994
No information
Cz
Low shy vs. high shy; low soc vs. high soc
EEG at F3/4, P3/4, A1/2 and O1/2
Yes
Stough et al., 2001
2030 years
No information
No information
Measures of openness (O), agreeableness (A) and conscientiousness (C) EEG at F3/4 7 and P3/4
EEG at FP1/2, F3/4, F7/8, T3/4, T5/6, C3/4, O1/2. BIS/BAS, BASBIS diff score EEG at F3/4, F7/8, T3/4, C3/4 and P3/4 General positive and negative affect (PA and NA) Eudaimonic well-being (EWB), hedonic well-being (HWB)
Yes
Sutton and Davidson, 1997
46 (23 female)
1822 years
LM
No
LFA, BAS RFA, BIS LFA, BAS-BIS diff HD = LFA in F3/4 and F7/8 LFA, PA LFA, NA
Tomarken and Davidson, 1994 Tomarken et al., 1992a
90 females
No information 1721 years
Cz
High defensive (HD) vs. low defensive (LD) EEG at F3/4, F7/8, T3/4, P3/4, C3/4
Yes
90 females
Cz, LM
No
Urry et al., in press
84
5760 years
AR, LM
EEG at 29 sites, including FPF1/2, FC3/4, FP1/2, F7/8, F3/4, and FC7/8
LFA in FC3/4, EWB LFA in FC3/4, HWB
RFA: right frontal activity, LFA: left frontal activity; RATA: right anterior temporal activity, LATA: left anterior temporal activity; RPA: right parietal activity, LPA: left parietal activity.
15
16
robust for BAS, but not so robust for BIS. While both Harmon-Jones and Allen and Coan and Allen found associations between relative left frontal activity and higher BAS scores, neither study detected a strong association between relative right frontal activity and BIS scores (though Coan and Allen did detect a statistical trend suggestive of this relationship). In both reports, it was suggested that the theoretical association between withdrawal motivations and the BIS is more complex than that between approach motivations and the BAS (Coan and Allen, 2003b; Harmon-Jones and Allen, 1997). Theoretically, Davidsons (1998a) withdrawal construct may be more general than that of the BIS. For example, the hypothetical systems underlying Davidsons withdrawal construct are thought to motivate or predispose organisms to withdraw from sources of aversive stimulation, whereas the BIS has been described as a system that, among other things, interrupts behavior, increases arousal and increases attention (Gray, 1994), none of which are necessarily involved in withdrawal behaviors or predispositions. Indeed, Davidsons withdrawal construct could overlap with functions included in any of the motivational systems described by Gray (for a more detailed discussion, see Coan and Allen, 2003b). By contrast, Davidsons approach and Grays BAS constructs (as well as the similar construct of the behavioral facilitation system; Depue and Collins, 1999; Depue and Iacono, 1989) may share substantial overlap. The rather robust relationship between frontal EEG asymmetry and the BAS is further bolstered from ndings designed to test the primarily motivational approach-withdrawal model of frontal asymmetry against a valence model (cf. Heller and Nitschke, 1997). For example, Harmon-Jones and Allen (1998), and more recently Harmon-Jones (2000), found that left frontal activity was associated with trait anger, a negatively-valenced but approach-related emotion conceivably related to BAS functions. Indeed, in at least one study, BAS sensitivity as measured by Carver and Whites (1994) BAS scale was associated with a greater likelihood of aggressive behavior (Wingrove and Bond, 1998). While the relationship between the self-reported BIS scores and trait withdrawal propensities remains uncertain, other evidence that trait frontal EEG asymmetries are related to both behavioral propensities to approach and withdraw is available. One source of this evidence derives from measures of social behavior. For example, Fox et al. (1995) found evidence that children with greater relative right frontal activity at rest were generally more inhibited socially, and scored lower on measures of social competency (Fox et al., 1995). Moreover, children were more sociable and more socially competent to the extent they had relatively greater left frontal activity. Schmidt and colleagues (Schmidt, 1999; Schmidt and Fox, 1994) investigated the relationship between frontal EEG asymmetry and measures of sociability in adults and found that individuals scoring low on measures of sociability showed relatively greater right frontal activity at rest (Schmidt and Fox, 1994). Further, Schmidt (1999) found that shyness was associated with relatively greater right frontal activity, while sociability was associated with relatively greater left frontal activity. Interestingly, Schmidt (1999) also found that shy individuals who nevertheless scored high on measures of sociability possessed greater left frontal activity than other shy individuals with low sociability scores. More recently, Kalin et al. (2000) have investigated the relationship between frontal EEG asymmetry and other physiological traits that may underlie processes related to those reviewed above. For example, they have reported a positive relationship between extreme right frontal EEG activity and high cerebrospinal uid concentrations of corticotrophin-releasing hormone (CRH) in rhesus monkeys (Kalin et al., 2000). CRH is
17
itself thought to mediate stress responses, as well as fear, anxiety and depression. According to this work, higher CRH levels are associated with higher levels of stress (De Souza, 1995; Kalin et al., 2000). Further, Kalin and colleagues have supplied evidence that rhesus monkeys showing extreme relative right frontal EEG activity have both elevated cortisol levels and exaggerated defensive behaviors (Kline et al., 1998).
5. Frontal EEG asymmetry as a possible moderator of emotional responding When considering how resting frontal EEG asymmetry may serve as a predictor of subsequent emotional responses, two non-mutually exclusive relationships are possible. 1. Resting frontal EEG asymmetry taps an individual difference that may facilitate or diminish an emotional response (e.g., happiness) across many classes of stimuli (e.g., happy, sad, angry, fearful, and disgusting stimuli). For example, greater relative left frontal activity might be expected to be associated with greater self-reported happiness, for positively- as well as negatively-valenced stimuli. 2. Resting frontal EEG asymmetry taps an individual difference that may facilitate or diminish emotional responses preferentially for some but not other classes of stimuli. For example, greater relative right frontal activity might be expected to potentiate startle in response to negatively-valenced stimuli, but not to positively-valenced or neutral stimuli. Strictly speaking, it is only the second case that qualies as a moderator variable (cf. Baron and Kenny, 1986), as will be detailed below. The rst instance is nonetheless a theoretically useful and interesting nding that frontal EEG asymmetry may serve as an individual difference variable that predicts emotional responses. Davidson (1998a) has proposed that trait EEG asymmetries index propensities for reacting in predictable ways to emotionally evocative stimuli. Davidson (1998a) has called this propensity affective style, and he proposes that frontal EEG asymmetry indexes a system that may have emotion-specic or valence-specic (p. 309) moderating inuences, with implications for risk for psychopathology. The intent of the following section is therefore to review the literature suggestive of frontal EEG asymmetry as a possible moderator of emotional responding. The question of whether and to what extent this evidence unequivocally implicates frontal EEG asymmetry as such a moderator, however, depends upon a deeper consideration of what is meant, not only conceptually but also statistically, by the term moderator. 5.1. Modeling frontal EEG asymmetry as a moderator Before describing the data analytic identication of a moderator variable, it is important to understand precisely what a moderator variable should look like in terms of statistical effects. Fig. 1 depicts a general moderator model that will serve as a useful guide to such a discussion. For the purposes of the illustration, assume that frontal EEG asymmetry may moderate the intensity rating of a fear experience following the stimulus presentation of an image of a poisonous snake versus some more affectively neutral image, such as that of an
18
Stimulus Type (snake vs. elephant) a
Moderator (EEG Asymmetry)
Emotional Response
c Stimulus Type X Moderator (EEG Asymmetry)

Fig. 1. Frontal EEG asymmetry as a moderator. Model adapted from Baron and Kenney (1986).
elephant. In this example, we might expect fear ratings in response to images of a poisonous snake to be relatively high and those to images of an elephant to be relatively low, with ratings in response to one, the other, or both images varying as a function of frontal EEG asymmetry. According to this model, there are three causal paths to the emotional responsethe rating of fear intensity according to our example. The rst of these paths, path a, represents the causal inuence of the independent variable, in this case, image type (poisonous snake verses an elephant). Path b would represent the unique contribution of trait frontal EEG activity. Path c represents the causal inuence of the interaction between the type of image presented and an individuals trait pattern of resting frontal EEG asymmetry. At its simplest, the moderator hypothesis states that path c is signicant. Baron and Kenny (1986) note, however, that a signicant contribution of this path alone does not necessarily present a strong case for a moderator effect. While a main effect of the independent variable is possible in this model, and in the case of this example even likely, it is preferred that the moderator variable be uncorrelated with both the stimulus and emotional response variables, if the moderator is specically to be identied as a moderator only (i.e. not simultaneously a moderator and mediator). It is easy to see that a correlation would not exist between frontal EEG asymmetry and the likelihood of encountering an image of a poisonous snake versus an elephant (at least in the experimental context). Less certain, however, is that frontal EEG asymmetry is unrelated to continuous measures of affective intensity, regardless of the particular stimulus being rated. If frontal EEG activity predicted fear intensity ratings independent of the eliciting stimulus, it could be somewhat difcult to disambiguate the moderating inuence (path c) of frontal EEG asymmetry on our specic construct of interestemotion responsivityfrom its main effect (path b) of predicting intensity. Such a condition would not preclude the identication of a moderator, however. A statistical method for dealing with this problem is outlined below. Moderator models of frontal EEG asymmetry assume that frontal EEG asymmetry itself is most likely a continuous variable. This is true, even if hemispheres act independently of each other to achieve their moderating effects. Raw EEG is always measured as a continuous variable, and while frontal EEG asymmetry can be articially dichotomized into categorical
19
variable, such an approach is not recommended, in part because doing so reduces statistical power and may result in articially obtained groups that distort the underlying continuous relationships. Following from Davidsons diathesis/stress theory of psychopathology and affective style, it makes the most sense to analyze frontal EEG asymmetry as a continuous moderator, and the following examples adhere to this. This data analytic approach is perhaps best illustrated with the snake/fear example, because the diathesis/stress theory models stress as an environmental event beyond an individuals direct control. Recall that this example assumes that any given individuals trait pattern of frontal EEG asymmetry will serve to either attenuate or amplify a fear response to the visual stimuli. Baron and Kenney (1986) present three possible models representing the interaction between an independent variable (e.g., snake versus elephant) and a continuous moderator variable (e.g., resting frontal EEG asymmetry). These are (1) the linear model, (2) the non-linear model, and (3) the stepfunction model. In the linear model, one would expect to nd a gradual, linear change in the effect of the independent variable on the outcome criterion as a function of the moderator. In terms of the example, the relationship between the interaction term (the snake/elephant by EEG asymmetry interaction) and fear ratings would be gradual and steady. That is, frontal EEG asymmetry should both attenuate and amplify the effect of the fear-relevant (but not fear-irrelevant) stimulus, depending on the relative difference between the left and right hemispheres. This model would then predict that individuals with relatively greater right frontal EEG asymmetry would be expected to show an amplied fear response, and individuals with relatively greater left frontal activity should show an equal and opposite attenuation in their fear response, with these effects specic to the snake stimulus. In the non-linear model, a quadratic equation would best exemplify the relationship between the interaction term and the outcome criterion. Following from the example, it might be expected that relative right frontal activity would correspond with a more intense fear rating in response to the snake image presentation, while nding that relative left frontal activity was not different from symmetrical activity in terms of inuencing fear ratings,3 with effects again specic to the snake and not the elephant stimulus. In the stepfunction model, some critical level of the moderator variable is assumed. In terms of the example, this model would state that fear responses are at a clearly dened normative level until a particular threshold in relative right frontal EEG asymmetry is crossed, the result of which would be a large jump in the intensity ratings of fear in response to the snake but not elephant image presentation. There is no evidence to suggest such a model is likely. In data analytic terms, each of these models can be tested with simple regression equations per the recommendations of Cohen and Cohen (1983) and Aiken and West (1986). This recommendation holds if the independent variable is continuous as well. Further, standard
3 The potentially independent contributions of the left and right frontal regions to the observed interaction would be important to examine in subsequent analyses. Although parsimonious to assume relatively comparable magnitude of effects from the left and right hemisphere in the linear model, this need not be the case, as a single hemisphere could be responsible for a majority of the effect, with the other hemisphere contributing little. In the case of the nonlinear model, it obviously cannot be assumed that both hemispheres contribute equally across the whole range of asymmetry scores, and underscores the desirability of follow-up analyses involving hemisphere (cf. Allen et al., this issue).
20
general linear model (GLM) approaches are essentially equivalent to regression approaches, and their exibility makes them in many cases more desirable, especially with regard to certain common statistical packages, such as SPSS, etc. In cases where the moderator variable is independently correlated with the criterion variable, it may be desirable to calculate ones general linear model using a type 1 sum of squares in order to control for multicolinearity with the moderator by independent variable interaction term. This is easily accomplished in most popular statistical packages. 5.2. Empirical support for frontal EEG asymmetry as a moderator Thumbnail reviews of a large number of studies relevant to this section can be found in Table 2. Early studies suggest that frontal EEG asymmetry may serve as a moderator involved in predictions of emotional reactions in infants. For example, Davidson and Fox (1989) found that infants who cried in response to maternal separation had greater right frontal activity at rest than those who did not. This result was replicated by Fox et al. (1992), who also found that this effect was modestly stable over 5 months. Similar ndings have been obtained in adults. For example, when asked by Tomarken et al. (1990) to report affective responses to emotional lm clips, individuals with greater right frontal activity at rest responded with more intense levels of negative affect to negativelyvalenced lm clips, particularly those involving fear (Tomarken et al., 1990). In a similar report, Wheeler et al. (1993) found that individuals with greater right frontal activity responded with more intense negative affect to negatively valenced lms, and that individuals with greater left frontal activity responded with more intense positive affect to positively valenced lms. These studies suggest that individual differences in emotional responding are in part a function of individual differences in trait frontal EEG asymmetry. Indeed, Davidson (1998a) has cited these ndings to argue that ones affective styleas indicated by ones trait frontal EEG asymmetrymay in part determine ones risk for certain affective disorders such as depression and anxiety, a proposal that clearly identies frontal EEG asymmetry as a moderator of emotion and related affective processes. While neither of these studies explicitly assessed the moderator hypothesis in terms of the recommendations of Baron and Kenny (1986), each adopted a similar approach. Each study can be conceptualized in terms of Fig. 1 as follows: (a) the stimulus type is the lm type (positive or negative); (b) resting frontal EEG asymmetry is the potential moderator; and (c) intensity rating would be the emotional response. In the end, both Wheeler et al. (1993) and Tomarken et al. (1990) provide compellingthough not denitiveevidence that resting frontal EEG asymmetry is a genuine moderator of emotional responding. Recently, Henderson et al. (2001) explicitly tested frontal EEG asymmetry recorded at 9 months of age as a moderator of the relationship between negative affectivity, also measured at 9 months, and social wariness at approximately 4 years of age. In this work, 9-month-old negative emotionality predicted social wariness at 4-year follow-up in infants whod shown relatively greater right frontal EEG activity at 9 months. No relationship was found between social wariness behaviors at follow-up and infant frontal EEG asymmetries favoring the left. It is noteworthy that these results are consistent with Baron and Kennys (1989) non-linear moderator model. More recently, Coan and Allen (2003c) conducted a moderator analysis of resting frontal EEG asymmetry and emotional experience using a
Table 2 Trait frontal asymmetry as a predictor of state dependent changes

Citation Davidson and Fox, 1989 Fox et al., 1992 N 13 infant females Age 10 months Handedness R (parents) Reference scheme Cz Independent variable Dependent variable EEG at F3/4, P3/4 Infant response to maternal separation (crying vs. not-crying) Infant response to maternal separation (crying vs. not-crying) Positive affect (PA), negative affect (NA), affective bias (AB), and generalized reactivity (GR), all in response to affective slides Hem Results summary
Yes
Criers, RFA non-criers, LFA
(1) 33 infants (17 female), (2) 13 infants (7 females)
(1) 1424 months, cross-sectional, (2) 712 months, longitud-inal 1944 years
R/L
Cz
EEG at F3/4, P3/4
Yes
Criers, RFA effects consistent over time
Hagemann et al., 37 (22 female) 1998
Cz, LM
EEG at F3/4, T3/4, C3.4, P3/4, A1/2
No
Cz, 8 min resting: LFA, PA Cz, 4 min eyes closed: LFA, PA LM reference: no effects R T3/4, NA Note: overall results equivocal with regard to A/W model RFA and NR, SW
Henderson et al., 97 infants (51 2001 females)
Longitud-inal; time 1, 9 months; time 2, 48 months 1741 years
No information AR
EEG at F3/4, C3/4, P3/4 and O1/2; negative reactivity (NR) EEG at F3/4, T3/4, P3/4, C3/4 EEG at F3/4, T3/4, P3/4, C3/4
Sociability (S) and social No wariness (SW)
Tomarken et al., 1990 Wheeler et al., 1993
32 females
Cz
No Reported positive affect (PA) and negative affect (NA) following lm clips Reported positive affect No (PA) and negative affect (NA) following lm clips
RFA, NA RFA, PA-NA difference RFA, fear report LFA, PA RFA, NA
26 females with 1721 years stable asymmetry across sessions
Cz
21
22
general linear model approach similar to that used by Henderson et al. (2001). Using data previously reported on in Coan et al. (2001); Coan and Allen (2003c) sought to assess the extent to which trait frontal EEG asymmetries moderated self-reported emotional experience in response to posed emotional facial expressions that depicted anger, disgust, fear, joy and sadness. To explicitly test a moderator hypothesis, Coan and Allen (2003c) constructed a single general linear model where emotion type, trait frontal EEG asymmetry, and an emotion by trait EEG asymmetry interaction were used to predict emotional experience. This model was able to accommodate both the categorical (emotion) and continuous (trait frontal EEG asymmetry) predictors, as well as their interaction, easily. Additionally, these predictors were interacted with the other categorical factors of reference scheme (average versus linked mastoid) and particular frontal region (F3/4, F7/8 and FTC1/2) to assess the degree to which these additional factors inuenced the effects of interest. If frontal EEG asymmetry moderates emotional experience and intensity, then this model would result in a signicant emotion by frontal EEG asymmetry interaction. Coan and Allen (2003c) found that relatively greater left frontal activity at rest predicted an increased likelihood of reporting an experience of anger, joy, disgust, but not sadness and fear (trait EEG asymmetry by emotion interaction, F(4, 865) = 3.53, P < 0.01) independently of reference scheme (average and linked mastoid) or specic frontal region (F7/8, F3/4 and FTC1/2). This moderator effect was, however, only partially consistent with predictions of the approach/withdrawal model of frontal brain asymmetry. Positive correlations between frontal EEG asymmetry and experience were found, as predicted, in the cases of joy and anger, but the predicted negative relationships between trait frontal EEG asymmetry and disgust, in addition to fear and sadness, were not. These ndings may in part be due to methodological idiosyncrasies such as the use of voluntary emotional facial expressions as emotional stimuli, but suggest that the role of frontal EEG asymmetry as a moderator of emotional response may only partially support of the approach/withdrawal model. Other studies suggest the same. Using normed emotion eliciting images, Hagemann et al. (1998) essentially failed to replicate earlier ndings. In their study, Hagemann et al. (1998) did indeed nd that individuals with relatively greater left frontal activity at rest tended to respond more positively to positively valenced images, but only when using a Cz reference (arguably the most problematic of all reference montages; see Hagemann et al., 2001). The effect did not generalize to frontal EEG recorded using a linked mastoids reference montage. Other results reported by Hagemann et al. (1998) were generally inconsistent with the approach/withdrawal model advocated by Davidson and colleagues. For example, Hagemann et al. (1998) found that relatively greater left anterior temporal activity at rest was associated with more intense experience reports associated with negative affect, whereas the approach/withdrawal model would predict precisely the opposite. Such discrepant ndings have prompted Davidson (1998b) to suggest that that Hagemann et al.s (1998) study was methodologically inconsistent with earlier ones supportive of the approach-withdrawal model, pointing out that in Wheeler et al.s (1993) study, only data from individuals with highly stable frontal EEG asymmetry patterns (across 3 weeks of measurement) were analyzed. Both authors have noted the need to resolve these inconsistencies with further research. Inconsistencies aside, if the possibility of frontal EEG activity as a moderator of emotional responses is ultimately borne out, there are likely to be consequences for affectively
23
based psychological disorders. Indeed, Davidsons theory of frontal EEG asymmetry as a partial determinant of affective style proposes explicitly that certain affective styles serve as diatheses that, in response to the appropriate stimuli, can increase risk for certain forms of psychopathology. Thus, evidence that frontal EEG asymmetry may moderate risk for psychopathology merits consideration.
6. Frontal EEG asymmetry as a risk factor and possible moderator of psychopathology Resting frontal EEG asymmetry may serve as an index of risk for a variety of emotionrelated disorders, including depression and anxiety. Whether it indexes a general propensity to respond in ways consistent with depression or anxiety, or a specic tendency to do so in some emotionally evocative situations and not others, remains an empirical question. Yet evidence of frontal EEG asymmetrys role in moderating emotional responsivity suggests a similar role in the development of various disorders involving affect. Researchers have identied a link between relatively greater right frontal resting activity or relatively lower left frontal resting activityand depression (e.g., Henriques and Davidson, 1990, 1991; Schaffer et al., 1983) including seasonal depression (Allen et al., 1993). In the rst report of this type, Schaffer et al. (1983) found that higher scores on the Beck Depression Inventory (BDI) were associated with relatively greater right frontal resting activity in their sample of undergraduate research participants (see Table 3). Subsequent studies have identied the same relationship in clinically diagnosed participants (Allen et al., 1993; Gotlib et al., 1998; Henriques and Davidson, 1991). In some of these studies, participant samples have included euthymic individuals who have suffered previous bouts of depression (e.g., Gotlib et al., 1998; Henriques and Davidson, 1990). Some researchers, however, have failed to replicate the relationship between resting frontal EEG asymmetry and depression (Reid et al., 1998). For example, despite the use of two separate and large samples, Reid et al. (1998) were unable to detect such a relationship. In interpreting the ndings, Reid et al. (1998) suggested that given the heterogeneity of depression, frontal EEG asymmetry might tap only one of several risk trajectories associated with depression. Reid et al. also speculated that some unidentied qualities of their research environment may have been to blame, noting that traits may interact in particular ways within particular environments and that these interactions may have masked preexisting asymmetries in their research participants. Evidence that infants of depressed mothers show relatively greater right frontal resting activity further suggests a link between frontal EEG asymmetry and depression (e.g., Dawson et al., 1999a; Field et al., 1995). For example, Dawson et al. (1997) found that infants of depressed mothers showed less left frontal activity than those of non-depressed mothers and that such lower left frontal activity discriminated infants whose mothers were diagnosed with major depression from those with mothers whose symptoms were sub-threshold. More recently, Dawson and colleagues have found that infants of depressed mothers are less affectionate and show evidence of relatively less left frontal activity not only at rest, but also while interacting with their mothers and while interacting with familiar strangers (Dawson et al., 1999b). In independent investigations, Field et al. (1995) have achieved similar ef-
24
Table 3 Trait frontal EEG asymmetry and measures of psychopathology

Citation Allen et al., 1993 Allen et al., in press N 8 females (4 with seasonal affective disorder; SAD) 30 depressed females Age No information Handedness R Reference scheme Cz Independent variable SAD (S) vs. control (C) groups; prepost bright light treatment Time of assessment to examine stability of asymmetry over time in depression Depressed (D) and non-depressed (ND) groups (BDI median split) Resting EEG at F3/4, FP1/2, F7/8, FC5/6, FT9/10, C3/4, T7/8, CP5/6, TP9/10, P3/4, P7/8, P9/10, O1/2 Social phobics vs. controlsanticipating public speech Dependent variable EEG at F3/4, P3/4 Hem Yes Results summary S LFA Unchanged by treatment
1845
Cz, LM, AR
EEG at F3/4, F7/8, FTC1/2, others
No
Asymmetry stable across 35 monthly assessments, median intraclass correlation = 0.61 D, pct time with RFA
Baehr et al., 1998
24 (13 depressed; no sex information)
4357 years
No information
Cz
Percent time spent with RFA vs. LFA in F3/4
No
Bruder et al., 2001
53 (28 females)
1865
No information
Nose
Recovery vs. non-recovery from depression following SSRI (uoxetine) treatment EEG at AF1/2, F3/4, F7/8, T3/4, P3/4, C3/4, Cz, and Fz in alpha 1 (810 Hz); and alpha 2 (1013 Hz) EEG at F3/4, P3/4
Yes
In women: non-responders, RFA
Davidson et al., 2000
28 (no sex information)
1968 years
LM
Yes
Alpha 1: phobics, RFA/RATA
Davidson et al., 1985
20 (10 depressed; 7 females in each group)
1823 years
Cz
Happy, sad and neutral face pictures, depressed vs. non-depressed, left visual eld (LVF) vs. right visual eld (RVF)
Yes
Group differences in frontal asymmetry between RVF and LVF presentations appears to account for group differences in self-report ratings of happiness in response to lateralized picture presentations
Dawson et al., 1997
117 infants (52 females)
1315 months
No information
LM
Depressed (D) vs. non-depressed (ND) mothers; major depression (MD) vs. sub-depression (SD) Depressed (D) vs. non-depressed (ND) mothers; interaction with mother vs. familiar adult Depressed (D) vs. non-depressed (ND) mothers Emotional faces during emotional stimuli; depressed (D) vs. non-depressed (ND) mothers Emotion conditions (play w/mother (P), stranger approach (SA), maternal separation (MS); depressed (D) vs. non-depressed (ND) mothers; secure (S) vs. insecure (IS) attachment Depressed (D) vs. non-depressed (ND) groups Pre and post biofeedback treatment to LFA Depressed (D) vs. non-depressed (ND) mothers
Infant EEG at F3/4 and P3/4
Yes
D, LFA MD, LFA compared to SD
Dawson et al., 1999a,b
1315 months
R/L
LM
Yes
D, LFA (across other conditions)
Dawson et al., 1999a,b Dawson et al., 1997
1315 months
No information
LM
Infant EEG at F3/4 and P3/4; affection behaviors (AB) Infant EEG at F3/4 and P3/4
Yes
D, AB (D and AB), LFA Bilateral in activation during negative faces in D group
1117 months
No information
Cz
Yes
Dawson et al., 1992
27 infants (no sex information)
1117 months
No information
Cz
Yes
S: If D, LFA during P D, RFA
Debener et al., 2000
37 (25 female)
2364 years
L/R (most R)
Linked earlobes
EEG at Fp1/2, F3/4, F7/8, C3/4, T3/4, T5/6, P3/4, O1/2 BDI score
Yes
D, LFA, but not stable over time C, temporal stability in asymmetry BDI score post treatment D, RFA (mothers and infants)
Earnest, 1999
1 female (case study)
14 years
No information
Cz
No
Field et al., 1995
32 (16 female)
36 months
R (mothers)
Cz
Infant and mother EEG at F3/4 and P3/4
Yes
25
26
Citation Field et al., 2000 N 160 depressed, 100 non-depressed, all females 96 (56 female) Age Mean = 17.8 Handedness No information Reference scheme Cz Independent variable Depressed (D) vs. non-depressed (ND) mothers and their respective infants EEG at F3/4, P3/4 and O1/2 Smoking (S) and non-smoking (NS) conditions, NEO personality inventory (NEO-PI), fagerstrom TOLERANCE questionnaire (FTQ), negative mood (NM) Previously depressed (PD), depressed (D) and never depressed (Nev) groups Never depressed (Nev) and previously depressed (PD) groups Depressed (D) vs. non-depressed (ND) groups Anxious (A) and control (C) groups; anxious arousal (Ar) vs. worry (W) tasks Music (Mu) vs. massage (Ma) therapy; pre, during and post tests Dependent variable Infant and mother EEG at F3/4 and P3/4 Hem Yes Results summary D, RFA (mothers and infants)
Fox et al., 1996 Gilbert et al., 1999
4662 months
No information
Cz
50 male smokers
Mean = 28.1
Sociability (S), externalizing (E) and internalizing (I) EEG at F3/4 and P3/4
Yes
( S and RFA), E ( S and RFA), I NM, RFA NS, RFA
No
Gotlib et al., 1998
Study 1: 77, study 2: 59 all female
No information
Cz
EEG at F3/4, mood/cognitive measures
Yes
PD, D, LFA No other effects, suggesting no cognitive mediation D, LFA D, RPA Cz: D, RFA AR: D, RFA LM: no effects A, LFA (A and Ar), RPA
Henriques and Davidson, 1990 Henriques and Davidson, 1991 Heller et al., 1997
14 (6 previously depressed) 28 (18 female)
Dmean : 37.4 years, Cmean : 34.7 years 3157 years
Cz, LM
EEG at F3/4, F7/8, T3/4, T5/6, P3/4, C3/4 EEG at F3/4, F7/8, T3/4, T5/6, P3/4, C3/4 EEG at F3/4, A1/2, P3/4
Yes
Cz, LM, AR
Yes
40 (22 female)
No information
LM
Yes
Jones and Field, 1999
30 (no sex information)
Mean = 18.8 years
No information
Cz
Depression measures and EEG at F3/4, P3/4
No
LFA from pre to during, and from during to post in both Mu and Ma
Jones et al., 1998 Jones et al., 1998 Jones et al., 1997 Kentgen et al., 2000
25 infants (25 female)
1 month
No information
Cz
Pre, during and post massage Depressed (D) vs. non-depressed (ND) mothers Depressed (D) vs. non-depressed (ND) mothers Depressed (D) vs. non-depressed (ND) adolescents, anxiety comorbidity (A), no anxiety Males (M) vs. females (F), history of childhood depression (D) vs. No depression history (ND) Reassurance seeking (RS); Beck Depression Inventory (BDI) Anxious apprehension (AAp), anxious arousal (AAr), depressed (D), co-morbid (CM) and control groups (C) Resting EEG and electrodermal activity (EDA) at two time points (T1 and T2); depression (D) and anxiety (A) measured
Infant EEG at F3/4 and P3/4 Infant EEG at F3/4 and P3/4, vagal tone Infant EEG at F3/4 and P3/4 EEG at F3/4, F7/8, C3/4, T7/8, P3/4, P7/8
No
RFA from pre to during and from during to post D, LFA D, vagal tone D, RFA RFA, Neg affect pattern stable No frontal relationships. D, RPA
63 infants (no sex information) 44 infants (no sex information) 25 females
1 week
No information
Cz
Yes
1 month, 3 months (longitudinal) 12.218.8 years
No information
Cz
Yes
Nose
Yes
Miller et al., 2002
110 (66 female)
Mean = 26.6
AR
EEG at F3/4, AF3/4, F7/8, FC1/2, FC5/6, C3/4, T7/8, P3/4, P7/8, O1/2 EEG at F3/4, FP1/2, F7/8, C3/4, T3/4, T5/6, P3/4, O1/2 EEG at F3/4, F7/8, T3/4, T5/6, P3/4
Yes
D and M, LFA D and F, RFA
Minnix et al., this volume Nitschke et al., 1999
12 (6 females)
1952 years
LE
No
RS and BDI, LFA RS and BDI, RFA AAr, LFA Aap no LFA
67 (39 female)
1720 years
LM
Yes
Papousek and Schulter, 2001
50 (25 females)
No information
Nose
EEG at FP1/2, F3/4, P3/4
No
If RFA AND anxiety, EDA If LFA and depression, EDA
27
28
Citation Papousek and Schulter, 2002 N Study 1: 56 (30 female), study 2: 128 (68 female) Age Study 1: 1836 years, study 2: 1831 years Handedness R Reference scheme Nose Independent variable Study 1: anxious tension (AT); negative mood (NM); Time 1 (T1) vs. Time 2 (T2), study 2: anxious tension (AT); depression (D); anxiety; Time 1 (T1) vs. Time 2 (T2) Pre and post rigorous exercise conditions Dependent variable EEG atFP1/2, F3/4, T3/4, P3/4 for both Alpha 1 ( 1) and Alpha 2 ( 2) Hem No Results summary In both studies, AT 2 covaries with asymmetry in the fronto-polar region
Petruzzello and Landers, 1994 Reid et al., 1998
19 males
Mean = 22.7 years
LM
EEG at F3/4, T3/4; state measures of anxiety level
Yes
Anxiety post-exercise
LFA post-exercise Study1: 36 (17 depressed), study 2: 27 (13 depressed), all female Study 1: mean = 18.53 years, study 2: mean = 27.54 years R Cz, LM, AR Study 1: depressed (D) vs. non-depressed (ND) groups (BDI); study 2: depressed (D) vs. non-depressed (ND) groups (SCID) EEG at F3/4 EEG at Fp1/2, F3/4, F7/8, T3/4, T5/6, P3/4, C3/4, O1/2, A1/2, FTC1/2, TCP1/2, PO1/2 No D and ND not different (both studies) left anterior temporal activation in depressed (study 2, in LM reference, trend in AR reference) Subjects with LFA at beginning of session show change from neg to pos affect BDI, RFA RE, RFA
Rosenfeld et al., 1996
5 (4 female)
No information
Cz
Pre and post therapy session reports of affect; affect change (AC) score EEG at F3/4, P3/4 EEG at F3/4, P3/4
No
Schaffer et al., 1983 Silva et al., 2002 Tomarken et al., this volume
15 (10 female) 55 females
R R
Cz AR
High vs. low BDI scores Restrained eaters (RE) vs. unrestrained eaters (UE) Family history of depression (F) vs. no family history (NF); socio-economic status (SES)
Yes Yes
38 (20 female)
Mean = 13 years
Cz, LE, AR
EEG at F3/4
No
F: LFA SES, LFA
Wiedemann et al., 1999
48 (39 female)
Mean = 36.55 years
Cz
Panic (P) vs. control groups (C); conditions: rest (R), neutral stim (N), panic stim (Pn), anxiety stim (A), emotional stim (E), motor task (M)
EEG at F3/4, P3/4
Yes
Rest: P, RFA A: P, RFA P, LFA when shown erotic pictures
J.A. Coan, J.J.B. Allen / Biological Psychology 67 (2004) 749 29
30
fects, reporting more right frontal activity (not less left frontal activity) in depressed versus non-depressed mothers, and correspondingly similar differences in their respective infants. Infant studies such as these are important not only because they add to the growing literature on frontal EEG asymmetry and depression, but because they hold the potential for suggesting the origin of trait frontal EEG asymmetries. EEG spectra are highly heritable (see Lykken et al., 1982), and though it is impossible to tell from these reports the degree to which infant frontal EEG asymmetries derive from genetic versus environmental inuences, the nding of mother/infant asymmetry similarities points to the need for research on the heritability of frontal EEG asymmetry, in addition to research designed to identify which maternal behaviors or other environmental factors may inuence infant trait asymmetries. To date, there is limited evidence bearing on whether frontal EEG asymmetries are heritable (Coan, 2003; MacDhomhail et al., 1999). For example, in a small sample of female mono- and di-zygotic twins, Coan (2003) estimated that genetic inuences may account for as much as 22% of the variance in resting mid-frontal EEG asymmetry, and that most of this genetic inuence is non-additive. Other designs would also prove informative in addressing this question, including an adoption design comparing infants of depressed versus non-depressed adoptive mothers. If frontal EEG asymmetry were indeed a moderator of risk for psychopathology, one would expect trait levels of frontal EEG asymmetry to be relatively stable, even as episodes of depression come and go. As a moderator, trait (stable) frontal EEG asymmetry would interact with other environmental variables to cause depressive episodes; environmental variables, which are potentially less consistent, would serve as the co-determinants of any particular episode. If, by contrast, frontal EEG asymmetry changes to a signicant degree with the presence or absence of an affective disorder, one would conclude that frontal EEG asymmetry potentially acts as a mediator of psychopathology. (And a more complex possibility exists, such that asymmetry has some trait-like predictive value in the face of episode-dependent changes, thus potentially serving as moderator and mediator.) A review of the literature suggests a somewhat inconsistent picture regarding the stability of trait frontal EEG asymmetry in relationship to affective psychopathology. For example, Jones and Field (1999) have found that music or massage therapy applied to depressed adolescents resulted in the attenuation of their group level relative right frontal activity, suggesting that relative decreases in the magnitude of psychopathology can result in corresponding decreases in relative right frontal resting activity. Further, Jones et al. (1998) have demonstrated that applying massage therapy can produce alterations in frontal EEG asymmetry in infants. In this work, applying massage therapy to 1-month-old infants decreased relative right frontal EEG activity from pre-test to mid-treatment and from mid-treatment to post-test (Jones et al., 1998). Two studies offer some insight into the stability of trait frontal EEG asymmetry across time in adults undergoing treatment for depression. Debener et al. (2000) examined 15 medicated depressed patients on two occasions separated by 2 weeks, they found (1) that control subjects showed relatively greater left frontal resting activity than did subjects who had depression (this nding was anticipated) and (2) although depressed subjects exhibited lower testretest stability of frontal EEG asymmetry than did controls, no systematic mood-dependent changes in asymmetry occurred across sessions in these depressed patients. Patients in this study received a variety of antidepressant compounds, most initiated
31
prior to the rst EEG assessment, and 11 of the 15 additionally received benzodiazepines, thus raising the possibility that the variability in the depressed patients could reect the acute effects of the initiation of a trial of medication; no evaluation of this possibility was performed. One study examined the stability of frontal EEG asymmetry in a non-medicated sample of depressed patients as they progressed through a non-pharmacological treatment (Allen et al., 2004; reported in preliminary form in Urry et al., 1999). Allen et al. (2004) found evidence of adequate stability in resting frontal EEG asymmetry in depressed patients. Depression and frontal EEG asymmetry were assessed at 4-week interals. Across three occasions of measurement the median ICC at frontal sites (across three reference schemes) was 0.56, across ve assessments the comparable value was 0.61. This stability was apparent despite rather substantial improvements in clinical status over the same interval. These values are comparable to those reported by Debener et al. (2000) for control subjects, and by Tomarken et al. (1992b) in unselected college students. Interestingly, just as with Debener et al., Allen et al. found that changes in frontal asymmetry over assessments was not related patients clinical status or mood. The most obvious difference between the two studies is that patients in Allen et al. (2004) received no medication, whereas those in the study of Debener et al. (2000) received a variety of antidepressant compounds and most additionally received benzodiazepines. Clearly a systematic investigation of the impact of antidepressant and antianxiety medications on frontal EEG asymmetry would be desirable, both in terms of understanding the extent to which they may alter asymmetry, and also in terms of informing questions related to mechanism of action. In aggregate, these studies tentatively suggest some trait-like stability in frontal EEG asymmetry across time in depressed subjects, but provide no evidence to suggest that variation in frontal EEG asymmetry across measurement occasions is due to changes in clinical status. From a moderator perspective, it would be expected that testretest be fairly high even in clinical populations as symptoms wax and wane. While the evidence for this is strong on one study, but not another, it is consistently the case that occasion-related variance in resting frontal activity is not linked to clinical state. Most of the work on frontal EEG asymmetry and psychopathology has concerned depression, but several studies have also examined anxiety disorders as well. Relatively greater right frontal activity has been associated with panic disorder (Wiedemann et al., 1999) and social phobia (Davidson et al., 2000), while anxious apprehension has been associated with relatively greater left frontal activity (Heller et al., 1997). Although the general pattern is consistent with the model suggested by Davidson (1998a,b), inconsistencies have prompted Heller and colleagues (e.g., Heller and Nitschke, 1998) to propose a revised valence model of frontal EEG asymmetry and anxiety; i.e., anxiety is comprised of two distinct though related processes, anxious apprehension and anxious arousal, and these processes are reected in frontal EEG asymmetries as relatively greater left frontal and right parietal activity, respectively (see also Nitschke et al., 1999). Although there are data supportive of this position (Heller et al., 1997; Nitschke et al., 1999), it is hard to reconcile with other ndings that place all forms, or nearly all forms, of anxiety squarely in the domain of relatively greater right frontal activity. In an attempt to resolve this discrepancy, Heller and Nitschke (1998) have argued that affective valence may account for the difference between their ndings and those of Davidson et al. (2000) and Wiedemann et al. (1999). In particular, Heller and
32
Nitschke (1998) have argued anxiety coupled with negative affective valence should be associated with the pattern of right frontal activity that other researchers have found in their studies of frontal EEG asymmetry and anxiety. Thus, ultimately Heller and colleagues argue that anxiety can become manifest in frontal EEG asymmetry as either relatively greater left frontal, relatively greater right frontal, or relatively greater right parietal activity, depending on the relative presence or absence of anxious apprehension, anxious arousal and negative affect.
7. Frontal EEG asymmetry as a possible mediator of emotional responding Systematic alterations in frontal EEG asymmetry in response to specic emotional stimuli suggest the possibility that the systems tapped by frontal EEG asymmetry may also fulll a mediating role in emotional responding. While this need not be true of all emotional responding, it may well be true of the motivational properties that form one or more of the components of emotional responding. Modeling frontal EEG asymmetry as a mediator. Just as with modeling frontal EEG asymmetry as a moderator, it will be useful to review the data analytic and statistical considerations underlying the specication of a mediator. Fig. 2 represents a mediational model of frontal EEG asymmetry in emotional responding, also adapted from Baron and Kenny (1986). For this illustration, consider again the independent variable example discussed previously, where subjects are presented with an image of either a poisonous snake (the fear stimulus) or an elephant (a neutral stimulus). Frontal EEG asymmetry could be said to be a mediator of emotional responding if a measurable change in it is necessaryat least in partfor the relationship between some emotion-eliciting stimulus and the subsequent emotional response. That is, it could be said to mediate emotional responses to the degree that it serves as, or is highly correlated with, the mechanism by which emotional stimuli have their effects. In the case of the example, arguing that frontal EEG asymmetry is a mediator of fear responses is tantamount to saying that the fear response would not occur, or would occur differently, or would occur at a lower level of self reported intensity, if there was no change in frontal EEG asymmetry in the direction of increased relative right activity (in accord with the approach/withdrawal model). In this way, the mediational models refer specically to frontal EEG activation as opposed to activity. In Fig. 2, this situation is represented by three causal pathways, identied as paths a, b and c. In path a, changes in the type of stimulus (in the example, the presentation of a
Mediator (EEG Asymmetry) a Stimulus Type (snake vs. elephant) b Emotional Response
Fig. 2. Frontal EEG asymmetry as a mediator. Model adapted from Baron and Kenney (1986).
Table 4 Frontal EEG activation asymmetry as a state measure

Citation Allen et al., 2001 N 18 females Age 1838 years Handedness R reference scheme Cz Independent variable Biofeedback training to move asymmetry towards greater left or right activity Wakefulness vs. various sleep stages (REM, StgII, SWS) Pre (PRE) and post (POST) EEG hook-up mood ratings Resting EEG at F3/4, F7/8, FTC1/2; state EEG at F3/4, F7/8, and FTC1/2 during voluntary facial expressions of anger (A), disgust (D), fear (F), joy (J) and sadness (S). Voluntary emotional facial expressions grouped according to approach (A), withdrawal (W) and control (C) conditions Emotional expression during emotional lms; happy (H), sad (S) and neutral (N) Emotional facial expressions during emotional lm clips Films of an actress performing happy vs. sad faces Meditation (M) vs. control (C); pre-treatment (T1), post-treatment (T2), 4 month follow-up (T3); antibody titers to the inuenza vaccine (HIA) Dependent variable Self report emotion, facial EMG Hem Yes Results summary
RFA caused positive affect, zygomatic, and

corrugator activity Waking EEG correlated with sleep (notably REM) in frontal and temporal regions Women: POST mood, LFA Men: POST mood, RFA
Benca et al., 1999
17 (4 female)
3563 years
LM
EEG at F3/4, F7/8, T3/4, T5/6, P3/4 O1/2 EEG at F3/4, FP1/2, F7/8, T3/4, T5/6, C3/4, P3/4, O1/2 Emotional intensity reports (IR) following voluntary facial expressions of anger (IR-A), disgust (IR-D), fear (IR-F), joy (IR-J) and sadness (IR-S)
Yes
Blackhart et al., in press Coan and Allen, in press
77 (41 female)
1639 years
LM
No
Resting EEG: 30; state EEG:31
No information
AR, LM
No
resting LFA, IR-A resting LFA, IR-D resting LFA, IR-J LFA during A, IR-A LFA during J, IR-J RFA during F, IR-F
Coan et al., 2001
36 (26 female)
1724 years
Cz, LM, AR
EEG at F3/4, F7/8, FTC1/2, P3/4
Yes
W, LFA A=C
Collet and Duclaux, 1986 Davidson et al., 1990
24 (13 female)
1845 years
AR
EEG at F3/4, T1/2, T3/4, T5/6, C3/4, P3/4 O1/2 EEG at F3/4, C3/4, T3/4, P3/4
No
No effects
11 females
1741 years
Cz
Yes
No effect of lms. Disgust (face), RATA Joy (face), LATA Happy, LFA No frontal effects. At T2 and T3, baseline activity in left central region (C3/4) in M In M, greater relative left activity in C3/4 from T1 to T2 corresponded with greater HIA.
Davidson and Fox, 1982 Davidson et al., 2003
24 infant females 41 (29 female)
10 months 2356 years
No information R
Cz LE
EEG at F3/4, P3/4 EEG at F3/4, FC7/8, T3/4, C3/4
Yes No
33
34
Citation Davidson et al., 1992 Davidson et al., 1993 N 9 rhesus monkeys (4 female) 9 rhesus monkeys (4 female) Age 12 months 12 months Handedness No information No information reference scheme LM LM Independent variable Benzodiazepam shot vs. vehicle Benzodiazepam shot vs. vehicle Dependent variable EEG at F3/4, P3/4 EEG at F3/4, P3/4, freezing time in response to challenge Hem Yes No Results summary Benzodiazepam, LFA Those with most LFA to Benzodiazepam showed longest duration freezing behavior
Dawson et al., 1992 Ekman and Davidson, 1993 Ekman et al., 1990
21 infants 14
21 months No information
Cz LM
Baseline (B) vs. mother out (MO) conditions Duchenne (D) vs. unfelt smiles (U) vs. anger face (A) Emotional facial expressions during emotional lm clips; Duchenne (D) vs. unfelt (U) smiles Stranger approach (SA), mother approach (MA), maternal separation (MS) condition
EEG at F3/4, P3/4 EEG at F3/4, F7/8, C3/4, T3/4, T5/6, P3/4, O1/2 EEG at F3/4, C3/4, T3/4, P3/4
No Yes
overall frontal activation

during MO D LFA, LATA D LFA than A
31 females
1741 years
Cz
Yes
D, LATA
U, RATA EEG at F3/4, P3/4 Yes MA, LFA (mother reach sub-condition); ( LFA if vocalizing) MS + crying, RFA ( LFA if vocalizing) D, LFA than U A and S (no crying), LFA A and S (crying), RFA
Fox and Davidson, 1986
35 infant females
10 months
R (parents)
Cz
35 infant females
10 months
R (parents)
Cz
Stranger approach (SA) vs. mother approach (MA); facial expressions of joy (J), anger (A), sadness (S); Duchenne (D) vs. unfelt (U) smiles Emotional facial expressions during taste conditions (sucrose [S], citric acid [CA], H2 O) Various self-report measures Occular artifacts vs. no occular artifacts in EEG recordings Anger manipulation (AM) vs. no anger (NA), symptoms associated with depression (D) and hypomania-plus-biphasia (HB)
EEG at F3/4, P3/4
Yes
16 infant (no sex information)
23 days
R (parents)
Cz
EEG at F3/4, P3/4
Yes
13 Hz band: H2 O, RFA S, LFA 612 Hz band: H2 O, RFA S, LFA
Gilbert et al., 1994 Hagemann and Naumann, 2001 Harmon-Jones et al., 2002
32 (16 female) 31 (19 female) 67 (33 female)
2135 years 1936 No information
R No information R
No information Cz LE
EEG at F3/4, T3/4, P3/4 (others) EEG at FP1/2, F3/4, F7/8, T3/4, C3/4, T5/6, P3/4, O1/2 EEG at F3/4, F7/8, FT7/8
Yes Yes Yes
BDI, RFA (in normals)

No effects of occular artifact in the alpha range
LFA during AM During AM, if HB, then

LFA During AM, if D, then RFA
Harmon-Jones and Sigelman, 2001
42 males
No information
LM
Baseline (B), insult (I), no-insult (NI) conditions
Self-reported anger (A) and aggression (AG); EEG at F3/4, F7/8, P3/4
Yes I produced, A, AG, LFA LFA correlated with anger in I, not NI LFA correlated with Aggression in I, not NI Yes H, LFA S, RFA D, RFA P, LFA during H N, RFA during H Yes V, LFA Yes
Jones and Fox, 1992
23 females
1822 years
Kline et al., 2000 Miller and Tomarken, 2001
49 females 60 (30 female)
Mean = 64.2 years Mean = 19
Emotional facial expressions during videos of anger (A), happiness (H), disgust (D), and sadness (S); positive (P) vs. negative (N) affectivity groups No information No information Odor conditions; vanilla (V), neutral (N), valarian (VN). Cz Incentive levels: large reward (LR), reward (R), no reward (NR), punish (P), large punish (LP); expectancy levels: high (HE), medium (ME) and low (LE); response levels: active (A) and passive (PS); hand response levels: left (LT) and right (RT) TV segments depicting positive (P) and negative (N) scenes Reward (R) vs. punishment (P) conditions
Cz
EEG at F3/4, T3/4, P3/4
EEG at Fp1/2, F3/4, F7/8, O1/2, P3/4, T5/6 EEG at F3/4, C3/4, P3/4, AF3/4
L and R
R, LFA Men: HE, LFA Women: LE, LFA LT, RFA RT, LFA
Reeves et al., 1989 Sabotka et al., 1992
16 (7 female) 15 (8 female)
2050 years 1825 years
R R
LM LM
EEG at F3/4, O1/2 Ratings of happiness (H) vs. sadness (S) during conditions; EEG at F3/4, F7/8, T3/4, C3/4, O1/2, TP3/4; Approach (nger press; FP) vs. withdrawal (nger lift; FL) responses
Yes P, LFA N, RFA Yes R, LFA P, RFA
Sanders et al., 2002
Study 1: 39 (29 female), study 2: 26 infants 24 males
Study 1: mean = 31 years No information Cz
Schmidt et al., 1999
1838 years
Cz
Studies 1 and 2: lavender odor (L) EEG at F3/4, P3 vs. rosemary odor (R); baseline left frontal (LF) vs. baseline right frontal (RF) Prednisone (P) vs. control groups; EEG at F3/4, C3/4, P3/4, O1/2 pre-treatment (T1) vs. post-treatment (T2) Imagination condition; depressed (D) vs. sexually aroused (S) EEG at F3/4, C3/4, P3/4, O1/2
No
Study 1: L, LFA Study 2: no effects
No
T2 and P, RFA
Tucker and Dawson, 1984
9 method actors (5 female)
No information
LM
Yes S RFA, compared to D
35
36
Citation Waldstein et al., 2000 Zinser et al., 1999 N 30 (18 female) 72 (no sex information) Age Mean = 24 years Mean = 26.3 Handedness R R reference scheme Cz Cz Independent variable Imagination and lm conditions; happiness (H) vs. anger (A) Cigarette deprivation (D) and control (C) groups by 1 cigarette anticipation (A) and 2 cigarette no wait (N) groups (2 2 factorial) Dependent variable EEG at F3/4, C3/4, P3/4, O1/2 EEG at F3/4 Hem Yes No Results summary H LFA compared to A D,A LFA Smoking itself, LFA
RFA: right frontal activation, LFA: left frontal activation; RATA: right anterior temporal activation, LATA: left anterior temporal activation; RPA: right parietal activation, LPA: left parietal activation.
37
snake versus an elephant) correspond to different changes in frontal EEG activity (frontal EEG activation). In path b, frontal EEG activation corresponds to changes in emotional responding. In path c, changes in levels of the independent variable (snake versus elephant) correspond to changes in emotional responding. Frontal EEG asymmetry can be said to be a mediator of emotional responding to the extent that path cthe direct relationship between emotional stimuli and emotional respondingis attenuated when paths a and b are statistically adjusted for. By this reasoning, frontal EEG asymmetry would be a perfect mediator of emotional responding if statistically adjusting for paths a and b resulted in a residual of zero for path c (thus identifying frontal EEG activation as a necessary condition for emotional responding). This idealized scenario is unlikely. Rather, frontal EEG asymmetry is likely to represent, if anything, one of several mediating processes in emotional responding. If this is the case, then it would be expected that paths a and b improve the overall t of the model, and that the residual in path c is signicantly decreased when paths a and b are statistically adjusted for. Such a nding would suggest that, while not necessary, frontal EEG activation may be a sufcient condition for generating emotional responses, an idea that could in theory be tested. In applying data analysis to this mediational model, one can start by conceptualizing each path (a, b and c) as separate regression equations. That is, one could regress frontal EEG activation (the mediator) on the independent variable, the emotional response (the criterion variable) on the independent variable, and the emotional response on frontal EEG activation. Baron and Kenny (1986) point out that coefcients from each equation can alone be used to suggest (or, more strongly, rule out) a mediating relationship. Using the example, by this method one would expect to nd (1) that frontal EEG activation is related to the differential presentation of a fearful or non-fearful stimulus (represented in Fig. 2 as path a), (2) that the intensity of a fear related response will be related to the presentation of a fearful versus a non-fearful stimulus (represented in Fig. 2 as path c), and (3) that the intensity of the fear related response will be related to frontal EEG activation (represented in Fig. 2 as path b). If any of these three conditions are not obtained, then the possibility of EEG activation serving a mediating role in the observed emotional responding can effectively be ruled out. If these conditions are obtained, however, then mediation has not been ruled out until controlling paths a and b show an observed impact on path c in Fig. 2.4 Another possible method of testing mediational models of this sort is to utilize structural equation modeling (SEM). The SEM approach is increasing in popularity, an indirect measure of which is its conspicuous appearance as a new component of various popular statistical programs, such as SAS, Statistica, etc. (Also useful are SEM dedicated programs
4 It should be noted that while it may be tempting to conclude, from analyses such as those described here, that mediation can be ruled in, this is in fact not generally true. With the mediational analyses described here, true mediation can at best only fail to be ruled out, unless statistically adjusting for paths a and b do indeed result in a residual of zero, which is exceedingly unlikely (Patrick McKnight, personal communication). Moreover, the Baron and Kenney method of addressing the status of a hypothesized mediator may suffer from unrealistic type I error rates and low statistical power with small or even medium sample sizes. The length and scope of this article precludes a detailed discussion of the many complexities involved in the determination of statistical mediation, such as issues of sample size, attempts to model multiple mediators, etc. Thus, our treatment of these data analytic issues are somewhat idealized. For additional information on mediation analyses, the reader is enthusiastically referred to Kenny et al. (1998) and MacKinnon et al. (2002).
38
such as LISREL and EQS. EQS is particularly recommended.) SEM approaches essentially reconstruct the three regression equations listed above as one model, where coefcients are estimated within the context of the entire covariance matrix of the models variables. This allows one to estimate the overall t of the model one has constructed, using various indices (for a thorough review of SEM approaches, see Loehlin, 1998). 7.1. Empirical support for frontal EEG asymmetry as a mediator A number of studies provide oblique evidence that frontal EEG asymmetry may indeed mediate emotional responses (see Table 4). Most of the evidence reviewed below simply demonstrates that frontal EEG asymmetry can be altered by emotional or quasi-emotional stimuli. Few studies measure other emotional responses as a correlate of changes in frontal EEG asymmetry, where each is in response to some emotional stimulus. No studies to date have explicitly tested the mediational hypothesis. The approach/withdrawal model of frontal EEG asymmetry accommodates state changes as well as individual differences in trait propensities. According to the model, stimuli intended to elicit approach-oriented responses should result in an observed relative left frontal EEG activation, while stimuli intended to elicit a withdrawal-oriented response should result in an observed relative right frontal EEG activation. Indeed, there is evidence to support this prediction (e.g., Coan et al., 2001; Ekman et al., 1990; Davidson and Fox, 1982). Some of the earliest work in this area was done by Davidson and Fox (1982), who found that 1012-month-old infants showed evidence of increased left frontal EEG activity in response to lms of an actress performing happy faces. Subsequently, Fox and Davidson (1986) found that infants as young as 23 days exhibited an increase in left frontal activity in response to drops of sugar water deposited on their tongues, while exhibiting more right frontal activity in response to neutrally avored drops (of water). Still later, Fox and Davidson (1987) found that 10 month-old infants who reached for their mothers during a mother approach task showed more concomitant left frontal activity than infants who did not, and that babies who cried in response to maternal separation showed a similar right frontal activity effect compared to other infants. Moreover, Fox and Davidson (1988) found that anger and sadness in response to maternal separation corresponded with relatively greater left frontal activity, unless the infants were crying, in which case both anger and sadness corresponded with relatively greater right frontal activity. (This last nding presents some interesting difculties for both the approach/withdrawal and valence models of EEG asymmetry, but nevertheless serves to illustrate the potential mediating role of processes indicated by anterior EEG alpha asymmetries.) Similar results have been obtained in studies of adults. Using emotional lms to investigate the relationship between emotional experience and frontal EEG asymmetry, Davidson et al. (1990) found that, although frontal EEG recordings averaged across the entire period of viewing emotional lms did not show evidence of differences in hemispheric activity, important differences did emerge during facially expressive emotional reactions to those lms. In particular, disgust lms elicited relatively greater right anterior temporal activity relative to baseline, while happy lms elicited more left anterior temporal activity (Davidson et al., 1990). In a reanalysis of the same data set, Ekman and colleagues (Ekman et al., 1990) found that individuals who exhibited Duchenne smiles (smiles involving activation of the
39
orbicularis pars lateralis muscle) in response to happy lms showed more concomitant left anterior temporal activity than did individuals who exhibited unfelt smiles. Subsequently, Ekman and Davidson (1993) asked participants to voluntarily perform Duchenne versus unfelt smiles and again found that Duchenne smiles resulted in greater left anterior temporal, as well as left frontal, activity. Thus, lateralized brain activity seems to be a potentially important element in the collection of properties that comprise at least some emotions or emotion families. This idea is borne out further by other work (e.g., Coan et al., 2001; Harmon-Jones and Sigelman, 2001; Jones and Fox, 1992). For example, Coan et al. (2001) used a voluntary directed facial action task (cf., Levenson et al., 1990) to elicit approach (joy and anger) and withdrawal (disgust, fear and sadness) related emotions, hypothesizing that approach-related emotions should result in left lateralized frontal activation while withdrawal-related emotions resulted in right lateralized frontal activation. Their predictions were partially conrmed; withdrawal-related emotions, particularly fear and sadness, did result in the expected relative right frontal activation compared to a control condition, but approach-related emotions did not result in a comparable relative left frontal activation (Coan et al., 2001). While Coan et al. (2001) did measure other emotional responses, such as subjective emotional experience reports, they did not test an explicit mediational model of frontal EEG asymmetry and emotion. Recently, Coan and Allen (2003c) used the extant data reported in Coan et al. (2001) to assess the relationship between state frontal EEG asymmetry and emotional experience during voluntary emotional facial expressions. They found that anger and, marginally, joy were more likely to be reported if their concomitant state EEG asymmetries involved greater left activity, and that fear was more likely to be reported if its concomitant state EEG asymmetry involved greater right activity (state EEG asymmetry by emotion interaction, F(4, 895) = 8.17, P < 0.001). These results proved to be independent of reference scheme or specic frontal region. Unfortunately, this data set did not include a single dependent measure that was recorded following each emotion task, and the lack of such an outcome measure precluded a proper mediational analysis. Had each participant been asked, for example, to rate the degree to which they experienced fear following each emotion induction, a mediational analysis, where emotion type would be the independent variable, frontal EEG asymmetry would be the mediator and fear experience would be the outcome measure, would be readily forthcoming. Further evidence of the relationship between emotional states and concomitant changes in frontal EEG asymmetry can be found in a series of studies conducted by Harmon-Jones and co-workers (Harmon-Jones and Sigelman, 2001; Harmon-Jones et al., 2003). In an investigation that probably comes closest to testing a true mediational model of frontal EEG asymmetry and emotion, Harmon-Jones and Sigelman (2001) observed that individuals who showed relative left frontal activation (again, change in frontal EEG asymmetry from rest) in response to an insult were more likely to report experiencing anger. Similarly, Harmon-Jones et al. (2003) observed that individuals showing left frontal activation under a different anger induction procedure also displayed more aggressive and retaliatory behavior. Interestingly, Harmon-Jones et al. (2003), also found that left frontal activation only occurred in response to an anger elicitation when coping or retaliatory responses were possible. In their study, college students were confronted either with a bogus radio broadcast conrming that a tuition increase was certain or with one that suggested that such an increase was merely
40
under consideration. Those subjects who were led to believe that the increase was merely under consideration (1) showed greater left frontal activation than those who believed the increase was certain and (2) were more likely to engage in coping actions such as signing and taking petitions (Harmon-Jones et al., 2003). Taken together, these studies suggest that frontal EEG asymmetry, or by inference the activity of the brain systems it measures, may mediate emotional responding, and that these systems are likely to be particularly motivational in nature, as predicted by the approach/withdrawal model.
8. Frontal EEG asymmetry as a trait-like moderator and a state-related mediator One important question regarding frontal EEG asymmetrys role as either a moderator or a mediator is its relative robustness as one or the other. That is, one could ask how robust trait-like individual differences in baseline frontal EEG asymmetry are across different emotional states? An equally important question could be asked of state dependent frontal EEG asymmetries: how robust are state frontal EEG asymmetries across individuals? Generalizability theory, or g-theory (Cronbach et al., 1972), provides a method of estimating the reliability of particular facets of any given measure across other facets of that measure. In g-theory, variance components are estimated and used to calculate specic intraclass coefcients, in this context referred to as generalizability or g coefcients. Thus, g-theory can be applied to answer the precise questions articulated above. Indeed, for this article, g-theory was applied to an extant data set described in Coan et al. (2001). In this data set, frontal EEG asymmetries were recorded from 36 research participants both at rest and during a voluntary directed facial action task wherein they were asked to perform voluntary facial expressions denoting anger, disgust, fear, joy and sadness (see Coan et al., 2001 for a detailed description of the methods used in this study). These emotions were then grouped according to the approach/withdrawal motivational model of emotion. By this scheme, frontal EEG asymmetries during anger and joy were arithmetically averaged into an approach condition and disgust, fear and sadness were arithmetically averaged into a withdrawal condition. An additional control condition was also employed. Generalizability theory allowed for analyses of these data such that the following questions could be evaluated empirically: (1) were state changes in frontal EEG asymmetry resulting from the emotional manipulation task reliably elicited in all subjects? And (2) were trait predispositions in frontal EEG asymmetry preserved within emotional state conditions? Results indicated rst that trait, state and the trait by state interaction accounted for approximately 8, 10 and 11% of the variance in frontal EEG asymmetry during state manipulations, respectively (see Table 5). These results suggest that variations in frontal EEG asymmetry attributable to traits, states and trait by state interactions are approximately equalaround 10%. In addition to this information, trait and state g-coefcients were 0.42 and 0.97, respectively. Thus, the answers to questions 1 and 2 above appear to be yes and somewhat, respectively. Indeed, it appears that state changes were strikingly robust to individual differences in this study. Put another way, to the extent that state changes in frontal EEG asymmetry occurred at all in response to the emotional manipulation, they occurred in nearly all of the subjects, regardless of their trait predispositions. By contrast, trait predispositions were only moderately preserved within state manipulations. Ultimately, these
J.A. Coan, J.J.B. Allen / Biological Psychology 67 (2004) 749 Table 5 Results of a state manipulated frontal EEG asymmetry generalizability study N Trait State Trait state Residual 36 3 Variance component 0.003 0.004 0.005 0.030 Relative error variance Trait State 0.002 0.0001 % variance 8.23 9.79 11.42 70.57
41
Generalizability coefcient 0.42 0.97
results raise the possibility that frontal EEG asymmetry may function more robustly as a mediator of emotional responses than as a moderator of emotional responses, although different results might be obtained if one assessed frontal EEG asymmetry across several occasions to provide a better and more stable estimate of trait asymmetry. Indeed, the SpearmanBrown Prophecy Formula suggests that the generalizability of trait frontal EEG asymmetry across emotional states would increase from 0.42 to 0.74 if averaged across 4 occasions of measurement. Such an approach may make it easier to assess the degree to which trait predispositions in frontal EEG asymmetry affect state changes in the same.
9. Concluding remarks While there is a great deal of evidence suggesting that frontal EEG asymmetry may function as both a moderator and a mediator in various aspects of emotion, it is striking how few studies provide explicit evidence of either moderation or mediation. Indeed, in the case of mediation, explicit tests are entirely lacking, although extant data sets could conceivably be used to explicitly test for both mediation and moderation. In this paper, recommendations have been made for the rigorous evaluation of frontal EEG asymmetry as either a moderator or a mediator. 9.1. Frontal EEG asymmetry as a moderator of emotion Thus far, compelling evidence suggests a moderating role for frontal EEG asymmetry in at least some emotions, supporting Davidsons (1998a,b) theory of affective style. Nevertheless, generalizability analyses presented here suggest that while frontal EEG asymmetry may function as a moderator, its inuence may only be modestly reliable across emotional states, and it is emotional states that trait frontal EEG asymmetries are supposed to be moderating. Others have obliquely pointed this out in various ways. For example, Davidson and colleagues (e.g., Davidson, 1998b; Henriques and Davidson, 1991; Tomarken et al., 1990; Wheeler et al., 1993) have long advocated for the use of multiple measures of resting frontal EEG asymmetry in its measurement as a robust trait. This group of researchers have taken different approaches to this, from averaging values across multiple sessions (thereby increasing the measures reliability; e.g. Sutton and Davidson, 1997) to selecting
42
only individuals whose resting frontal EEG asymmetry remains stable over 2 or >2 weeks (e.g., Wheeler et al., 1993). Such practices have many psychometric virtues. Indeed, the SpearmanBrown prophecy formula would predict that with four occasions of measurement, the reliability of trait frontal EEG asymmetry, even across emotional states, should increase to a quite respectable range. 9.2. Frontal EEG asymmetry as a moderator of psychopathology To the extent that trait frontal EEG asymmetry moderates emotional responses, it may do the same for psychopathology. That is, as Davidson (1998a) has argued, an individuals particular affective style may create a predisposition to risk for psychopathology. If this is true, the empirical data to date remain mixed in their support this proposition. While several studies have suggested a relationship between frontal EEG asymmetry and psychopathology (Allen et al., 1993; Bruder et al., 1997; Henriques and Davidson, 1990, 1991; Gotlib et al., 1998; Debener et al., 2000; Davidson et al., 2000; Wiedemann et al., 1999), some have suggested otherwise (Reid et al., 1998), and none have attempted to rigorously model frontal EEG asymmetry as a moderator in the explicit fashion recommended by Baron and Kenny (1986). Importantly, however, establishing a moderating effect of frontal EEG asymmetry will require that data sets include measures of risk-related events or traits that frontal EEG asymmetry can moderate. So for example, if it is found that a particular cognitive vulnerability (for example, hopelessness; Abramson et al., 2002) is related to depression, then one could ask if that relationship is especially strong for those lowest in left frontal activity. Ideally, of course, one would obtain prospective data, such that hopeless cognitions and frontal EEG were assessed in a relatively high-risk but euthymic population, and the development of depression at a later timepoint would be assessed. One might object to this last remark on the grounds that modeling moderator effects for psychopathology is not possible given the difculty in implementing the kind of large, prospective studies that such an approach would require, but such declarations are probably premature. There may be ways to model frontal EEG asymmetry as a genuine moderator of psychopathology that do not require such costly commitments. For example, Allen et al. (1993) studied participants suffering intermittently from seasonal affective disorder (SAD)a psychological difculty whose course, treatment and eliciting environmental stimulus is relatively well understood, and moreover, relatively manipulable by investigators (via techniques such as phototherapy and/or waiting for the seasons to change). In studying psychopathology with an episodic course, of which SAD is a prime example, one could construct a GLM that would straightforwardly test the moderator model such as the following: SAD = season + trait frontal EEG asymmetry + season trait frontal EEG asymmetry. One could also substitute a phototherapy manipulation for the Season variable in the case of SAD, or a psychotherapy treatment variable in the case of non-seasonal depression. Other possibilities no doubt exist. For example, various negative life events have been associated with depression (e.g., Goodman, 2002). One could imagine that individuals who have just experienced the death of a close loved one may be at greater risk for depression than would those who have not. One could imagine further, indeed Davidsons diathesis/stress model predicts, that individuals with relatively greater right frontal EEG asymmetry would
43
be particularly likely to become depressed following the death of a close loved one. As a general linear model, one could state this as follows. Depression = loss status (yes/no) + trait frontal EEG asymmetry + loss trait frontal EEG asymmetry. Similarly, adjustment disorders with depressed features frequently strike freshmen undergraduate students away from home for the rst time. The approach/withdrawal diathesis/stress model would predict that individuals who show relatively greater right frontal EEG activity at rest are at greater risk for such adjustment disorders during their freshman year of college. 9.3. Frontal EEG asymmetry as a mediator While no study to date has explicitly modeled frontal EEG asymmetry as a mediator of emotional responses, several studies provide evidence that this is at least a distinct possibility, and the generalizability analyses reported here suggest that if frontal EEG asymmetry does indeed function as a mediator, its effects are quite robust to individual differences in trait frontal EEG asymmetry. As for modeling frontal EEG asymmetry as a mediator, it will be important for individuals to record measures of emotional response other than frontal EEG asymmetry per se. That is, many studies of state changes in frontal EEG asymmetry regard frontal EEG asymmetry as a dependent variable only. In order to assess its function as a possible mediator, some other criterion of emotional response must be established. A good example of this, though also one in which frontal EEG asymmetry is not explicitly modeled as a mediator, was conducted by Davidson et al. (2000). These researchers identied state changes in frontal EEG asymmetry in social phobics as a function of anticipating giving a speech. In addition to demonstrating changes in frontal EEG asymmetry, other measures of physiological and self-reported distress were obtained, allowing at least the possibility of an explicit mediational model. Similarly, Harmon-Jones et al. (2003), in identifying state changes in frontal EEG asymmetry resulting from bogus anger inducing radio broadcasts, also measured coping behaviors, such as signing and taking petitions, again allowing for the possibility of an explicitly mediational model. Ultimately, if frontal EEG asymmetries prove to satisfy criteria as a mediator of emotion, then activity in the brain systems tapped by frontal EEG asymmetry may not simply be an output of emotion, but rather may be something that facilitates the emotional response. 9.4. Future directions As a sub-eld of emotion and motivation, the study of frontal EEG asymmetry holds substantial promise. In terms of basic science, this sub-eld promises to inform us regarding the fundamental properties of emotion, both in terms of how emotions occur and what properties they entail. In more applied settings, the possibility yet remains that frontal EEG asymmetry may serve as a useful liability marker for depression and anxiety. Regardless of the application of this measure, it is increasingly important that theoretical predictions surrounding frontal EEG asymmetry are put to more rigorous tests, especially those provided by testing explicit mediational and moderational models. In the absence of such explicit tests, the eld will remain a collection of studies merely suggestive of moderating and mediating inuences, around which much exciting and potentially important speculation
44
and theory is generated. It is possible to more explicitly test that speculation and theory with models that more adequately can support or refute such inferences. The results of these explicit tests will guide the eld in investigating whether and how frontal EEG asymmetry may serve as a risk marker for psychopathology, and in investigating what underlying physiological systems inuence and are inuenced by emotion. Gaining a deeper understanding of the fundamental properties of emotion will require clear thinking in terms of how the various components of emotion and emotional experience are related. It is in this spirit that the present remarks are offered.
Acknowledgements This work was supported, in part, by a Young Investigator award from NARSAD (John Allen) and a Graduate Research Fellowship from the National Science Foundation (James Coan). Portions of the present work appear in Coan and Allen (2003a) and appear with permission. Address for correspondence: James A. Coan, W.M. Keck Center for Functional Brain Imaging and Behavior, Waisman Center, University of Wisconsin, Madison, WI 53705, USA. E-mail: jacoan@wisc.edu or John J.B. Allen, Department of Psychology, University of Arizona, Tucson, AZ 85721-0068, USA. E-mail: jallen@u.arizona.edu.
References
Allen, J.J.B., Coan, J.A., Nazarian, M. Whats the difference? Issues and assumptions in the use of difference scores and other metrics of anterior brain asymmetry in emotion. Biological Psychology (Special issue), this volume. Allen, J.J.B., Harmon-Jones, E., Cavender, J.H., 2001. Manipulation of frontal EEG asymmetry through biofeedback alters self-reported emotional responses and facial EMG. Psychophysiology 38, 685693. Allen, J.J., Iacono, W.G., Depue, R.A., Arbisi, P., 1993. Regional electroencephalographic asymmetries in bipolar seasonal affective disorder before and after exposure to bright light. Biological Psychiatry 33, 642646. Allen, J.J.B., Urry, H.L., Hitt, S.K., Coan, J.A., 2004. Stability of resting frontal EEG asymmetry across different clinical states of depression. Psychophysiology 41, 269280. Baehr, E., Rosenfeld, J.P., Baehr, R., Earnest, C., 1998. Comparison of two EEG asymmetry indices in depressed patients vs. normal controls. International Journal of Psychophysiology 31, 8992. Benca, R.M., Obermeyer, W.H., Larson, C.L., Yun, B., Dolski, I., Kleist, K.D., Weber, S.M., Davidson, R.J., 1999. EEG alpha power and alpha power asymmetry in sleep and wakefulness. Psychophysiology 36, 430436. Blackhart, G.C., Kline, J.P., Donohue, K.F., LaRowe, S.D., Joiner, T.E. Affective responses to EEG preparation and their link to resting anterior EEG asymmetry. Personality and Individual Differences, in press. Bruder, G.E., Fong, R., Tenke, C.E., Leite, P., Towey, J.P., Stewart, J.E., McGrath, P.J., Quitkin, F.M., 1997. Regional brain asymmetries in major depression with or without an anxiety disorder: a quantitative electroencephalographic study. Biological Psychiatry 41, 939948. Bruder, G.E., Stewart, J.W., Tenke, C.E., McGrath, P.J., Leite, P., Bhattacharya, N., Quitkin, F.M., 2001. Electroencephalographic and perceptual asymmetry differences between responders and nonresponders to an SSRI antidepressant. Biological Psychiatry 49, 416425. Buss, K.A., Schumacher, J.R.M., Dolski, I., Kalin, N.H., Goldsmith, H.H., Davidson, R.J., 2003. Right frontal brain activity, cortisol, and withdrawal behavior in 6-month-old infants. Behavioral Neuroscience 117, 1120. Carver, C.S., White, T.L., 1994. Behavioral inhibition, behavioral activation, and affective responses to impending reward and punishment: the BIS/BAS scales. Journal of Personality and Social Psychology 67, 319333.
45
Coan, J.A., 2003. The heritability of trait frontal EEG asymmetry and negative emotionality: sex differences and genetic nonadditivity (Doctoral dissertation, University of Arizona). Dissertation Abstracts Int. 64, 2382. Coan, J.A., Allen, J.J.B., 2003a. The state and trait nature of frontal EEG asymmetry in emotion. In: Hugdahl, K., Davidson, R.J. (Eds.), The Asymmetrical Brain. MIT Press, Cambridge, MA, pp. 565615. Coan, J.A., Allen, J.J.B., 2003b. Frontal EEG asymmetry and the behavioral activation and inhibition systems. Psychophysiology 40, 106114. Coan, J.A., Allen, J.J.B., 2003c. Varieties of emotional experience during voluntary emotional facial expressions. Annals of the New York Academy of Sciences 1000, 375379. Coan, J.A., Allen, J.J.B., Harmon-Jones, E., 2001. Voluntary facial expression and hemispheric asymmetry over the frontal cortex. Psychophysiology 38, 912925. Collet, L., Duclaux, R., 1986. Hemispheric lateralization of emotions: absence of electrophysiological arguments. Physiology and Behavior 40, 215220. Cronbach, L.J., Gleser, G.C., Nanda, H., Rajaratnam, N., 1972. The Dependability of Behavioral Measurements: Theory of Generalizability of Scores and Proles. Wiley, New York. Davidson, R.J., 1993. Cerebral asymmetry and emotion: conceptual and methodological conundrums. Cognition and Emotion 7, 115138. Davidson, R.J., 1998a. Affective style and affective disorders: perspectives from affective neuroscience. Cognition and Emotion 12, 307330. Davidson, R.J., 1998b. Anterior electrophysiological asymmetries, emotion, and depression: conceptual and methodological conundrums. Psychophysiology 35, 607614. Davidson, R.J., Chapman, J.P., Chapman, L.J., Henriques, J.B., 1990. Asymmetrical brain electrical activity discriminates between psychometrically-matched verbal and spatial cognitive tasks. Psychophysiology 27, 528543. Davidson, R.J., Coe, C.C., Dolski, I., Donzella, B., 1999. Individual differences in prefrontal activatin asymmetry predict natural killer cell activity at rest and in response to challenge. Brain, Behavior and Immunity 13, 93108. Davidson, R.J., Ekman, P., Saron, C.D., Senulis, J.A., Friesen, W.V., 1990. Approach-withdrawal and cerebral asymmetry: emotional expression and brain physiology I. Journal of Personality and Social Psychology 58, 330341. Davidson, R.J., Fox, N.A., 1982. Asymmetrical brain activity discriminates between positive and negative affective stimuli in human infants. Science 218, 12351236. Davidson, R.J., Fox, N.A., 1989. Frontal brain asymmetry predicts infants response to maternal separation. Journal of Abnormal Psychology 98, 127131. Davidson, R.J., Kabat-Zinn, J., Schumacher, J., Rosenkranz, M., Muller, D., Santorelli, S.F., Urbanowski, F., Harrington, A., Bonus, K., Sheridan, J.F., 2003. Alterations in brain and immune function produced by mindfulness meditation. Psychosomatic Medicine 65, 564570. Davidson, R.J., Kalin, N.H., Shelton, S.E., 1992. Lateralized effects of diazepam on frontal brain electrical asymmetries in rhesus monkeys. Biological Psychiatry 32, 438451. Davidson, R.J., Kalin, N.H., Shelton, S.E., 1993. Lateralized response to diazepam predicts temperament style in rhesus monkeys. Behavioral Neuroscience 107, 11061110. Davidson, R.J., Marshall, J.R., Tomarken, A.J., Henriques, J.B., 2000. While a phobic waits: regional brain electrical and autonomic activity in social phobics during anticipation of public speaking. Biological Psychiatry 47, 8595. Davidson, R.J., Schaffer, C.E., Saron, C., 1985. Effects of lateralized presentations of faces on self-reports of emotion and EEG asymmetry in depressed and non-depressed subjects. Psychophysiology 22, 353364. Dawson, G.D., Frey, K., Panagiotides, H., Osterling, J., Hessl, D., 1997. Infants of depressed mothers exhibit atypical frontal brain activity: a replication and extension of previous ndings. Journal of Child Psychology and Psychiatry 38, 179186. Dawson, G.D., Frey, K., Panagiotides, H., Yamada, E., Hessl, D., Osterling, J., 1999a. Infants of depressed mothers exhibit atypical frontal brain activityduring interactions with mother and with familiar, non-depressed adults. Child Development 70, 10581066. Dawson, G., Frey, K., Self, J., Panagiotides, H., Hessl, D., Yamada, E., Rinaldi, J., 1999b. Frontal brain electrical activity in infants of depressed and nondepressed mothers: relation to variations in infant behavior. Development and Psychopathology 11, 589605.
46
Dawson, G., Klinger, L.G., Panagiotides, H., Hill, D., Spieker, S., 1992. Infants of mothers with depressive symptoms: electroencephalograhic and behavioral ndings related to attachment status. Child Development 63, 725737. Dawson, G., Panagiotides, H., Klinger, L.G., Hill, D., 1992. The role of frontal lobe functioning in the development of infant self-regulatory behavior. Brain and Cognition 20, 152175. Dawson, G., Panagiotides, H., Klinger, L.G., Spieker, S., 1997. Infants of depressed and non-depressed mothers exhibit differences in frontal brain electrical activity during expression of negative emotions. Developmental Psychology 33, 650656. Debener, S., Baeudecel, A., Nessler, D., Brocke, B., Heilemann, H., Kayser, J., 2000. Is resting anterior EEG alpha asymmetry a trait marker for depression? Neuropsychobiology 41, 3137. De Souza, E.B., 1995. Corticotropin-releasing factor receptors: physiology, pharmacology, biochemistry and role in central nervous system and immune disorders. Psychoneuroendochrinology 20, 789819. Earnest, C., 1999. Single case study of EEG biofeedback for depression: an independent replication in an adolescent. Journal of Neurotherapy 3, 2835. Ehlers, C.L., Wall, T.L., Garcia-Andrade, C., Phillips, E., 2001. EEG asymmetry: relationship to mood and risk for alcoholism in mission indian youth. Biological Psychiatry 50, 129136. Ekman, P., Davidson, R.J., 1993. Voluntary smiling changes regional brain activity. Psychological Science 4, 342345. Ekman, P., Davidson, R.J., Friesen, W.V., 1990. The Duchenne smile: emotional expression and brain physiology II. Journal of Personality and Social Psychology 58, 342353. Field, T., Diego, M., Hernandez-Reif, M., Schanberg, S., Kuhn, C., 2002. Relative right versus left frontal EEG in neonates. Developmental Psychobiology 41, 147155. Field, T., Fox, N.A., Pickens, J., Nawrocki, T., 1995. Relative right frontal EEG activation in 3 to 6-month-old infants of depressed mothers. Developmental Psychology 31, 358363. Field, T., Pickens, J., Prodromidis, M., Malphurs, J., Fox, N.A., Bendell, D., Yando, R., Schanberg, S., Kuhn, C., 2000. Targeting adolescent mothers with depressive symptoms for early intervention. Adolescence 35, 381414. Fox, N.A., Bell, M.A., Jones, N.A., 1992. Individual differences in response to stress and cerebral asymmetry. Developmental Neuropsychology 8, 161184. Fox, N.A., Davidson, R.J., 1986. Taste-elicited changes in facial signs of emotion and the asymmetry of brain electrical activity in human newborns. Neuropsychologia 24, 417422. Fox, N.A., Davidson, R.J., 1987. Electroencephalogram asymmetry in response to the approach of a stranger and maternal seperation in 10-month-old infants. Develomental Psychology 23, 233240. Fox, N.A., Davidson, R.J., 1988. Patterns of brain electrical activity during facial signs of emotion in 10-month-old infants. Develomental Psychology 24, 230246. Fox, N.A., Rubin, K.H., Calkins, C.D., Marshall, T.R., Coplan, R.J., Porges, S.W., Long, J.M., Shannon, S., 1995. Frontal activation asymmetry and social competence at four years of age. Child Development 66, 17701784. Fox, N.A., Schmidt, L.A., Calkins, C.D., Rubin, K.H., Coplan, R.J., 1996. The role of frontal activation in the regulation and dysregulation of social behavior during the preschool years. Development and Psychopathology 8, 89102. Gilbert, D.G., McClernon, F.J., Rabinovich, N.E., Dibb, W.D., Plath, L.C., Hiyane, S., Jensen, R.A., Meliska, C.J., Estes, S.L., Gehlbach, B.A., 1999. EEG, physiology, and task-related mood fail to resolve across 31 days of smoking abstinence. Experimental and Clinical Psychopharmacology 7, 427443. Gilbert, D.G., Meliska, C.J., Wesler, R., Estes, S.L., 1994. Depression, personality, and gender inuence EEG, cortisol, beta-endorphin, heart rate, and subjective responses to smoking miltiple cigarettes. Personality and Individual Differences 36, 247264. Goodman, S.H., 2002. Depression and early adverse experiences. In: Gotlib, I.H., Hammen, C.L. (Eds.), Handbook of Depression. The Guilford Press, New York, NY, pp. 245267. Gotlib, I.H., Ranganath, C., Rosenfeld, J.P., 1998. Frontal EEG asymmetry, depression, and cognitive functioning. Cognition and Emotion 12, 449478. Gray, J.A., 1972. The psychophysiological basis of introversionextraversion: a modication of Eysencks theory. In: Nebylitsyn, V.D., Gray, J.A. (Eds.), The Biological Bases of Individual Behavior. Academic Press, San Diego, CA, pp. 182205. Gray, J.A., 1987. The Psychology of Fear and Stress. Cambridge University Press, Cambridge, England.
47
Hagemann, D., Naumann, E., 2001. The effects of occular artifacts on (lateralized) broadband power im EEG. Clinical Neuropsychology 112, 215231. Hagemann, D., Naumann, E., Becker, G., Maier, S., Bartussek, D., 1998. Frontal brain asymmetry and affective style: a conceptual replication. Psychophysiology 35, 372388. Hagemann, D., Naumann, E., Luerken, A., Bartussek, D., 1999. EEG trait asymmetry and affective style I: latent state and trait structure of resting asymmetry scores, Psychophysiology, S57. Hagemann, D., Naumann, E., Thayer, J.F., 2001. The quest for the EEG reference revisited: A glance from brain asymmetry research, Psychophysiology 38, 847857. Hagemann, D., Naumann, E., Luerken, A., Becker, G., Maier, S., Bartussek, D., 1999. EEG asymmetry, dispositional mood and personality. Personality and Individual Differences 27, 541568. Harmon-Jones, E., Abramson, L.Y., Sigelman, J., Bohlig, A., Hogan, M.E., Harmon-Jones, C., 2002. Proneness to hypomania/mania symptoms or depression symptoms and asymmetrical frontal cortical responses to an anger-evoking event. Journal of Personality and Social Psychology 82, 610618. Harmon-Jones, E., 2000. Relationship between anger and asymmetrical frontal cortical activity. Psychophysiology, S18. Harmon-Jones, E., Allen, J.J.B., 1997. Behavioral activation sensitivity and resting frontal EEG asymmetry: covariation of putative indicators related to risk for mood disorders. Journal of Abnormal Psychology 106, 159163. Harmon-Jones, E., Allen, J.J.B., 1998. Anger and frontal brain activity: EEG asymmetry consistent with approach motivation despite negative affective valence. Journal of Personality and Social Psychology 74, 13101316. Harmon-Jones, E., Sigelman, J., 2001. State anger and prefrontal brain activity: Evidence that insult-related relative left-prefrontal activation is associated with experienced anger and aggression, Journal of Personality and Social Psychology 80, 797803. Harmon-Jones, E., Sigelman, J.D., Bohlig, A., Harmon-Jones, C., 2003. Anger, coping, and frontal cortical activity: The effect of coping potential on anger-induced left frontal activity, Cognition and Emotion 17, 124. Heller, W., Nitschke, J.B., 1997. Regional brain activity in emotion: a framework for understanding cognition in depression. Cognition and Emotion 11, 637661. Heller, W., Nitschke, J.B., Etienne, M.A., Miller, G.A., 1997. Patterns of regional brain activity differentiate types of anxiety. Journal of Abnormal Psychology 106, 0010010. Henderson, H.A., Fox, N.A., Rubin, K.H., 2001. Temperamental contributions to social behavior: the moderating roles of frontal EEG asymmetry and gender. Journal of the American Academy of Child and Adolescent Psychiatry 40, 6874. Henriques, J.B., Davidson, R.J., 1990. Regional brain electrical asymmetries discriminate between previously depressed and healthy control subjects. Journal of Abnormal Psychology 99, 2231. Henriques, J.B., Davidson, R.J., 1991. Left frontal hypoactivation and depression. Journal of Abnormal Psychology 100, 535545. Jackson, D.C., Mueller, M.J., Dolski, I., Dalton, K.M., Nitschke, J.B., Urry, H.L., Rosenkranz, M.A., Ryff, C.D., Singer, B.H., Davidson, R.J., 2003. Now you feel it, now you dont: frontal brain electrical asymmetry and individual differences in emotion regulation. Psychological Science 14, 612617. Jacobs, G., Snyder, D., 1996. Frontal brain asymmetry predicts affective style in men. Behavioral Neuroscience 110, 36. Jones, N.A., Field, T., 1999. Massage and music therapies attenuate frontal EEG asymmetry in depressed adolescents. Adolescence 34, 529535. Jones, N.A., Field, T., Davalos, M., 1998. Massage therapy attenuates right frontal EEG asymmetry in one-month-old infants of depressed mothers. Infant behavior and Development 21, 527530. Jones, N.A., Field, T., Davalos, M., Pickens, J., 1997. EEG stability in infants/children of depressed mothers. Child Psychiatry and Human Development 28, 5970. Jones, N.A., Field, T., Fox, N.A., Davalos, M., Lundy, B., Hart, S., 1998. Newborns of mothers with depressive symptoms are physiologically less developed. Infant Behavior and Development 21, 537541. Jones, N.A., Field, T., Fox, N.A., Lundy, B., Davalos, M., 1997. EEG activation in 1-month-old infants of depressed mothers. Development and Psychopathology 9, 491505. Jones, N.A., Field, T., Fox, N.A., Davalos, M., Malphurs, J., Carraway, K., Schanburg, S., Kuhn, C., 1997. Infants of intrusive and withdrawn mothers. Infant Behavior and Development 20, 175186.
48
Jones, N.A., Fox, N.A., 1992. Electroencephalogram asymmetry during emotionally evocative lms and its relation to positive and negative affectivity. Brain and Cognition 20, 280299. Kalin, N.H., Shelton, S.E., Davidson, R.J., 2000. Cerebrospinal uid corticotropin-releasing hormone levels are elevated in monkeys with patterns of brain activity associated with fearful temperament. Biological Psychiatry 47, 579585. Kang, D.H., Davidson, R.J., Coe, C.L., Wheeler, R.E., Tomarken, A.J., Ershler, W.B., 1991. Frontal brain asymmetry and immune function. Behavioral Neuroscience 105, 860869. Kenny, D.A., Kashy, D.A., Bolger, N., 1998. Data analysis in social psychology. In: Gilbert, D.T., Fiske, S.T., Lindzey, G. (Eds.), The Handbook of Social Psychology. McGraw-Hill, Boston, pp. 233265. Kentgen, L.M., Tenke, C.E., Pine, D.S., Fong, R., Klein, R.G., Bruder, G.E., 2000. Inuence of comorbidity with anxiety disorders. Journal of Abnormal Psychology 109, 797802. Kline, J.P., Allen, J.J.B., Schwartz, G.E., 1998. Is left frontal brain activation in defensiveness gender specic? Journal of Abnormal Psychology 107, 149153. Kline, J.P., Blackhart, G.C., Joiner, T.E. Sex, lie scales and electrocaps: an interpersonal context for defensiveness and anterior electroencephalographic asymmetry. Personality and Individual Differences, in press. Kline, J.P., Blackhart, G.C., Woodward, K.M., Williams, S.R., Schwartz, G.E.R., 2000. Anterior electroencephalographic asymmetry changes changes in elderly women in response to a pleasant and unpleasant odor. Biological Psychology 52, 241250. Kline, J.P., Knapp-Kline, K., Schwartz, G.E.R., Russek, L.G.S. Anterior asymmetry, defensiveness, and perceptions of parental caring. Personality and Individual Differences, in press. Kline, J.P., LaRowe, S.D., Donohue, K.F., Minnix, J., Blackhart, G.C., 2003. Adult experimental psychopathology. In: Roberts, M.C., Ilardi, S.S., (Eds.), Methods of Research in Clinical Psychology: A Handbook. Blackwell, Malden, MA. Loehlin, J.C., 1998. Latent Variable Models: An Introduction to Factor, Path, and Structural Analysis, third ed. Lawrence Erlbaum Associates, Mahwah, NJ. Lykken, D.T., Tellegen, A., Iacono, W.G., 1982. EEG spectra in twins: evidence for a neglected mechanism of genetic determination. Phsyiological Psychology 10, 6065. MacDhomhail, S., Allen, J.J.B., Katsanis, J., Iacono, W.G., 1999. Heritability of frontal alpha asymmetry. Psychophysiology, S74. MacKinnon, D.P., Lockwood, C.M., Hoffman, J.M., West, S.G., Sheets, V., 2002. A comparison of methods to test mediation and other intervening variable effects. Psychological Methods 7, 83104. McManis, M.H., Kagan, J., Snidman, N.C., Woodward, S.A., 2002. EEG asymmetry, power, and temperament in children. Developmental Psychobiology 41, 169177. Merckelbach, H., Muris, P., Pool, K., DeJong, P., Schouten, E., 1996. Reliability and validity of a paper and pencil test measuring hemisphere preference. European Journal of Personality 10, 221231. Miller, A., Fox, N.A., Cohn, J.F., Forbes, E.E., Sherrill, J.T., Kovacs, M., 2002. Regional patterns of brain activity in adults with a history of childhood-onset depression: gender differences and clinical variability. American Journal of Psychiatry 159, 934940. Miller, A., Tomarken, A.J., 2001. Task-dependent changes in frontal brain asymmetry: effects of incentive cues, outcome expectencies, and motor responses. Psychophysiology 38, 500511. Minnix, J.A., Kline, J.P., Blackhart, G.C., Pettit, J.W., Perez, M., Joiner, T.E. Relative left frontal activity is associated with increased depression in high reassurance-seekers. Biological Psychology (Special issue), this volume. Moss, E.M., Davidson, R.J., Saron, C., 1985. Cross-cultural differences in hemisphericity: EEG asymmetry discriminates between Japanese and Westerners. Neuropsychologia 23, 131135. Nitschke, J.B., Heller, W., Palmieri, P.A., Miller, G.A., 1999. Contrasting patterns of brain activity in anxious apprehension and anxious arousal. Psychophysiology 36, 628637. Papousek, I., Schulter, G., 2001. Associations between EEG asymmetries and electrodermal lability in low vs. high depressive and anxious normal individuals. International Journal of Psychophysiology 41, 105117. Papousek, I., Schulter, G., 2002. Covariations of EEG asymmetries and emotional states indicate that activity at frontopolar locations is particularly affected by state factors. Psychophysiology 39, 350360. Petruzzello, S.J., Landers, D.M., 1994. State anxiety reduction and exercise: does hemispheric activation reect such changes? Medicine and Science in Sports and Exercise 26, 10281035.
49
Reeves, B., Lang, A., Thorson, E., Rothschild, M., 1989. Emotional television scenes and hemispheric specialization. Human Communication Research 15, 493508. Reid, S.A., Duke, L.M., Allen, J.J.B., 1998. Resting frontal electroencephalographic asymmetry in depression: inconsistencies suggest the need to identify mediating factors. Psychophysiology 35, 389404. Rosenfeld, J.P., Baehr, E., Baehr, R., Gotlib, I.H., Ranganath, C., 1996. Preliminary evidence that daily changes in frontal alpha asymmetry correlate with changes in affect in therapy sessions. International Journal of Psychophysiology 23, 137141. Sabotka, S.S., Davidson, R.J., Senulis, J.A., 1992. Anterior brain electrical asymmetries in response to reward and punishment. Electroencephalography and Clinical Neurophysiology 83, 236247. Sanders, C., Diego, M., Fernandez, M., Field, T., Hernandez-Reif, M., Roca, A., 2002. EEG asymmetry responses to lavender and rosemary aromas in adults and infants. International Journal of Neuroscience 112, 13051320. Schaffer, C.E., Davidson, R.J., Saron, C., 1983. Frontal and parietal electroencephalogram asymmetry in depressed and non-depressed subjects. Biological Psychiatry 18, 753759. Schmidt, L.A., 1999. Frontal brain electrical activity in shyness and sociability. Psychological Science 19, 316321. Schmidt, L.A., Fox, N.A., 1994. Patterns of cortical electrophysiology and autonomic activity in adults shyness and sociability. Biological Psychology 38, 183198. Schmidt, L.A., Fox, N.A., Goldberg, M.C., Smith, C.C., Schulkin, J., 1999. Effects of acute prednisone administration on memory, attention and emotion in healthy human adults. Psychoneuroendocrinology 24, 461483. Silva, J.R., Pizzagalli, D.A., Larson, C.L., Jackson, D.C., Davidson, R.J., 2002. Frontal brain asymmetry in restrained eaters. Journal of Abnormal Psychology 111, 676681. Stough, C., Donaldson, C., Scarlata, B., Ciorciari, J., 2001. Psychophysiological correlates of NEO-PI-R openness, agreeableness and concientiousness: preliminary results. International Journal of Psychophysiology 41, 8791. Sutton, S.K., Davidson, R.J., 1997. Prefrontal brain asymmetry: a biological substrate of the behavioral approach and inhibition systems. Psychological Science 8, 204210. Tomarken, A.J., Davidson, R.J., 1994. Frontal brain activation in repressors and non-repressors. Journal of Abnormal Psychology 103, 339349. Tomarken, A.J., Davidson, R.J., Henriques, J.B., 1990. Resting frontal brain asymmetry predicts affective responses to lms. Journal of Personality and Social Psychology 59, 791801. Tomarken, A.J., Davidson, R.J., Wheeler, R.E., Doss, R.C., 1992a. Individual differences in anterior brain asymmetry and fundemental dimensions of emotion. Journal of Personality and Social Psychology 62, 676687. Tomarken, A.J., Davidson, R.J., Wheeler, R.E., Kinney, L., 1992b. Psychometric properties of resting anterior EEG asymmetry: temporal stability and internal consistency. Psychophysiology 29, 576592. Tomarken, A.J., Dichter, G.S., Garber, J., Simien, C. Relative left frontal hypo-activation in adolescents at risk for depression. Biological Psychology (Special issue), this volume. Tucker, D.M., Dawson, S.L., 1984. Asymmetric changes as method actors generated emotions. Biological Psychiatry 19, 6375. Urry, H.L., Nitschke, J.B., Dolski, I., Jackson, D.C., Dalton, K.M., Mueller, C.J., Rosenkranz, M.A., Ryff, C.D., Singer, B.H., Davidson, R.J. Making a life worth living: neural correlates of well-being. Psychological Science, in press. Urry, H.L., Hitt, S.K., Allen, J.J.B., 1999. Internal consistency and test-retest stability of resting EEG alpha asymmetry in major depression, Psychophysiology 36, S116. Waldstein, S.R., Kop, W.J., Schmidt, L.A., Hauer, A.J., Krantz, D.S., Fox, N.A., 2000. Frontal electrocortical and cardiovascular reactivity during happiness and anger. Biological Psychology 55, 323. Wheeler, R.E., Davidson, R.J., Tomarken, A.J., 1993. Frontal brain asymmetry and emotional reactivity: a biological substrate of affective style. Psychophysiology 30, 8289. Wiedemann, G., Pauli, P., Dengier, W., Lutzenberger, W., Birbaumer, N., Buchkremer, G., 1999. Frontal brain asymmetry as a biological substrate of emotions in patients with panic disorders. Archives of General Psychiatry 56, 7884. Zinser, M.C., Fiore, M.C., Davidson, R.J., Baker, T.B., 1999. Manipulating smoking motivation impact on an electrophysiological index of approach motivation. Journal of Abnormal Psychology 108, 240252.
Review
Contributions from research on anger and cognitive dissonance to understanding the motivational functions of asymmetrical frontal brain activity
Eddie Harmon-Jones
Department of Psychology, University of Wisconsin-Madison, 1202 West Johnson Street, Madison, WI 53706, USA
Abstract Research has suggested that approach-related positive emotions are associated with greater left frontal brain activity and that withdrawal-related negative emotions are associated with greater right frontal brain activity. Different explanations have been proposed. One posits that frontal asymmetry is due to emotional valence (positivity/negativity), one posits that frontal asymmetry is due to motivational direction (approach/withdrawal), and one posits that frontal asymmetry is due to a combination of emotional valence and motivational direction (positive-approach/negative-withdrawal). Because research had confounded emotional valence and motivational direction, the theoretical explanation was muddled. Solely supporting the motivational direction model, recent research has revealed that anger and cognitive dissonance, emotions with negative valence and approach motivational tendencies, are related to relatively greater left frontal activity. 2004 Elsevier B.V. All rights reserved.
Keywords: Frontal EEG asymmetry; Emotion; Anger; Cognitive dissonance
1. Introduction Over the past two decades, a variety of research approaches have pointed to the importance of the left and right frontal brain regions in emotion and motivation. Research has suggested that left frontal brain activity is associated with positive emotions and approach behavior and right frontal brain activity is associated with negative emotions and withdrawal behavior. This research has created an impression that high levels of left frontal
Tel.: +1-608-265-5504; fax: +1-608-262-4029. E-mail address: eharmonj@facstaff.wisc.edu (E. Harmon-Jones).
52
E. Harmon-Jones / Biological Psychology 67 (2004) 5176
activity are preferred to low levels of left frontal activity. Several popular press articles have appeared in the New York Times, Parade Magazine, and other sources echoing the message advanced by several researchers (e.g., Fox et al., 2001; Davidson, 1998) that relatively high left frontal brain activity is more psychologically and physically healthy than relatively less left frontal brain activity. This theoretical model is widely accepted, even among researchers who are not involved in research on the frontal asymmetry (e.g., Oatley and Jenkins, 1996; Zajonc and McIntosh, 1992). Indeed, researchers have begun to use the past ndings regarding the association of left frontal activity with more positive outcomes in creating treatment strategies such as biofeedback (e.g., Baehr et al., 1997; Rosenfeld et al., 1995). However, in this article, I review research that casts doubt on the assumption that increased left frontal activity is always benecial. Although several studies have found left frontal activity to be associated with positive emotions, recent research has indicated that these ndings resulted because the past research confounded approach motivation with positive emotional valence. Appetitive motivations are not always associated with positive affects. Anger, greed, lust, and mania are some examples of approach motivations that may have deleterious consequences. In this article, I will briey review research on the relationship between emotion/motivation and asymmetrical frontal brain activity. I will then describe several theoretical explanations of the research results. After reviewing the various theoretical explanations, I will review recent research on the emotion of anger and the emotive state of cognitive dissonance that has favored the motivational direction view over the other views. Finally, I broadly consider this recent research for theories and research on emotion.
2. Asymmetrical frontal brain activity and emotion Broadly speaking, research on frontal asymmetry has proceeded along three lines: (1) examinations of the relationship between behavioral or experiential indices of trait affect/motivation and resting electroencephalograph activity (EEG); (2) examinations of the relationship between resting EEG and responses to emotion-eliciting stimuli; and (3) examinations of EEG changes during exposure to emotionally evocative situations.1 To begin, I will briey review representative ndings from each of these empirical approaches.
1 Much of the research on asymmetrical frontal brain activity has assessed the activity using alpha frequency band activity derived from the electroencephalograph (EEG). Research has revealed that alpha power is inversely related to regional brain activity using hemodynamic measures (Cook et al., 1998) and behavioral tasks (Davidson et al., 1990). Additional data from individuals with brain damage supports the research reviewed in this article (e.g., Robinson and Downhill, 1995). Because the majority of research on the frontal asymmetry has used EEG alpha power, the review focuses on this research. In addition, in the review, I use the term brain activity to refer to the inverse of alpha power, as is commonly done in this literature. Moreover, I reserve the use of the term activation to refer to state-induced changes in EEG, whereas activity can refer to state or trait (baseline) EEG. Finally, the term relative left frontal activity (or activation) is used to describe greater left than right frontal activitya difference or asymmetry score.
53
3. Examinations of the relationship between indices of trait affect/motivation and resting EEG 3.1. Depression Depression has been found to relate to resting frontal asymmetrical activity, with depressed individuals showing relatively less left than right frontal brain activity. This relationship between depression and asymmetrical frontal activity has been found in individuals identied by self-report indices of depression (Jacobs and Snyder, 1996; Schaffer et al., 1983), and individuals identied through clinical interviews (Allen et al., 1993). Moreover, relatively less left frontal activity has been found in individuals who were previously clinically depressed but were in remission status (Henriques and Davidson, 1990). 3.2. Positive and negative affect Other research has revealed that trait positive affect is associated with greater left than right frontal brain activity, whereas trait negative affect is associated with greater right than left frontal brain activity (e.g., Tomarken et al., 1992a). In this past research, trait positive affect and negative affect were assessed using the Positive and Negative Affect Schedule (Watson et al., 1988). These affects are best described as activated positive affect and activated negative affect (Watson et al., 1999), as the measures include words such as interested and active on the positive affect scale, and afraid and distressed on the negative affect scale. 3.3. Behavioral activation/behavioral inhibition Other research has found that trait behavioral activation sensitivity (BAS) relates to greater left than right frontal brain activity (Coan and Allen, 2003; Harmon-Jones and Allen, 1997; Sutton and Davidson, 1997). In this research, BAS was measured by Carver and Whites (1994) behavioral inhibition sensitivity (BIS)/BAS questionnaire, which includes items such as When I want something, I usually go all-out to get it. Regarding the relationship of behavioral inhibition sensitivity (BIS; I worry about making mistakes.), studies have produced inconsistent results, with one nding a signicant relationship between BIS and greater right than left frontal activity (Sutton and Davidson, 1997) and the two others nding a non-signicant relationship (Coan and Allen, 2003; Harmon-Jones and Allen, 1997). It is possible that BIS is not equivalent to withdrawal motivation (see Harmon-Jones and Allen, 1997).
4. Examinations of the relationship between resting EEG and responses to emotion-eliciting stimuli Resting baseline frontal asymmetrical activity has been found to predict emotional responses. Individuals with relatively greater right than left frontal activity exhibit larger negative affective responses to negative emotion-inducing lms (fear and disgust) and smaller
54
positive affective responses to positive emotion-inducing lms (happiness) (Tomarken et al., 1990; Wheeler et al., 1993). In a related vein, research has found that resting baseline frontal asymmetrical activity predicted evaluative responses to merely exposed stimuli (Harmon-Jones and Allen, 2001). That is, individuals with relative right frontal activity reported more favorable attitudes toward familiarized stimuli (safety) than did individuals with relative left frontal activity. Other research has found that relative right frontal activity at baseline predicts crying in response to maternal separation in 10-month-old infants (Davidson and Fox, 1989). Although these effects are based on correlational evidence and hence subject to alternative explanations, recent research has more strongly suggested that the frontal asymmetry is causally involved in the production of these emotional responses. In this experiment, biofeedback training was used to manipulate the frontal asymmetry (Allen et al., 2001). Participants were randomly assigned to receive biofeedback training designed to increase right frontal relative to left frontal activity or to receive training in the opposite direction. Five consecutive days of biofeedback training provided signals of reward (300 Hz reward tone) or nonreward (150 Hz nonreward tone) depending on whether the difference between right and left frontal activity exceeded a criterion value. Systematic alterations of frontal asymmetry were observed as a function of biofeedback training. Moreover, subsequent self-reported affect in response to emotionally evocative lm clips were signicantly inuenced by the direction of biofeedback training. That is, individuals trained to increase left frontal activity reported more positive affect in response to the happy lm clip than individuals trained to increase right frontal activity. These results suggest that the frontal asymmetry can be altered using biofeedback training and that this alteration can affect emotional responses. Taken together with previous research, this experiment suggests that the frontal asymmetry is causally involved in the production of emotional experience.
5. Examinations of EEG activity during exposure to emotionally evocative situations Research has also demonstrated that asymmetrical frontal brain activity is associated with state emotional responses. For instance, Davidson and Fox (1982) found that 10-month-old infants exhibited increased left frontal activation in response to a lm clip of an actress generating a happy facial expression as compared to responses exhibited to a lm clip of an actress generating a sad facial expression. Frontal brain activity has been found to relate to facial expressions of positive and negative emotions, as well. For example, Ekman and Davidson (1993) found increased left frontal activation during voluntary facial expressions of smiles of enjoyment (i.e., activation of zygomatic major with concurrent activation of orbicularis oculi, pars lateralis) as compared to voluntary facial expressions of smiles not associated with enjoyment (i.e., activation of zygomatic major without orbicularis oculi, pars lateralis). More recently, Coan et al. (2001) found that voluntary contractions of the facial musculature to form a happy facial expression produced relatively greater left frontal activity, and that voluntary contractions of the facial musculature to form a fearful facial expression produced relatively less left frontal activity.
55
6. Explanations of the relationship between asymmetrical frontal brain activity and emotion Primarily, there have been three conceptual models designed to explain the observed results. The rst view has posited that the left frontal brain region is involved in the experience and expression of positive emotion and that the right frontal brain region is involved in the expression and experience of negative emotion (e.g., Ahern and Schwartz, 1985; Gotlib et al., 1998; Heller, 1990; Heller and Nitschke, 1998; Silberman and Weingartner, 1986). Indeed, this theoretical model is widely accepted (e.g., Oatley and Jenkins, 1996; Zajonc and McIntosh, 1992). Most of the results can be explained with this model, which I refer to as the valence model. A second view has posited that the left frontal brain region is involved in expression of approach-related emotions and that the right frontal brain region is involved in expression withdrawal-related emotions (Davidson, 1995; Fox, 1991; Harmon-Jones and Allen, 1997; Sutton and Davidson, 1997). Again, the obtained results can be accommodated by this model, which I refer to as the motivational direction model. That is, the affects and emotions that have been examined in the research are all associated with approach or withdrawal motivation. The approach-related emotions have been found to be associated with relatively greater left frontal activity, whereas the withdrawal-related emotions have been found to be associated with relatively greater right frontal activity. A third view has posited that the left frontal brain region is involved in the expression and experience of positive, approach-related emotions and that the right frontal brain region is involved in the expression and experience of negative, withdrawal-related emotions (Davidson, 1998; Tomarken and Keener, 1998). The obtained results can also be accommodated by this model, which I refer to as the valenced motivation model. That is, the positive affects and emotions that have been examined in the research are all associated with approach motivation, and the negative affects and emotions that have been examined are all associated with withdrawal motivation. Because the previously conducted research confounded the valence of the emotion with the direction of motivation, it is unable to address whether the frontal asymmetry reects the valence of the emotion, the direction of the motivation, or a combination of valence and motivation. Often, positive emotion is associated with approach-related motivation, whereas negative emotion is associated with withdrawal-related motivation. Indeed, most contemporary theories of emotion posit that positive emotion is always associated with approach motivation and that negative emotion is always associated with withdrawal motivation emotion (e.g., Watson, 2000; for a different point of view, see Carver, 2001). However, not all emotions behave in accord with this presumed relationship between the valence of emotion and direction of motivation. Anger is one of the best examples of a violation of the relationship, because anger is negative in valence (e.g., Lazarus, 1991; Watson et al., 1999), but it often evokes approach motivation (e.g., Berkowitz, 1999; Darwin, 1872/1965; Plutchik, 1980; Young, 1943). To address the primary emotional/motivation functions of asymmetrical frontal brain activity, my colleagues and I have been examining the emotion of anger, as it is often considered a negative emotion that evokes approach motivational tendencies. By examining the
56
emotion of anger, we are in a position to answer precisely what the emotional/motivational functions of asymmetrical frontal brain activity are.
7. Anger and approach motivation Before reviewing the research on anger and asymmetrical frontal activity, it is important to consider whether anger is associated with approach motivation. Several lines of research suggest that anger elicits behavioral approach or approach motivation tendencies. In what follows, I briey review representative research supporting the idea that anger is associated with approach inclinations. 7.1. Behavioral evidence In the animal behavior literature, a distinction has been made between offensive or irritable aggression and defensive aggression (Flynn et al., 1970; Moyer, 1976). It has been posited that irritable aggression results from anger and that pure irritable aggression involves attack without attempts to escape from the object being attacked (Moyer, 1976, p. 187). A number of aggression researchers have suggested that offensive aggression is associated with anger, attack, and no attempts to escape, whereas defensive aggression is associated with fear, attempts to escape, and attack only if escape is impossible (Blanchard and Blanchard, 1984; Lagerspetz, 1969; Moyer, 1976). In demonstrating that organisms evidence offensive aggression and that this is an approach behavior, Lagerspetz (1969) found that under certain conditions mice would cross an electried grid to attack another mouse. In other research with adult humans, Baron (1977) demonstrated that angry individuals are reinforced positively by signs of their tormentors pain. The participants who had been deliberately provoked by another individual had a sanctioned opportunity to assault him in return. Indications that their rst attacks were hurting their target led to intensied aggression even though the unprovoked participants reduced the intensity of their punishment at learning of the others pain. The initial signs of their victims suffering showed the angry persons they were approaching their aggressive goal and thus evoked even stronger assaults from them. Other research is consistent with these ndings (e.g., Berkowitz et al., 1981). In addition, Lewis et al. (1992) found that infants who expressed anger during extinction maintained interest during subsequent relearning, whereas infants who expressed sadness during extinction evidenced decreased interest during relearning. Thus, subsequent to frustrating events, anger may maintain and increase task engagement and approach motivation. 7.2. Subjective evidence Additional support for the idea that anger is associated with approach motivation comes from research testing the conceptual model that integrated reactance theory with learned helplessness theory (Wortman and Brehm, 1975). According to this model, how individuals respond to uncontrollable outcomes depends on their expectation of being able to control the outcome and the importance of the outcome. When an individual expects to be able to control outcomes that are important, and those outcomes are found to be uncontrollable,
57
psychological reactance should be aroused. Thus, for individuals who initially expect control, the rst few bouts of uncontrollable outcomes should arouse reactance, a motivational state aimed at restoring control. After several exposures to uncontrollable outcomes, these individuals should become convinced that they cannot control the outcomes and should show decreased motivation (i.e., learned helplessness). In other words, reactance will precede helplessness for individuals who initially expect control. In one study testing this model, individuals who exhibited angry feelings in response to one unsolvable problem had better performance and were presumably more approach motivated on a subsequent cognitive task than did participants who exhibited less anger (Mikulincer, 1988). Other research has revealed that state anger relates to high levels of self-assurance, physical strength, and bravery (Izard, 1991), inclinations associated with approach motivation. Additionally, Lerner and Keltner (2001) found that anger (both trait and state) is associated with optimistic expectations, whereas fear is associated with pessimistic expectations. Moreover, happiness was associated with optimism, making anger and happiness appear more similar to each other in their relationship with optimism than fear and anger. Although Lerner and Keltner (2001) interpreted their ndings as being due to the appraisals associated with anger, it seems equally plausible that it was the approach motivational character of anger that caused the relationship of anger and optimism. That is, anger creates optimism because anger engages the approach motivational system and produces greater optimistic expectations. 7.3. Hormonal and physiological evidence Further evidence supporting the conceptualization of anger as involving approach and not withdrawal comes from research on testosterone, which has been found to be associated with anger and aggression in humans (e.g., Olweus, 1986). In this research, testosterone treatments have been found to decrease withdrawal (fear) responses in a number of species (e.g., Boissy and Bouissou, 1994; Vandenheede and Bouissou, 1993). Other research has demonstrated that damage to the amygdala, a brain region involved in defensive behavior, has no effect on offensive aggression but reduces reactivity to nonpainful threat stimuli (Blanchard and Takahashi, 1988; Busch and Bareld, 1974). 7.4. Individual differences evidence Other evidence supporting the idea that anger is associated with an approach-orientation comes from research on bipolar disorder. The emotions of euphoria and anger often occur during manic phases of bipolar disorder (Cassidy et al., 1998; Depue and Iacono, 1989; Tyrer and Shopsin, 1982). Both euphoria and anger may be approach-oriented processes, and a dysregulated or hyperactive approach system may underlie mania (Depue and Iacono, 1989; Fowles, 1993). Research suggests that hypomania/mania involves increased left frontal brain activity and approach motivational tendencies. In this research, it has been found that individuals who have suffered damage to the right frontal cortex are more likely to evidence mania (see review by Robinson and Downhill, 1995). Thus, this research is consistent with the view that mania may be associated with increased left frontal activity and increased approach tendencies, because the approach motivation functions of the left frontal cortex are released and not restrained by the withdrawal system in the right frontal
58
cortex. Furthermore, lithium carbonate, a treatment for bipolar disorder, reduces aggression (Malone et al., 2000), suggesting that anger and aggression correlate with the other symptoms of bipolar disorder. In addition, trait anger has been found to relate to high levels of assertiveness and competitiveness (Buss and Perry, 1992). These lines of research suggest that anger is associated with a number of approach-related individual differences characteristics. Recently, two additional individual differences studies were conducted to test the hypothesis that trait anger is related to trait approach motivation, or more specically, trait BAS (Harmon-Jones, 2003). In both studies, trait BAS, as assessed by Carver and Whites (1994) scale, was positively related to trait anger at the simple correlation level, as assessed by the Buss and Perry (1992) aggression questionnaire. One of the two studies found that trait anger also related to trait behavioral inhibition sensitivity at the simple correlation level. In both studies, general negative affect was statistically controlled. This was done because angers association with general negative affect (Berkowitz, 1999, 2000; Watson, 2000) may cause the association of BIS and anger. That is, the affect of anger has two subcomponents: a nonspecic component that reects the contribution of general negative affect (Berkowitz, 1999; Watson, 2000) and a more specic component that reects the unique qualities of anger (Watson, 2000). In other words, at the simple correlation level, anger may be associated with BIS, but when controlling for negative affect, anger will not be associated with BIS but will only be associated with BAS. Results supported this prediction in both studies. Additional results in Study 2 revealed that BAS was positively correlated with physical aggression, and simultaneously regressing aggression onto BAS, BIS, and general negative affect revealed that physical aggression was positively related to BAS, negatively related to BIS, and positively related to negative affect. These results support the hypothesis that anger is related to approach motivation and strongly challenge theoretical models that assume that approach motivation is only associated with positive affect. 7.5. Summary of section Because of the large body of evidence suggesting that anger is often associated with approach motivation, we examined the relationship between anger and relative left frontal activation to test whether the frontal asymmetry is due to emotional valence, motivational direction, or a combination of emotional valence and motivational direction.
8. Anger and asymmetrical frontal brain activity 8.1. Asymmetrical frontal activity and trait anger In one of the rst studies examining the relationship between anger and asymmetrical frontal brain activity (Harmon-Jones and Allen, 1998), EEG of young adolescents (M age = 13 years) was recorded as they sat quietly for 6 min. Trait anger was measured using the Buss and Perry (1992) aggression questionnaire. Results indicated that trait anger was positively related with relatively greater left than right frontal brain activity. Moreover, additional analyses revealed that high levels of trait anger were associated with increased left frontal
59
activity and decreased right frontal activity. These results suggest that the frontal asymmetry is associated with motivational direction (approach versus withdrawal) rather than emotional valence. In addition, in this study, general activated positive and negative affect, as assessed by the Positive and Negative Affect ScheduleChildrens version (PANAS-C; Laurent et al., 1994), related to the frontal asymmetry in magnitudes similar to those found in previous research (Sutton and Davidson, 1997). That is, positive affect related to relative left frontal activity and negative affect related to relative right frontal activity. Presumably, positive affect and anger related to relative left frontal activity because both emotions are approach-related. Moreover, controlling for positive and negative affect, separately and together, did not alter the magnitude of the angerfrontal asymmetry relationship. These results suggest that anger is related to relative left frontal activity, independent of general activated positive and negative affect. A second study was conducted to assess the validity of an alternative explanation for the relationship between anger and relative left frontal brain activity (Harmon-Jones, 2004). According to this explanation, individuals with high levels of trait anger might regard anger as a positive emotion, and this positive feeling or attitude toward anger is responsible for anger being associated with relative left frontal activity. To understand why anger might be a positive emotion, it is important to consider what is meant by positivity and negativity of emotion. Emotions can be regarded as positive or negative because of the conditions that evoked the emotion or because of the emotions subjective feel. The emotion of anger can be viewed as negative when considering the conditions that evoked the emotion, and positive or negative when considering the subjective feel or evaluation of the emotion. Indeed, past writers have suggested in line with this latter denition that some individuals take considerable pleasure in the experience of anger (Ekman and Friesen, 1975, p. 81). To address whether the relationship of anger with increased left frontal activation could be explained by the fact that individuals with high levels of trait anger enjoy the experience of anger more than individuals with low levels of trait anger, studies were conducted to rst create and assess the reliability and validity of a scale that would assess attitude toward anger and second to examine the relationship of attitude toward anger with the resting frontal asymmetry (Harmon-Jones, 2004). A reliable and valid measure of feelings toward anger was created, and then it and a measure of trait anger were related to the resting frontal asymmetry. Results indicated that trait anger related to relative left frontal activity2 ,
2 Two other studies have provided additional examinations of the association of anger with the frontal EEG asymmetry. However, the results were not strongly supportive of an anger-left frontal activity relationship. In one of these experiments (Dawson et al., 1992), 21-month-old infants facial expressions were recorded during emotion-eliciting situations. During expressions of anger, relatively greater left than right frontal activity occurred, and during expressions of sadness, relatively greater right than left activity occurred. However, the frontal asymmetry was signicantly different from neutral expressions only for sad expressions and not for angry expressions. In another experiment (Fox and Davidson, 1988), 10-month-old infants evidenced greater left than right frontal activity during angry facial expressions not accompanied by crying, and greater right than left frontal activity during angry facial expressions accompanied by crying. Effects of identical direction occurred for sad facial expressions; that is, greater left than right frontal activity during sad facial expressions not accompanied by crying, and greater right than left frontal activity during sad facial expressions accompanied by crying. Because sadness and anger showed similar relations with the frontal asymmetry, it is difcult to infer from these effects whether the frontal asymmetry reects valence or direction. Finally, both of these studies were correlational and subject to interpretational difculties associated with correlational studies.
60
replicating the previous study (Harmon-Jones and Allen, 1998). Moreover, while trait anger was directly associated with a more positive attitude toward anger, the attitude toward anger did not relate to relative left frontal activity. In addition, statistically controlling for attitude toward anger did not alter the magnitude of the relationship between trait anger and relative left frontal activity. This study suggested that the relationship between trait anger and relative left frontal activity was not due to relative left frontal activity being associated with a more positive feeling toward anger. In addition, this study replicated the results of the previous one, but with a sample of young adults, whereas the rst study examined adolescents who might be less controlled and more likely to translate anger into approach behavior. That anger related with relative left frontal activity in young adults (M age 19 years) suggests that the relationship is not limited to young adolescents. 8.2. Asymmetrical frontal activity and state anger While evidence suggests that anger is a negative and approach-oriented emotion, and that trait anger is related to relative left frontal activity, it is not known how state anger relates to asymmetrical frontal brain activity. Moreover, the trait anger-relative left frontal evidence presented thus far is entirely correlational and subject to the interpretational difculties associated with correlational results. Because past research has found asymmetrical frontal activity to be responsive to manipulations of positive-approach and negative-withdrawal states (see above review as well as Hagemann et al., 2002), it was therefore important to examine whether manipulated anger would increase relative left frontal activity. In this state anger research, we also examined the relationship between anger-related asymmetrical frontal activity and aggression. Emotions such as anger can be conceived of as having motivational functions and as generating action tendencies (Brehm, 1999; Frijda, 1986). Of course, emotions may generate action tendencies that may not be manifest in overt behavior. However, anger often generates approach-related action tendencies that are generally aimed at resolving the anger-producing event. In the case of an insult, the action tendency may be aggression. If anger-induced relative left frontal activity is involved in approach motivational processes, then greater anger-induced left frontal activity may relate to increased aggression. Our research on trait aggression and asymmetrical frontal activity has yielded some support for this prediction, in that relative left frontal activity at baseline has been found to relate to somewhat greater self-reported trait aggression (Harmon-Jones and Allen, 1998). To assess the relationship of relative left frontal activity with aggression, we included a behavioral measure of aggression. We predicted that relative left frontal activity in response to an anger-eliciting event would relate to increased aggression. To test these predictions, participants were randomly assigned to a condition in which they were insulted or not insulted (Harmon-Jones and Sigelman, 2001). They were informed that the study concerned personality, psychophysiology, and perception. They were also told that another participant was in another room with another experimenter, and that the study would be conducted in connection with this other participants study. Then, they were told that there were two perception studies, the rst involving person perception and the second involving taste perception. In the rst ostensible study, participants wrote an essay on a social issue that they found important and they argued in support of the side of the issue
61
they favored (e.g., legal drinking age). Then, the other ostensible participant supposedly read and evaluated their essay. Finally, participants read the feedback provided by the other participant. The feedback was designed to be insulting or not. Immediately after the feedback manipulation, EEG was recorded. Then, participants completed the second perception study that involved taste perception. This study allowed us to obtain a behavioral measure of aggression. In this study, participants were told that it was very important for experimenters to remain blind to the type of tastes to which participants are exposed in taste perception studies. The experimenter explained that one way to keep experimenters blind to the tastes is to have one participant assign the tastes to the other participant. He also explained that the other participant would have to drink all that he is given. He then showed that participant six types of beverages, which consisted of 11 oz of water with 1, 2, or 3 teaspoons of sugar, apple juice, lemon juice, salt, vinegar, or hot sauce mixed into the water. Thus, each of the six types of beverages had three concentration levels. It was explained that most persons nd the sugar water most pleasant and the hot sauce most unpleasant, and that the other beverages were rated in between these two extremes, with those closer to sugar being more pleasant and those closer to hot sauce being more unpleasant (presented in the following order: sugar, apple juice, lemon juice, salt, vinegar, hot sauce). Participants were told to select one of the six types of beverages for the other participant, to pour some of each of the three concentrations into cups, and to cover the cups with lids when done. The participants were also told to label the concentration level on the bottom of each cup. The experimenter indicated that the participants may choose which type of beverage to administer and how much to administer to the other participant. The participants were also given a black sheet to cover the unused beverages when they were nished administering the beverages, to keep the experimenter blind to the type of beverage they chose. Aggression was calculated by assigning each beverage a value that corresponded to its unpleasantness. This measure of aggression is similar to a technique developed by other researchers (Lieberman et al., 1999; McGregor et al., 1998). Results indicated that participants in the insult condition reported feeling more angry and were more aggressive than participants in the no insult condition. More importantly, participants in the insult condition evidenced greater relative left frontal activity than participants in the no insult condition. Finally, within the insult condition, participants who evidenced greater relative left frontal activity in response to the insult reported feeling more angry and behaved more aggressively. This research supports the prediction that manipulated anger causes increased relative left frontal brain activity. In conjunction with trait-based research (Harmon-Jones, 2004; Harmon-Jones and Allen, 1998), the research demonstrates that asymmetrical frontal brain activity reects motivational direction rather than emotional valence. In addition, results suggested that relative left frontal activity during an anger-evoking situation related to behavioral aggression. In addition to the reviewed evidence, other research is consistent with the hypothesis that anger is associated with left frontal activity.2 For example, DAlfonso et al. (2000) recently used slow repetitive transcranial magnetic stimulation (rTMS) to inhibit the left or right prefrontal cortex. They found that rTMS of the right prefrontal cortex caused selective attention towards angry faces whereas rTMS of the left prefrontal cortex caused selective
62
attention away from angry faces. Because slow rTMS produces inhibition of cortical excitability, these results suggests that the rTMS of the right prefrontal cortex decreases its activation and caused the left prefrontal cortex to become more active. The same holds true for rTMS of the right prefrontal cortex and activation of the left prefrontal cortex. The increase in activation left prefrontal led participants to attentionally approach angry faces, as in an aggressive confrontation. In contrast, the increase in activation right prefrontal led participants to attentionally avoid angry faces, as in a frightening confrontation. The interpretation of these results, which dAlfonso et al. advanced, concurs with other research that has demonstrated that attention toward angry faces is associated with high levels of self-reported anger and that attention away from angry faces is associated with high levels of cortisol (van Honk et al., 1998, 1999, 2001), which is associated with fear. 8.3. Manipulating the intensity of approach motivation in an anger-evoking situation According to the motivational direction model of asymmetrical frontal activity, approach motivation is related to left frontal activity and withdrawal motivation is related to right frontal activity. Thus, increased left frontal activation occurs in response to anger-inducing situations because the increase in relative left frontal activity increases approach motivational tendencies that would assist in behavior that may rectify the anger-inducing situation. From this perspective, it follows that if no approach behavior could be taken to deal with the anger-provoking situation, then this increase in relative left frontal activation should be less. In other words, if approach and withdrawal motivational tendencies do underlie asymmetrical frontal activity, then alterations in motivational intensity should affect the degree of activation in the frontal brain regions. Several motivational theories posit that the expectancy of success or perceived task difculty should affect motivational intensity (for reviews, see Brehm and Self, 1989; Wright and Kirby, 2001). For the emotion of anger, if a situation creates anger and the individual believes that she can successfully act to alter the situation, then motivational intensity should be relatively high. If, on the other hand, the individual believes that no action can be taken, then motivational intensity should be relatively low. A similar prediction follows from the idea of secondary coping (Lazarus, 1991). Negative emotions including anger, sadness, guilt and fear occur when persons nd themselves in aversive situations. According to Lazarus (1991), the type of negative emotion evoked by a situation may be determined by coping potentialhow persons appraise their ability to deal with the aversive situation. If something can be done to resolve the situation, then anger, an active and negative emotion, should be aroused. In contrast, if nothing can be done to resolve the situation, then a passive and negative emotion, sadness, may be aroused. Theorists make similar predictions for appraisals of power (Roseman, 1991), power and control (Scherer, 1993), and likelihood of reinstatement of the goal state (Levine, 1995). Unfortunately, little research has addressed whether appraisals of higher coping potential lead to more anger. However, research by Levine (1995) found that, when 5-year-old children were presented with scenarios in which a child experienced a negative outcome, they expected the protagonist to experience more anger and less sadness when they judged the possibility of goal reinstatement more likely and less anger and more sadness when they judged goal reinstatement less likely.
63
Based on the integration of ideas from the motivational model of asymmetrical frontal activity with theories of motivational intensity and how coping potential relates to anger, we predicted that greater left frontal activation would occur in response to an anger-producing event when persons believe that action can be taken to resolve the situation as compared to when persons believe that no action can be taken to resolve the situation. To test these predictions, university students who paid a sizable portion of their tuition and who were opposed to a tuition increase were invited to an experiment ostensibly concerned with reactions to pilot radio broadcasts. They then heard an editorial in which the speaker argued forcefully for a tuition increase. Immediately prior to hearing the editorial, participants were informed that the tuition increase may occur in the future and that petitions were being circulated to attempt to prevent the increase (action-possible condition), or they were informed that the tuition increase would denitely occur (action-impossible condition). Immediately after listening to the editorial, EEG was recorded, and then participants completed a self-report emotion questionnaire. Finally, participants in the action-possible condition were given the opportunity to sign a petition and take as many petitions as they wanted to have others sign. Results revealed that participants in the action-possible condition evidenced greater relative left frontal activity than did participants in the action-impossible condition. Moreover, within the action-possible condition, this increase in relative left frontal activity directly related to self-reported anger and behaviors aimed at rectifying the anger-producing event (i.e., whether or not they signed the petition and number of petitions taken). Interestingly, self-reported anger did not differ between the action-possible and action-impossible conditions. Both conditions reported feeling much more angry after hearing the editorial as compared to before hearing the editorial. These results suggest that the appraisal of coping potential inuenced relative left frontal activity but not angry feelings (Harmon-Jones et al., 2003b). 8.4. Individual differences that predict increased relative left frontal activity during anger The research reviewed thus far suggests that trait and state anger are related to relative left frontal activity when anger is associated with approach motivational tendencies. As mentioned earlier, past research has revealed that individual differences in behavioral approach sensitivity are related to relatively greater left frontal activity (Harmon-Jones and Allen, 1997; Sutton and Davidson, 1997). Thus, greater BAS should predispose individuals to respond with greater left frontal activity when angered. Indeed, the BAS has been posited to be involved in predatory aggression (Depue and Iacono, 1989; Gray, 1982). In addition, the BAS has been proposed to underlie types of psychopathology, with depression involving a hypoactive BAS (Fowles, 1988, 1993) and mania/hypomania involving a hyperactive BAS (Depue and Iacono, 1989; Depue et al., 1987; Meyer et al., 1999). Based on these ideas, research examined the relationship between proneness to hypomania/mania and anger-related left frontal activity (Harmon-Jones et al., 2002). In the study, individuals with proneness toward hypomania/mania or depression symptoms were exposed to the anger-evoking radio broadcast used in the previously mentioned study (Harmon-Jones et al., 2003b). Results indicated that individuals with proneness toward hypomania/mania
64
evidenced greater left frontal activation when confronted with the anger-evoking situation, whereas individuals with proneness toward depression symptoms evidenced less left frontal activation when confronted with the same anger-evoking situation. These results support predictions derived from models relating mania/hypomania to BAS activity (Depue and Iacono, 1989; Fowles, 1988, 1993), models of the motivational functions of asymmetrical frontal activity (Harmon-Jones and Allen, 1997; Sutton and Davidson, 1997), and models that consider anger as part of the BAS (Depue and Iacono, 1989; Harmon-Jones and Allen, 1998, 2001; Harmon-Jones and Sigelman, 2001; Harmon-Jones et al., 2003b). This research extends the past research on anger and frontal brain activity by revealing individual difference characteristics that predict who is more likely, as well as less likely, to respond with increased left frontal activity in anger-inducing situations. That is, individuals with proneness toward hypomania/mania symptoms evidence greater relative left frontal activity, whereas individuals with proneness toward depressive symptoms evidence lesser relative left frontal activity when confronted with an anger-evoking event. From these results, it seems plausible to predict that proneness toward hypomania/mania symptoms may predispose persons toward responding with increased approach (and decreased withdrawal) motivational tendencies given challenging or frustrating situations, whereas proneness toward depressive symptoms may predispose persons toward responding with decreased approach (and increased withdrawal) motivational tendencies given these same situations. In other words, proneness toward hypomania/mania symptoms may lead to reactance-like responses and proneness toward depressive symptoms may lead to helpless responses in the face of challenges (e.g., Abramson et al., 1989; Mikulincer, 1988; Wortman and Brehm, 1975). 8.5. On the reduction of anger-related left frontal activity Whereas the reviewed evidence cogently links approach anger to relative left frontal activity, it would be important to establish that manipulations that reduce angry approach behaviors also reduce relative left frontal activity. To address this issue, we tested whether sympathy would reduce the relative left frontal activity typically observed during anger. Past research has suggested that experiencing sympathy for another individual can reduce aggression toward that individual (e.g., see review by Miller and Eisenberg, 1988). We hypothesized that sympathy may reduce aggression by reducing the relative left frontal activity associated with anger. To test this hypothesis, participants were told that the study concerned personality, perception, and brain activity, and that they and another student would be writing essays and evaluating each other based on the essays. Participants then wrote a persuasive essay arguing either for or against a 10% tuition increase. Then, the experimenter returned to the participants room and handed them a folder containing a reading perspective, an essay, and a questionnaire. The reading perspective instructions asked participants to remain completely objective (low sympathy) or to try to imagine how the other person must feel (high sympathy), as in much past sympathy research (Batson, 1991; Harmon-Jones et al., 2003a). The participant then read the essay ostensibly written by the other participant. In the essay, the other participant described his/her difculties with having multiple sclerosis. Following the reading of the essay, the participant received an evaluation of his/her essay ostensibly written by the
65
other participant. The evaluation contained either neutral ratings and comments (no insult) or insulting ratings and comments (insult). Immediately after feedback manipulation, EEG was collected. Then, the participant completed a questionnaire assessing impressions of the other participant and a questionnaire assessing emotions. Results indicated that the insult condition evoked greater self-reported anger than the no-insult condition. Also, the high sympathy condition evoked greater self-reported sympathy than the low-sympathy condition. No signicant insult X sympathy condition interactions emerged for either anger or sympathy. Moreover, no signicant condition differences emerged for self-reported sadness, happiness, fear, or distress. More importantly, the insult condition evoked greater left frontal activity but only when high levels of sympathy were not rst evoked for the insulting person. That is, the low sympathy/insult condition produced greater relative left frontal activity than every other condition. In addition, the low sympathy/insult condition evoked greater left frontal activity and lesser right frontal activity than every other condition, when separate estimates of left and right frontal activity were examined using methods suggested by Wheeler et al. (1993). Thus, when participants rst experienced sympathy for the target person, they did not evidence increased left frontal activity when insulted. Moreover, they expressed less hostile attitudes toward the insulting person than did participants who did not rst experience sympathy for the insulter. That is, the low sympathy/insult condition differed from each of the other conditions. The experiment thus suggested that the alteration of relative left frontal activity via sympathy can reduce angry aggression.
9. Dissonance and left frontal activity The anger research has strongly supported the motivational direction model over the emotional valence model of the frontal asymmetry. In addition to this anger research, some recent evidence suggests that cognitive dissonance is associated with left frontal activity. Because cognitive dissonance is associated with negative affect (e.g., Elliot and Devine, 1994; Harmon-Jones, 2000a; Zanna and Cooper, 1974; for a review, see Harmon-Jones, 2000b), this research further supports the motivational direction model over the valence model. The idea that dissonance caused by commitment to a course of action might be associated with left frontal activity was derived from the action-based model of cognitive dissonance (Harmon-Jones, 1999; Harmon-Jones and Harmon-Jones, 2002). The action-based model proposes that inconsistency between cognitions makes persons uncomfortable because inconsistency has the potential to interfere with effective action. From the viewpoint of the action-based model, cognitions are important because they guide the actions of an organism. When an individual holds two relatively important cognitions that are inconsistent, the potential to act in accord with them is undermined. To reduce the inconsistency and resulting negative affect, individuals engage in a variety of cognitive strategies. For example, dissonance results when one freely chooses to engage in behavior that is inconsistent with an attitude or belief. The free choice is subtly induced by the experimenter in experimental research. Numerous experiments have found that when individuals engage in such behavior, they often change their attitudes to be consistent with their recent behavior.
66
In other research, it has been found that after making difcult decisions (which have been found to cause dissonance), individuals value the chosen alternative and devalue the rejected alternative more than they did prior to the decision (for reviews, see Beauvois and Joule, 1996; Brehm and Cohen, 1962; Harmon-Jones, 1999). In both of these dissonance-evoking situations, dissonance occurs because there are cognitions that are inconsistent with a chosen course of action. That is, in the former situation, the past attitude is inconsistent with the current behavior. In the latter situation, the positive aspects of the rejected alternative and the negative aspects of the chosen alternative are inconsistent with the decision. The dissonance thus has the potential of interfering with the translation of the decision into effective action. According to the action-based model, attitude change produced by dissonance is the result of following through with the commitment to the behavior. The attitude change is posited to be one of a number of processes that would assist with the translation of the commitment into effective and unconicted action. Thus, according to the action-based model, dissonance evokes a negative affective state that signals the organism that something is wrong and motivates the organism to engage in behavior to correct the problem. The correction of the problem often involves following through with the commitment to the behavior or decision. This view of dissonance is consistent with past as well as present theorizing on the function of dissonance and dissonance reduction (e.g., Beckmann and Kuhl, 1984; Harmon-Jones, 1999, 2000a; Jones and Gerard, 1967; McGregor et al., 1999; NewbyClark et al., 2002). To assist in translating the intention into effective action, approach motivational processes should be activated, as the individual works to successfully implement the new commitment. Thus, the increase in approach motivation should activate the left frontal cortex. Interestingly, past research ndings are consistent with the idea that the left frontal cortical region may be involved in approach motivational processes aimed at resolving inconsistency. For example, event-related functional magnetic resonance imaging research has found that the left dorsolateral prefrontal cortex is more active during preparation for color naming than during preparation for word naming in a Stroop task (MacDonald et al., 2000). Moreover, more activity in this brain region was associated with less conict (i.e., smaller reaction time interference effects). MacDonald et al. suggested that these ndings support the hypothesis that the left dorsolateral prefrontal cortex is involved in the implementation of control, by representing and actively maintaining the attentional demands of the task (p. 1837). They also suggested that greater activity in the left dorsolateral prefrontal cortex, which implements control, should cause less conict. Other research has suggested that activity in the anterior cingulate cortex is involved in monitoring the occurrence of errors or the presence of response conict (e.g., Carter et al., 1998; Gerhing et al., 1993). Importantly, recent research has found increased anterior cingulate cortex activity, as measured by the event-related potential known as the error-related negativity, when behavior conicts with the self-concept (Amodio et al., 2004). This nding suggests that even higher level conicts, the type with which dissonance theory has been most concerned, also activate the anterior cingulate cortex. Based on this past research, it seems plausible that dissonance (or potential response conict, in action-based model terms) activates the anterior cingulate cortex, and then activates left dorsolateral prefrontal cortex, which assists in resolving the conict. To test the prediction that dissonance was associated with increased left frontal cortical activity, uni-
67
versity students who were opposed to a tuition increase participated in a study ostensibly concerned with attitudes and personality (Harmon-Jones et al., 2000c). They were randomly assigned to one of two choice conditions. In the low-choice condition, participants were told they were to write an essay supporting a 10% tuition increase at their university. In the high-choice condition, participants were told that writing the essay in favor of the tuition increase was their choice and completely voluntary. However, the instructions subtly encouraged them to write such an essay. EEG was assessed for 1 min following the beginning of the writing of the counterattitudinal essay, as past research has revealed that dissonance is greatest at this point in time (Beauvois and Joule, 1996). Moreover, the commitment alone (and not the complete essay writing) is sufcient to evoke dissonance (e.g., Beauvois and Joule, 1996; Rabbie et al., 1959). Then, participants completed an attitude measure. Replicating much past research, results revealed that high-choice participants changed their attitudes more than low-choice participants. Supporting the primary prediction, results also revealed that high-choice participants evidenced greater relative left frontal activity than low-choice participants (main effect of choice manipulation in an ANOVA with lateral frontal and mid-frontal asymmetry as repeated measures). It is important to note that high- and low-choice conditions did not differ in compliance rates; that is, an equal number of high- and low-choice individuals engaged in the counterattitudinal behavior (the illusion of choice is the critical variable in this paradigm). Moreover, they did not differ in baseline resting frontal asymmetry. These results suggest that a selection bias did not produce the above asymmetry results. In addition, the two conditions did not differ in how convincing the essays were or in essay length, suggesting that the differences observed on attitude and relative left frontal activity were not due to differences in the strength or length of the counterattitudinal statements. We have replicated this nding using a different attitudinal issue and in this experiment, we measured EEG immediately after commitment and before essay writing began (Harmon-Jones et al., 2000c, Experiment 2). The results of this recent research suggest that dissonance, a negative emotive state, is associated with greater relative left frontal activity. While dissonance and left frontal activity co-occur, it is not yet empirically clear how they relate to each other. Based on other research, I would suggest that the initial experience of dissonance may activate the anterior cingulate cortex, which then activates the left frontal cortex to engage the approach motivational system, which ultimately may assist with the reduction of dissonance. Indeed, anger may relate to left frontal cortical activity in much the same way; that is, the left frontal cortex may become active during anger to engage the approach motivational system to ultimately assist with the reduction of anger. The ndings of Harmon-Jones et al. (2003b) are consistent with such a view. Only when individuals expected to be able to rectify the anger-evoking situation did increased left frontal activity result. And the left frontal activity related directly to self-reported anger and actions taken to resolve the anger-evoking event. However, when individuals did not expect to be able to rectify the anger-evoking situation, increased left frontal activity did not occur. Interestingly, both experimental conditions caused an increase in self-reported anger. Thus, while left frontal activity may relate to the negative emotions of dissonance and anger, its relation is not due to the valence of the emotions but is instead due to the approach motivational tendencies activated with these emotions.
68
10. Discussion of some issues The reviewed research provides strong support for the motivational direction model of asymmetrical frontal activity and directly contradicts the other two models that have been offered to explain the relationship between emotions and the frontal asymmetry. If asymmetrical frontal activity reected emotional valence, then anger and dissonance would have directly related to relative right frontal activity. That the results of the reviewed studies were signicantly opposite to the prediction derived from emotional valence models provides particularly strong support for the motivational direction model. Before concluding, it is important to discuss a few issues concerning the interpretations of asymmetrical frontal brain activity and its relation to emotion and motivation, and to discuss a few broad implications these ndings have for research on and theories of emotions. 10.1. On the relationship of angry feelings and relative left frontal activity The research examining the relationship between asymmetrical frontal brain activity and anger began with the assumption that anger is often associated with approach motivation. Indeed, as indicated earlier, much past research has revealed that anger is associated with approach motivation and behavior. However, it is important to note that angry feelings are not inevitably associated with approach motivation and left frontal activity. For instance, in the study in which we examined the relationship between coping potential and left frontal activity (Harmon-Jones et al., 2003b), we found that regardless of whether individuals expected to be able to potentially resolve the anger-producing event, equally intense feelings of anger were produced by the aversive event. However, relative left frontal activity was only increased when individuals expected to be able to potentially rectify the anger-producing event. In addition, feelings of anger were associated with left frontal activity only in this latter experimental condition. Thus, feelings of anger are often, but not inevitably, associated with approach motivation and left frontal activity. Results from another study are consistent with this interpretation. In this study, results revealed a dissociation between angry feelings and left frontal activity when high levels of sympathy were rst aroused for the insulting person (Harmon-Jones et al., 2004). In this case, angry feelings were equally high whether or not sympathy was aroused for the insulting person. In contrast, left frontal activity differed between these two conditions, such that left frontal activity was increased following an insult by a person for whom sympathy had not previously been aroused, whereas left frontal activity was not increased when participants rst empathized with the insulting person. In fact, in this latter condition, feelings of anger were directly associated with right frontal activity, suggesting that that the anger experienced while sympathizing with the other person was associated with withdrawal motivation. It is possible that individuals in this condition were motivated to withdraw so that their anger would not lead them to harm the person with whom they had earlier sympathized. In sum, research has revealed that whereas angry feelings are often associated with approach motivation and relative left frontal activity, they are not inevitably associated with these constructs, as has been revealed in recent research (Harmon-Jones et al., 2003b; Harmon-Jones et al., 2004). Future research is needed to assess whether a type of anger that evokes withdrawal and right frontal activity exists.
69
10.2. Dening emotional valence Consistent with current theories of emotion, the perspective advanced in this paper assumes that anger is a negative emotion. However, it is possible that anger could be a positive emotion and thus the valence model of the frontal asymmetry could explain the evidence linking anger to increased left frontal activity. Most perspectives on emotion rarely discuss what is meant by the valence of emotion. Most scientists, like most laypersons, know that joy and enthusiasm are positive emotions, and that anger and fear are negative emotions. On the rare occasions when valence is discussed, emotions are dened as positive or negative (1) because of the cause of the emotion; (2) because of the emotions adaptive consequences; (3) or because of the emotions subjective feel (Lazarus, 1991). Indeed, Lazarus (1991) noted that denition 1whether the personenvironment relationship is benecial or harmfulis the most common, implicit, use of the terms for positive and negative emotion (p. 6). By this denition, anger is indeed negative. Thus, the reviewed research demonstrates that a negative emotion with approach tendencies is associated with increased left frontal activity. However, it is still possible that the frontal asymmetry is a function of emotional valence when emotional valence is dened in an alternative manner. In our research, we sought to test whether the valence model of asymmetrical frontal brain activity could explain the relationship of anger and left frontal activity when emotional valence was dened using the subjective feel denition. In the research, we rst created a trait questionnaire that demonstrated that there were reliable individual differences in persons subjective feelings about anger. Then, we examined the relationship between the subjective feel of anger and resting baseline left frontal activity. We found that while trait anger related to both left frontal activity and a more positive subjective feeling about anger, the subjective feel of anger did not relate to left frontal activity (Harmon-Jones, 2004). Thus, even when using a more arcane denition of emotion, we were unable to nd a relationship between positivity of anger and left frontal activity. Given the difculties inherent in dening adaptational consequences, we have yet to examine relationships between the adaptational consequences of anger and asymmetrical frontal brain activity. However, in some denitions of adaptation, it does appear that anger is associated with increased left frontal activity regardless of the adaptive consequences. That is, regardless of whether anger is associated with constructive action (e.g., working to prevent an injustice; Harmon-Jones et al., 2003b) or destructive action (e.g., behavioral aggression; Harmon-Jones and Sigelman, 2001), anger has been found to relate to increased left frontal activity. In sum, the evidence strongly suggests that anger is a negatively valenced emotion that is related to relative left frontal activity because of its association with approach motivation. 10.3. Evoking emotions Of the emotions examined in the research on brain mechanisms involved in emotions, anger has received less attention than emotions such as fear, sadness, and disgust, which can be induced using established stimuli (e.g., pictures, lms). Part of this relative neglect of anger may be that fear, sadness, and disgust are relatively easier to evoke in the laboratory because active, involving situations are not necessarily required to elicit the emotions.
70
Because the evocation of anger often requires the use of more active and involving situations, the study of anger may require the assistance of social psychologists who are familiar with creating such high-impact settings (see e.g., Harmon-Jones et al., in press). On a related note, examinations of emotions other than anger should consider implementing more active and involving situations, as the more active and involving situations may elicit vastly different brain responses than the ones obtained in more passive situations. Indeed, our research has revealed that the manipulation of coping potential (i.e., the belief that one can take action to resolve the anger-producing situation) affects the activation of asymmetrical frontal brain activity but not the experience of anger (Harmon-Jones et al., 2003b). That is, the belief that one can take action to resolve the anger-producing situation produced an increase in left frontal activity and self-reported anger, whereas the belief that one could not act to resolve the anger-producing situation produced an increase in self-reported anger but not an increase in left frontal activity. Given that one of the primary functions of emotions is the motivation of behavior (Brehm, 1999; Frijda, 1986), it is all the more important for emotion researchers to consider the behavioral context in which emotions are evoked. 10.4. The health consequences of increased left frontal cortical activity Of course, the frontal cortex is a complex structure, and is involved in several psychological processes. The presently reviewed research, in conjunction with past research, suggests that the frontal cortex is also asymmetrically involved in approach and withdrawal motivation. Although many have viewed left frontal cortical activity as only associated with positive valence and positive outcomes, the reviewed research indicates that left frontal activity is also associated with anger, at both trait and state levels of analysis. Thus, it is most likely that left frontal cortical activity is associated with both positive and negative outcomes. Whether left frontal activity (and approach motivation) is associated with positive or negative outcomes likely depends on a variety of factors (e.g., environment) and perspectives (e.g., type of measure; individual versus society). That is, while left frontal activity may be associated with heightened immune system functioning, it may also be associated with other negative consequences for the individual and society. Individuals high in approach motivation, who have greater left frontal activity (Harmon-Jones and Allen, 1997; Sutton and Davidson, 1997), may also be grandiose, manic (Harmon-Jones et al., 2002; Meyer et al., 2001), angry (Harmon-Jones, 2003, in press; Harmon-Jones and Allen, 1998), defensive (Kline et al., 1998; Tomarken and Davidson, 1994), and at greater risk for cardiovascular problems (e.g., James et al., 1983). Moreover, some of the characteristics associated with left frontal activity may provide both benets and harms. For example, angry approach-oriented behaviors may constitute highly effective short-term strategies for achieving specic goals, asserting positions of social inuence, and even causing positive subjective feelings of satisfaction and accomplishment. On the other hand, those very same strategies could contribute to a variety of problems, from domestic abuse to international terrorism. Similarly, such putatively left frontal approach-oriented strategies could, over time, harm the body, such as occurs in the relationship between trait anger and cardiovascular disease or John Henryism (a pattern of active coping with adversity by trying harder and harder against obstacles that may be insurmountable).
71
Most of the evidence cited by Davidson (this issue) in raising the possibility that relative left frontal activity produces positive consequences and may be associated with aspects of resilience is correlational and subject to the limitations of correlational evidence (in particular, third variables). Given such correlational designs, it is prudent to assess the potential moderating or mediating role of frontal brain asymmetry (Coan and Allen, this issue). Davidson (this issue) argues in this regard that left frontal activity may not mediate emotional responses but only moderate them. In support, he reviews research that has found inverse relationships between left frontal activity and amygdala activity (as measured by startle eyeblink response and hemodynamic imaging) during fear and disgust, withdrawal-related emotions. In contrast, we have recently found that anger increases left frontal activity and amydala activity (as measured by the startle eyeblink response; Harmon-Jones et al., 2004). This evidence suggests that during an approach-related emotion, left frontal activity is not suppressing amygdala activity but is positively covarying with it. When emotions that naturally confound valence and motivational direction (e.g., fear) are used to test hypotheses, explanations can be muddled. The point of this review is not to suggest that anger is inevitably associated with increased left frontal activity, or that left frontal activity is inevitably associated with negative outcomes. Our research has indicated that anger is not inevitably associated with left frontal activity. It does, however, suggest that approach motivation, which can involve anger and other negative affective states, is associated with left frontal activity. Moreover, it suggests that a valence model explanation of asymmetrical frontal cortical activity is no longer viable.
11. Conclusion For the last few decades, several models of emotion have considered pleasant to unpleasant dimension of emotion an important organizing principleone that assists in understanding trait mood and situational reactions to signicant stimuli at both subjective and physiological levels of analysis. However, recent developments have suggested that this focus on the valence dimension may not adequately capture emotional space. Consideration of motivational direction in the analysis of emotion, particularly as it relates to asymmetrical frontal brain activity, seems especially important. In addition, as psychological science focuses more attention on the empirically neglected positive aspects to psychological life, it is important to keep in mind that approach motivations are not inevitably associated with positive subjective feelings or positive outcomes.
Acknowledgements The research presented in this article was funded by grants from the National Institute of Mental Health (MH60747-01 and MH52662), by a grant from the National Science Foundation (BCS-9910702), and by grants from the Wisconsin/Hilldale Undergraduate/Faculty Research Fund. The comments John Allen (editor), Cindy Harmon-Jones, Steve Sutton, Jim Coan, and an anonymous reviewer provided on this article are much appreciated.
72
References
Abramson, L.Y., Metalsky, G.I., Alloy, L.B., 1989. Hopelessness depression: a theory-based subtype of depression. Psychological Review 96, 358372. Ahern, G.L., Schwartz, G.E., 1985. Differential lateralization for positive and negative emotion in the human brain: EEG spectral analysis. Neuropsychologia 23, 745755. Allen, J.J., Iacono, W.G., Depue, R.A., Arbisi, P., 1993. Regional EEG asymmetries in bipolar seasonal affective disorder before and after phototherapy. Biological Psychiatry 33, 642646. Allen, J.J.B., Harmon-Jones, E., Cavender, J., 2001. Manipulation of frontal EEG asymmetry through biofeedback alters self-reported emotional responses and facial EMG. Psychophysiology 38, 685693. Amodio, D.M., Harmon-Jones, E., Devine, P.G., Curtin, J.J., Hartley, S., Covert, A., 2004. Neural signals for the control of unintentional race bias. Psychological Science 15, 8893. Baehr, E., Rosenfeld, J.P., Baehr, R., 1997. The clinical use of an alpha asymmetry protocol in the neurofeedback treatment of depression: two case studies. Journal of Neurotherapy 2, 1023. Baron, R.A., 1977. Effects of victims pain cues, victims race, and level of prior instigation upon physical aggression. Journal of Applied Social Psychology 9, 103114. Batson, C.D., 1991. The Altruism Question: Toward a SocialPsychological Answer. Lawrence Erlbaum, Hillsdale, NJ. Beauvois, J. L., Joule, R. V., 1996. A radical dissonance theory. London: Taylor and Francis. Beckmann, J., Kuhl, J., 1984. Altering information to gain action control: Functional aspects of human information processing in decision making. Journal of Research in Personality 18, 224237. Berkowitz, L., 1999. Anger. In: Dalgleish, T., Power, M.J. (Eds.), Handbook of Cognition and Emotion. John Wiley and Sons, Chichester, England, pp. 411428. Berkowitz, L., Cochran, S., Embree, M., 1981. Physical pain and the goal of aversively stimulated aggression. Journal of Personality and Social Psychology 40, 687700. Berkowitz, L., 2000. Causes and consequences of feelings. Cambridge: Cambridge University Press. Blanchard, D.C., Blanchard, R.J., 1984. Affect and aggression: an animal model applied to human behavior. Advances in the Study of Aggression 1, 162. Blanchard, D.C., Takahashi, S.N., 1988. No change in intermale aggression after amygdala lesions which reduce freezing. Physiology and Behavior 42, 613616. Boissy, A., Bouissou, M.F., 1994. Effects of androgen treatment on behavioral and physiological responses of heifers to fear-eliciting situations. Hormones and Behavior 28, 6683. Brehm, J.W., 1999. The intensity of emotion. Personality and Social Psychology Review 3, 222. Brehm, J. W., Cohen, A. R., 1962. Explorations in cognitive dissonance. New York: Wiley. Brehm, J.W., Self, E., 1989. The intensity of motivation. In: Rosenzweig, M.R., Porter, L.W (Eds.), Annual Review of Psychology, vol. 40. Annual Reviews Inc., Palo Alto, CA, USA, pp. 109131. Busch, D.E., Bareld, R.J., 1974. A failure of amygdaloid lesions to alter agonistic behavior in the laboratory rat. Physiology and Behavior 12, 887892. Buss, A.H., Perry, M., 1992. The aggression questionnaire. Journal of Personality and Social Psychology 63, 452459. Carter, C.S., Braver, T.S., Barch, D.M., Botvinick, M.M., Noll, D., Cohen, J.D., 1998. Anterior cingulate cortex, error detection, and the online monitoring of performance. Science 280, 747749. Carver, C.S., 2001. Affect and the functional bases of behavior: on the dimensional structure of affective experience. Personality and Social Psychology Review 5, 345356. Carver, C.S., White, T.L., 1994. Behavioral inhibition, behavioral activation, and affective responses to impending reward and punishment: the BIS/BAS scales. Journal of Personality and Social Psychology 67, 319 333. Cassidy, F., Forest, K., Murry, E., Carroll, B.J., 1998. A factor analysis of the signs and symptoms of mania. Archives of General Psychiatry 55, 2732. Coan, J. A., Allen, J. J. B., this issue. Frontal EEG asymmetry as a moderator and mediator of emotion. Biological Psychology Coan, J.A., Allen, J.J.B., 2003. Frontal EEG asymmetry and the behavioral activation and inhibition systems. Psychophysiology 40, 106114.
73
Coan, J.A., Allen, J.J.B., Harmon-Jones, E., 2001. Voluntary facial expression and hemispheric asymmetry over the frontal cortex. Psychophysiology 38, 912925. Cook, I.A., OHara, R., Uijtdehaage, S.H.J., Mandelkern, M., Leuchter, A.F., 1998. Assessing the accuracy of topographic EEG mapping for determining local brain function. Electroencephalography and Clinical Neurophysiology 107, 408414. DAlfonso, A.A.L., van Honk, J., Hermans, E., Postma, A., de Haan, E.H.F., 2000. Laterality effects in selective attention to threat after repetitive transcranial magnetic stimulation at the prefrontal cortex in female subjects. Neuroscience Letters 280, 195198. Darwin, C., 1872/1965. The Expression of the Emotions in Man and Animals. The University of Chicago Press, Chicago, IL. Davidson, R. J., this issue. What does the prefrontal cortex do in affect? Perspectives on frontal EEG asymmetry research. Biological Psychology. Davidson, R. J., 1995. Cerebral asymmetry, emotion, and affective style. In: R. J. Davidson, K. Hugdahl (Eds.), Brain Asymmetry, Cambridge, MA: Massachusetts Institute of Technology, pp. 361387. Davidson, R.J., 1998. Anterior electrophysiological asymmetries, emotion, and depression: conceptual and methodological conundrums. Psychophysiology 35, 607614. Davidson, R.J., Fox, N.A., 1982. Asymmetrical brain activity discriminates between positive and negative affective stimuli in human infants. Science 218, 12351236. Davidson, R.J., Fox, N.A., 1989. Frontal brain asymmetry predicts infants response to maternal separation. Journal of Abnormal Psychology 98, 127131. Davidson, R.J., Chapman, J.P., Chapman, L.J., Henriques, J.B., 1990. Asymmetrical brain electrical activity discriminates between psychometrically-matched verbal and spatial cognitive tasks. Psychophysiology 27, 528543. Dawson, G., Panagiotides, H., Klinger, L.G., Hill, D., 1992. The role of frontal lobe functioning in the development of infant self-regulatory behavior. Brain and Cognition 20, 152175. Depue, R.A., Iacono, W.G., 1989. Neurobehavioral aspects of affective disorders. Annual Review of Psychology 40, 457492. Depue, R.A., Krauss, S., Spoont, M.R., 1987. A two-dimensional threshold model of seasonal bipolar affective disorder. In: Magnusson, D., hman, A. (Eds.), Psychopathology: An Interactional Perspective. Academic Press, New York, pp. 95123. Ekman, P., Davidson, R.J., 1993. Voluntary smiling changes regional brain activity. Psychological Science 4, 342345. Ekman, P., Friesen, W.V., 1975. Unmasking the face: A Guide to Recognizing Emotions from Facial Clues. Prentice-Hall, Englewood Cliffs, NJ. Elliot, A.J., Devine, P.G., 1994. On the motivation nature of cognitive dissonance: Dissonance as psychological discomfort. Journal of Personality and Social Psychology 67, 382394. Flynn, J., Vanegas, H., Foote, W., Edwards, S.B., 1970. Neural mechanisms involved in a cats attack on a rat. In: Whalen, R., Thompson, R.F., Verzeano, M., Weinberger, N. (Eds.), The Neural Control of Behavior. Academic Press, New York, pp. 135173. Fowles, D.C., 1988. Psychophysiology and psychopathology: a motivational approach. Psychophysiology 25, 373391. Fowles, D.C., 1993. Behavioral variables in psychopathology: a psychobiological perspective. In: Sutker, P.B., Adams, H.E. (Eds.), Comprehensive Handbook of Psychopathology, second ed. Plenum Press, New York, pp. 5782. Fox, N.A., 1991. If its not left, its right. American Psychologist 46, 863872. Fox, N.A., Davidson, R.J., 1988. Patterns of brain electrical activity during facial signs of emotion in 10-month-old infants. Developmental Psychology 24, 230236. Fox, N.A., Henderson, H.A., Rubins, K.H., Calkins, S.D., Schmidt, L.A., 2001. Continuity and discontinuity of behavioral inhibition and exuberance: psychophysiological and behavioral inuences across the rst four years of life. Child Development 72, 121. Frijda, N.H., 1986. The Emotions. Cambridge University Press, Cambridge, England. Gerhing, W.J., Goss, B., Coles, M.G.H., Meyer, D.E., Donchin, E., 1993. A neural system for error detection and compensation. Psychological Science 4, 385390.
74
Gotlib, I.H., Ranganath, C., Rosenfeld, J.P., 1998. Frontal EEG alpha asymmetry, depression, and cognitive functioning. Cognition and Emotion 12, 449478. Gray, J.A., 1982. The Neuropsychology of Anxiety: An Enquiry into the Functions of the Septo-hippocampal System. Oxford University Press, New York. Hagemann, D., Naumann, E., Thayer, J., Bartussek, D., 2002. Does resting electroencephalograph asymmetry reect a trait? An application of latent state-trait theory. Journal of Personality and Social Psychology 82, 619641. Harmon-Jones, E., 1999. Toward an understanding of the motivation underlying dissonance processes: is feeling personally responsible for the production of aversive consequences necessary to cause dissonance effects? In: Harmon-Jones, E., Mills, J., Cognitive Dissonance: Perspectives on a Pivotal Theory in Social Psychology. American Psychological Association, Washington, DC, pp. 7199. Harmon-Jones, E., 2003. Anger and the behavioural approach system. Personality and Individual Differences 35, 9951005. Harmon-Jones, E., 2000a. Cognitive dissonance and experienced negative affect: Evidence that dissonance increases experienced negative affect even in the absence of aversive consequences. Personality and Social Psychology Bulletin 26, 14901501. Harmon-Jones, E., 2000b. A cognitive dissonance theory perspective on the role of emotion in the maintenance and change of beliefs and attitudes. In: N. H. Frijda, A. R. S., Manstead, S. Bem (Eds.), Emotions and beliefs, Cambridge: Cambridge University Press, pp. 185211. Harmon-Jones, E., Fearn, M., Gerdjikov, T., Harmon-Jones, C., 2000c. Dissonance and left frontal cortical activity: A test of the action-based model of cognitive dissonance. Unpublished manuscript. Harmon-Jones, E., 2004. On the relationship of anterior brain activity and anger: examining the role of attitude toward anger. Cognition and Emotion 18, 337361. Harmon-Jones, E., Allen, J.J.B., 1997. Behavioral activation sensitivity and resting frontal EEG asymmetry: covariation of putative indicators related to risk for mood disorders. Journal of Abnormal Psychology 106, 159163. Harmon-Jones, E., Allen, J.J.B., 1998. Anger and frontal brain activity: EEG asymmetry consistent with approach motivation despite negative affective valence. Journal of Personality and Social Psychology 74, 13101316. Harmon-Jones, E., Allen, J.J.B., 2001. The role of affect in the mere exposure effect: evidence from psychophysiological and individual differences approaches. Personality and Social Psychology Bulletin 27, 889898. Harmon-Jones, E., Harmon-Jones, C., 2002. Testing the action-based model of cognitive dissonance: The effect of action-orientation on post-decisional attitudes. Personality and Social Psychology Bulletin 28, 711723. Harmon-Jones, E., Sigelman, J., 2001. State anger and frontal brain activity: evidence that insult-related relative left prefrontal activation is associated with experienced anger and aggression. Journal of Personality and Social Psychology 80, 797803. Harmon-Jones, E., Abramson, L.Y., Sigelman, J., Bohlig, A., Hogan, M.E., Harmon-Jones, C., 2002. Proneness to hypomania/mania or depression and asymmetrical frontal cortical responses to an anger-evoking event. Journal of Personality and Social Psychology 82, 610618. Harmon-Jones, E., Amodio, D.M., Zinner, L., in press. Social psychological methods of emotion induction. In: Coan, J.A., Allen, J.J.B. (Eds.), Handbook of Emotion Elicitation and Assessment. Oxford Unversity Press. Harmon-Jones, E., Peterson, H., Vaughn, K., 2003a. The dissonance-inducing effects of an inconsistency between experienced empathy and knowledge of past failures to help: support for the action-based model of dissonance. Basic and Applied Social Psychology 25, 6978. Harmon-Jones, E., Sigelman, J.D., Bohlig, A., Harmon-Jones, C., 2003b. Anger, coping, and frontal cortical activity: the effect of coping potential on anger-induced left frontal activity. Cognition and Emotion 17, 124. Harmon-Jones, E., Vaughn, K., Mohr, S., Sigelman, J., Harmon-Jones, C., in press. The effect of manipulated sympathy and anger on left and right frontal cortical activity. Emotion. Harmon-Jones, E., Fearn, M., Burck, A., Welken, E., Moody, M., 2004. Asymmetrical frontal cortical and startle eyeblink responses to anger-evoking pictures, unpublished raw data. Heller, W., 1990. The neuropsychology of emotion: developmental patterns and implications for psychopathology. In: Stein, N.L., Leventhal, B., Trabasso, T. (Eds.), Psychological and Biological Approaches to Emotion. Lawrence Erlbaum Associates, Hillsdale, NJ, pp. 167211. Heller, W., Nitschke, J.B., 1998. The puzzle of regional brain activity in depression and anxiety: the importance of subtypes and comorbidity. Cognition and Emotion 12, 421447.
75
Henriques, J.B., Davidson, R.J., 1990. Regional brain electrical asymmetries discriminate between previously depressed and healthy control subjects. Journal of Abnormal Psychology 99, 2231. Izard, C.E., 1991. The Psychology of Emotions. Plenum Press, New York. Jacobs, G.D., Snyder, D., 1996. Frontal brain asymmetry predicts affective style in men. Behavioral Neuroscience 110, 36. James, S., Hartnett, S., Kalsbeek, W., 1983. John Henryism and blood pressure differences among black men. Journal of Behavioral Medicine 6, 259278. Jones, E. E., Gerard, H. B., 1967. Foundations of social psychology. New York: Wiley. Kline, J.P., Allen, J.J.B., Schwartz, G.E., 1998. Is left frontal brain activation in defensiveness gender specic? Journal of Abnormal Psychology 107, 149153. Lagerspetz, K.M.J., 1969. Aggression and aggressiveness in laboratory mice. In: Garattini, S., Sigg, E.B. (Eds.), Aggressive Behavior. Wiley, New York, pp. 7785. Laurent, J., Potter, K., Catanzaro, S.J., 1994. Assessing positive and negative affect in children: the development of the PANAS-C. In: Paper presented at the 26th Annual Convention of the National Association of School Psychologists. Seattle, WA. Lazarus, R.S., 1991. Emotion and Adaptation. Oxford, New York. Lerner, J.S., Keltner, D., 2001. Fear, anger, and risk. Journal of Personality and Social Psychology 81, 146159. Levine, L.J., 1995. Young childrens understanding of the causes of anger and sadness. Child Development 66, 697709. Lewis, M., Sullivan, M.W., Ramsey, D.S., Alessandri, S.M., 1992. Individual differences in anger and sad expressions during extinction: antecedents and consequences. Infant Behavior and Development 15, 443452. Lieberman, J.D., Solomon, S., Greenberg, J., McGregor, H.A., 1999. A hot new way to measure aggression: hot sauce allocation. Aggressive Behavior 25, 331348. Malone, R.P., Delaney, M.A., Luebbert, J.F., Cater, J., Campbell, M., 2000. A double-blind placebo-controlled study of lithium in hospitalized aggressive children and adolescents with conduct disorder. Archives of General Psychiatry 57, 649654. McGregor, H.A., Lieberman, J.D., Greenberg, J., Solomon, S., Arndt, J., Simon, L., Pyszczynski, T., 1998. Terror management and aggression: evidence that mortality salience motivates aggression against worldview threatening others. Journal of Personality and Social Psychology 74, 590605. McGregor, I., Newby-Clark, I. R., Zanna, M. P., 1999. Epistemic discomfort is moderated by simultaneous accessibility of inconsistent elements. In: E. Harmon-Jones, J. Mills (Eds.), Cognitive dissonance: Progress on a pivotal theory in social psychology, Washington, DC: American Psychological Association, pp. 325353. MacDonald, W.III., Cohen, J.D., Stenger, V.A., Carter, C.S., 2000. Dissociating the role of the dorsolateral prefrontal and anterior cingulate cortex in cognitive control. Science 288, 18351838. Meyer, B., Johnson, S.L., Carver, C.S., 1999. Exploring behavioral activation and inhibition sensitivities among college students at risk for bipolar spectrum symptomatology. Journal of Psychopathology and Behavioral Assessment 21, 275292. Meyer, B., Johnson, S.L., Winters, R., 2001. Responsiveness to threat and incentive in bipolar disorder: relations of the BIS/BAS scales with symptoms. Journal of Psychopathology and Behavioral Assessment 23, 133143. Mikulincer, M., 1988. Reactance and helplessness following exposure to unsolvable problems: the effects of attributional style. Journal of Personality and Social Psychology 54, 679686. Miller, P.A., Eisenberg, N., 1988. The relation of empathy to aggressive and externalizing/antisocial behavior. Psychological Bulletin 103, 324344. Moyer, K.E., 1976. The Psychobiology of Aggression. Harper and Row, New York. Newby-Clark, I. R., McGregor, I., Zanna, M. P., 2002. Thinking and caring about cognitive inconsistency: When and for whom does attitudinal ambivalence feel uncomfortable? Journal of Personality and Social Psychology, 82, 157166. Oatley, K., Jenkins, J.M., 1996. Understanding Emotions. Blackwell Publishers Inc., Oxford, England, UK. Olweus, D., 1986. Aggression and hormones: Behavioral relationship with testosterone and adrenaline. In: D. Olweus, J. Block, M. Radke-Yarrow (Eds.), Development of antisocial and prosocial behavior: Research, theories, and issues, Orlando: Academic Press, pp. 5172. Plutchik, R., 1980. Emotion: A Psychoevolutionary Synthesis. Harper and Row, New York. Rabbie, J.M., Brehm, J.W., Cohen, A.R., 1959. Verbalization and reactions to cognitive dissonance. Journal of Personality 27, 407417.
76
Robinson, R.G., Downhill, J.E., 1995. Lateralization of psychopathology in response to focal brain injury. In: Davidson, R.J., Hugdahl, K. (Eds.), Brain Asymmetry. Massachusetts Institute of Technology, Cambridge, MA, pp. 693711. Roseman, I.J., 1991. Appraisal determinants of discrete emotions. Cognition and Emotion 5, 161200. Rosenfeld, J.P., Cha, G., Blair, T., Gotlib, I.H., 1995. Operant (biofeedback) control of left-right frontal alpha power differences: potential neurotherapy for affective disorders. Biofeedback and Self-Regulation 20, 241258. Schaffer, C.E., Davidson, R.J., Saron, C., 1983. Frontal and parietal electroencephalogram asymmetry in depressed and nondepressed subjects. Biological Psychiatry 18, 753762. Scherer, K.R., 1993. Studying the emotion-antecedent appraisal process: an expert system approach. Cognition and Emotion 7, 325355. Silberman, E.K., Weingartner, H., 1986. Hemispheric lateralization of functions related to emotion. Brain and Cognition 5, 322353. Sutton, S.K., Davidson, R.J., 1997. Prefrontal brain asymmetry: a biological substrate of the behavioral approach and inhibition systems. Psychological Science 8, 204210. Tomarken, A.J., Davidson, R.J., 1994. Frontal brain activation in repressors and nonrepressors. Journal of Abnormal Psychology 103, 339349. Tomarken, A.J., Keener, A.D., 1998. Frontal brain asymmetry and depression: a self-regulatory perspective. Cognition and Emotion 12, 387420. Tomarken, A.J., Davidson, R.J., Henriques, J.B., 1990. Resting frontal brain asymmetry predicts affective responses to lms. Journal of Personality and Social Psychology 59, 791801. Tomarken, A.J., Davidson, R.J., Wheeler, R.E., Doss, R., 1992a. Individual differences in anterior brain asymmetry and fundamental dimensions of emotion. Journal of Personality and Social Psychology 62, 676687. Tyrer, S., Shopsin, B., 1982. Symptoms and assessment of mania. In: Paykel, E.S. (Ed.), Handbook of Affective Disorders. Guilford Press, New York, pp. 1223. Vandenheede, M., Bouissou, M.F., 1993. Effect of androgen treatment on fear reactions in ewes. Hormones and Behavior 27, 435448. van Honk, J., Tuiten, A., van den Hout, M., Koppeschaar, H., Thijssen, J., de Haan, E., Verbaten, R., 1998. Baseline salivary cortisol levels and preconscious selective attention for threat: A pilot study. Psychoneuroendocrinology 23, 741747. van Honk, J., Tuiten, A., Verbaten, R., van den Hout, M., Koppeschaar, H., Thijssen, J., de Haan, E., 1999. Correlations among salivary testosterone, mood, and selective attention to threat in humans. Hormones and Behavior 36, 1724. van Honk, J., Tuiten, A., de Haan, E., van den Hout, M., Stam, H., 2001. Attentional biases for angry faces: Relationships to trait anger and anxiety. Cognition and Emotion 15, 279297. Watson, D., 2000. Mood and Temperament. Guilford Press, New York. Watson, D., Clark, L.A., Tellegen, A., 1988. Development and validation of brief measures of Positive and Negative Affect: the PANAS scales. Journal of Personality and Social Psychology 54, 10631070. Watson, D., Wiese, D., Vaidya, J., Tellegen, A., 1999. The two general activation systems of affect: structural ndings, evolutionary considerations, and psychobiological evidence. Journal of Personality and Social Psychology 76, 820838. Wheeler, Davidson, R.J., Tomarken, A.J., 1993. Frontal brain asymmetry and emotional reactivity: a biological substrate of affective style. Psychophysiology 30, 8289. Wortman, C.B., Brehm, J.W., 1975. Responses to uncontrollable outcomes: an integration of reactance theory and the learned helplessness model. In: Berkowitz, L. (Ed.), Advances in Experimental Social Psychology, vol. 8. Academic Press, New York, pp. 278336. Wright, R.A., Kirby, L.D., 2001. Effort determination of cardiovascular response: an integrative analysis with applications in social psychology, In: Zanna, M.P. (Ed.), Advances in Experimental Social Psychology, vol. 33, pp. 255307. Young, P.T., 1943. Emotion in Man and Animal: Its Nature and Relation to Attitude and Motive. John Wiley and Sons, New York. Zajonc, R.B., McIntosh, D.N., 1992. Emotions research: some promising questions and some questionable promises. Psychological Science 3, 7074. Zanna, M.P., Cooper, J., 1974. Dissonance and the pill: An attribution approach to studying the arousal properties of dissonance. Journal of Personality and Social Psychology 29, 703709.
Resting frontal brain activity: linkages to maternal depression and socio-economic status among adolescents
Andrew J. Tomarken a, , Gabriel S. Dichter a , Judy Garber b , Christopher Simien a
a
Department of Psychology, College of Arts and Sciences, Vanderbilt University, 301 Wilson Hall, Nas hville, TN 37203, USA b Department of Psychology and Human Development, Peabody College, Vanderbilt University, Nashville, TN 37203, USA
Abstract We tested the prediction that resting frontal brain asymmetry would be a marker of vulnerability for depression among adolescents. Baseline electroencephalographic (EEG) activity was recorded from 12 to14-year-old adolescents whose mothers had a history of depression (high risk group) and whose mothers were lifetime-free of axis I psychopathology (low risk group). High risk adolescents demonstrated the hypothesized pattern of relative left frontal hypo-activity on alpha-band measures. Such effects were specic to the mid-frontal region and generally consistent across reference montages. Socio-economic status (SES) also predicted alpha asymmetry. When the effects of SES and risk status were jointly assessed, SES contributed unique variance to the prediction of frontal brain asymmetry. The implications of the observed relations among maternal depression, SES, and frontal brain asymmetry are discussed. 2004 Elsevier B.V. All rights reserved.
Keywords: Frontal EEG asymmetry; Emotion; Risk for depression; Socio-economic status
1. Frontal brain asymmetry and depression In this study, we assessed whether adolescents who are at-risk for depression differ in patterns of resting frontal brain activity when compared to low risk adolescents. We
Corresponding author. Tel.: +1-615-322-4177; fax: +1-615-343-8449. E-mail address: andrew.j.tomarken@vanderbilt.edu (A.J. Tomarken).
78
A.J. Tomarken et al. / Biological Psychology 67 (2004) 77102
addressed this question because of evidence from a variety of sources indicating a linkage between unipolar depression and decreased activity of left relative to right hemisphere frontal brain regions (for reviews, see Davidson, 1995, 1998b; Davidson et al., 2002; Tomarken and Keener, 1998). Consistent with this conclusion are: (a) neurological studies indicating that the severity of depressive symptomatology is correlated with the proximity of a left-hemisphere lesion to the frontal pole (e.g., Morris et al., 1996; Pohjasvaara et al., 2002; Robinson and Downhill, 1995; Shimoda and Robinson, 1999); (b) regional cerebral blood ow (rCBF) studies showing that clinically depressed participants demonstrate relative decreases in left frontal activity when compared to non-depressed control participants (e.g., Baxter et al., 1989; Bench et al., 1992; Ebert et al., 1991; Martinot et al., 1990); (c) resting electroencephalographic (EEG) studies showing that clinically depressed individuals (Allen et al., 1993) or individuals characterized by elevated scores on the Beck Depression Inventory (Beck and Steer, 1987) demonstrate relative left frontal hypo-activity when compared to controls (Schaffer et al., 1983); and, (d) studies showing that transcranial magnetic stimulation of the left frontal cortex (which may well increase left frontal activity) produces clinical improvement in depressed individuals (e.g., George et al., 1997; Pascual-Leone et al., 1996). As a cautionary note, we should add that not all ndings have been consistent in these areas (e.g., Dam et al., 1989; House et al., 1990; MacHale et al., 1998). In addition to these empirical linkages are theoretical perspectives proposing that left frontal dysfunction may be a neural substrate of core features of unipolar depression. For example, consistent with the approach-withdrawal model of frontal asymmetry (Davidson and Tomarken, 1989; Fox, 1991), several commentators have proposed that anhedonia reects a decit in a neural approach system, one component of which is the left frontal cortex (e.g., Davidson, 1998a).
2. Left frontal hypo-activity and vulnerability to depression: children of depressed mothers Of prime importance in the present context are EEG studies indicating that resting frontal brain asymmetry may indicate heightened vulnerability to depression. For example, two studies have found that currently euthymic individuals who have a history of depression demonstrate left frontal hypo-activity relative to control participants (Allen et al., 1993; Henriques and Davidson, 1990). Unfortunately, studies of individuals with remitted depression cannot distinguish whether left frontal hypo-activity is a vulnerability factor for depression or a consequence of depression (Alloy et al., 1999). An assessment of at-risk populations who have not yet manifested depression represents a more direct test of whether left frontal hypo-activity indicates vulnerability. One such population is children of depressed parents. Such children exhibit a range of negative outcomes and psychiatric diagnoses compared to children of parents without a psychiatric history (Downey and Coyne, 1990; Gelfand and Teti, 1990) and appear to be at particularly heightened risk for developing depression (Hammen, 1991; Warner et al., 1992; Weissman et al., 1992, 1997). Several studies have found that infants of depressed mothers do in fact exhibit left frontal hypo-activity (e.g., Dawson et al., 1997; see also Field et al., 1995; Jones et al., 1997).
79
Dawson and her colleagues also have linked left frontal hypo-activity in infants of depressed mothers to decreased positive affect and increased negative affect during interactions (Dawson et al., 1992a, 1999a,b). To date, studies of frontal brain asymmetry in children of depressed mothers have concentrated on infants. One goal of the present study was to investigate whether adolescent offspring of mothers with a history of depression demonstrated the same pattern of relative left frontal hypo-activity that has been observed in previous studies with currently depressed adults, adults with a history of depression, and high risk infants. To test this hypothesis, we compared patterns of resting frontal EEG asymmetry in young adolescent (1214-year-old) offspring of depressed and non-depressed mothers. There were several reasons why we assessed children in this age range. First, the presence of left frontal hypo-activity in adolescent offspring of depressed mothers would indicate a continuity of risk beginning in infancy and extending into adolescence. Second, the peak age of onset of depression in children of depressed parents ranges from 14 to 20 years (Weissman et al., 1997). Thus, left frontal hypo-activity in adolescent offspring of depressed mothers would indicate the presence of a potential indicator of risk at a point in time that immediately antedates the dramatic increase in the incidence of depression. Third, although the prevalence rate of depression in preadolescence is comparable in boys and girls, by adulthood nearly twice as many women are diagnosed with depression as men (Nolen-Hoeksema and Girgus, 1994). Thus, both increases in depressive symptoms and sex differences in such symptoms rst emerge during adolescence. Although prior studies have not typically found sex differences in the linkages between frontal brain asymmetry and depression, this is a largely unexplored area. We sought to examine the interrelations among sex, differential risk for depression, and frontal brain asymmetry in adolescents at high and low risk for depression.
3. Relations with socio-economic status We also assessed the relation between socio-economic status (SES) and frontal brain asymmetry. A number of epidemiological studies have indicated an inverse relation between social class and rates of unipolar depression (Kaplan et al., 1987; Leventhal and Brooks-Gunn, 2000; Murphy et al., 1991; Pearlin and Johnson, 1977; Weissman and Myers, 1978), as well as other psychiatric illnesses (Dohrenwend, 2000; Goodman, 1999; Hollingshead and Redlich, 1958; Rushing and Ortega, 1979; Srole, 1962; Williams, 1990). However, not all ndings have been consistent (e.g., Weissman et al., 1991; Weissman and Myers, 1978), and, in other cases, there is debate concerning the causal direction of the linkage between social class and unipolar depression (e.g., Fox, 1990; Rodgers and Mann, 1993). Despite these caveats, the available evidence suggests that social class may often be implicated in the etiology of depression. If lowered socio-economic status is a risk factor, it may moderate relations that have been observed between other risk variables (e.g., maternal history) and frontal brain asymmetry or between frontal brain asymmetry and depression. Alternatively, from a mediational perspective, SES could be a distal causal factor, the effects of which are mediated by other more proximal variables (e.g., maternal history). Unfortunately, prior ndings linking
80
frontal brain asymmetry and depression have failed to adequately examine the potential role played by SES. First, many investigators have not reported the socio-demographic composition of their samples (e.g., Allen et al., 1993; Dam et al., 1989; Dawson et al., 1992a). Second, among studies that have reported socio-demographic information, most have not examined linkages between SES and both the occurrence of mood disorders and patterns of frontal brain asymmetry (e.g., Henriques and Davidson, 1990; Morris et al., 1996; Robinson et al., 1984; Robinson and Price, 1982; Robinson and Szetela, 1981; Starkstein et al., 1987). Third, some samples have been characterized by restricted range on SES measures. For example, in some EEG studies the clear majority of the participants were economically disadvantaged mothers and their infants (e.g., Dawson et al., 1992a,b; Field et al., 1995; Jones et al., 1997), whereas in other studies college students, a relatively homogenous group, have served as participants (Davidson et al., 1985; Schaffer et al., 1983). To our knowledge, only one frontal EEG study has examined the correlation between frontal brain asymmetry and SES (Henriques and Davidson, 1991). Although this study failed to nd a signicant relation, a more systematic investigation of this question is necessary. The current study used a sample recruited from a metropolitan community, and thus participants varied widely with respect to SES. We sought to examine the main and interactive effects of SES and risk status on frontal brain asymmetry.
4. Effects of reference montage We assessed the relations among risk status, SES, and frontal asymmetry using three different EEG reference montages. There is evidence that the linkages between EEG asymmetry measures and measures of emotion or psychopathology are not always consistent across difference referencing schemes (for reviews, see Davidson, 1998b; Hagemann et al., 1998; Reid et al., 1998). Illustratively, Reid et al. (1998) assessed frontal brain asymmetry in depressed and non-depressed adults and found signicant group differences only during the rst 2 min of EEG recording using a linked-mastoid reference. There were no signicant between-group differences using average and Cz reference montages. Published ndings examining the relation between frontal brain asymmetry and risk for depression in infants have typically examined Cz-referenced data (e.g., Dawson et al., 1992a,b; Field et al., 1995), although others (e.g., Dawson et al., 1999b) have employed a linked-mastoid reference strategy. In the present study, to attend to the consistency issue, we included three reference montages (Cz, computer-averaged-ears, and average). In sum, we addressed whether high and low risk adolescents differ in patterns of resting frontal brain asymmetry. In addition, we assessed the roles of sex and SES in predicting frontal brain asymmetry and whether these two factors moderated the effect of risk status on frontal brain asymmetry. We predicted that adolescent offspring of mothers with a history of depression would demonstrate relative left frontal hypo-activity compared to adolescent offspring of non-depressed mothers. We did not have strong a priori hypotheses concerning the link between frontal brain asymmetry and SES or sex. However, we expected that, if relations were found, left frontal hypo-activity would be linked to female sex and to lower SES. To test these hypotheses, we used multiple EEG reference montages.
81
5. Method 5.1. Participants Participants were recruited from a larger sample of 240 adolescents and their mothers who were already participating in a study investigating the development of depression in adolescents. This larger sample was 54.2% female, 82% Caucasian, 14.7% AfricanAmerican, and 3.3% other (Hispanic, Asian, Native American). The sample was pre-dominantly lower-middle class to middle class with a mean SES (Hollingshead, 1975) of 41.84 (S.D. = 13.25). Participants for the larger study were recruited by letters sent to parents of children in the fth grade in the Nashville metropolitan public schools. Parents were invited to participate and were asked to complete a brief health history questionnaire indicating whether they ever experienced any of 24 medical conditions such as diabetes, cancer, heart disease, and depression, or if they had ever taken any of 34 medications. Of the 1495 parents, who returned these questionnaires, telephone screening interviews were conducted with the 587 who had endorsed either a history of depression, use of antidepressants, or no history of psychopathology. Based on these screening calls, 349 mothers who reported a history of depression or no history of psychiatric problems were interviewed in person with the structured clinical interview for DSM (SCID, Spitzer et al., 1990). To assess inter-rater reliability of the SCID, a subset (20%) of audio-taped interviews was evaluated by a second interviewer blind to the ratings of the original interviewer. The kappa coefcient (Cohen, 1960) indicating chancecorrected agreement was 0.88 for SCID diagnoses of mood disorders.. Families were excluded if mothers indicated a history of solely non-affective psychiatric disorders, or if a parent or child had serious medical problems. The nal high risk sample included 185 mothers who indicated a history of depressive disorders (i.e., major depression, dysthymia, depression not otherwise specied, and adjustment disorder with depressed mood). The low risk group consisted of 55 mothers who were lifetime-free of psychiatric diagnoses. Research staff, unaware of the mothers psychiatric history, administered a battery of questionnaires to the parents and their children. Only those assessment instruments relevant to the present study are described here. Families participated in yearly assessments after the initial interviews. Participants were asked if they were interested in participating in future studies, and those indicating such an interest were contacted to participate in the current investigation. Adolescents in the current study were recruited between the rst and second yearly follow-up interviews. Of the 50 high risk and 20 low risk participants from the larger study who were contacted to participate in the current study, 32 high risk and 15 low risk participants agreed to participate in the resting EEG recording. Among those who participated, two high risk participants and one low risk participant were excluded from analyses because they were not right-handed, as assessed by the Edinburgh inventory (Oldeld, 1971). Five high risk participants and one low risk participant were excluded from analyses due to other issues that compromised the validity of the resting EEG recording for the present purposes (e.g., excessive fatigue during the session, medication use). Thus, 25 high risk (11 male) and 13 low risk (7 male) participants were included in the current analyses. Participants were
82
compensated US$ 30. The relatively greater number of high risk than low risk participants in the current study reects the distribution of the larger participant pool from which these participants were drawn. A primary goal of the larger ongoing study was to investigate the conditions under which at-risk youths develop depression. High risk adolescents included 24 Caucasians and one African American. Low risk adolescents included 12 Caucasians and one Native American. The racial distributions of participants in the current study and in the larger sample from which participants were drawn did not statistically differ, 2 (4, N = 283) = 5.08, p > 0.25. Participants in the present study ranged in age from 12.2 to 14.0-year-old at the time of their EEG recording, high risk Mean = 13.1 (S.D. = 0.3); low risk Mean = 13.0 (S.D. = 0.4). The two groups did not signicantly differ with respect to age, F(1,37) = 2.20, P > 0.45. Among participants in the current study, the SES of the low risk group (Mean = 53.2, S.D. = 6.9) was signicantly higher than the high risk group (Mean = 37.3, S.D. = 13.3), F(1,37) = 16.38, P < 0.001. Participants in the current study and those from the larger cohort who did not participate in this study did not differ with respect to SES, F(1,216) = 0.01, P > 0.90. In addition, there was no signicant interaction between risk group status and participant status (participated/did not participate) in SES, F(1,216) = 2.61, P > 0.11. 5.2. Measures 5.2.1. Psychopathology Adolescent psychopathology (i.e., mood disorders, anxiety disorders, behavior disorders, and substance use disorders) was assessed at the rst evaluation with the Schedule for Affective Disorders and Schizophrenia for School-Age Children Epidemiological Version (K-SADS-E, Orvaschel et al., 1982) and with the Longitudinal Interval Follow-up Evaluation for Children (K-LIFE, Keller and Neilsen, 1988) at each follow-up assessment. All interviews were audiotaped. A second interviewer who was unaware of the rst interviewers ratings reviewed a randomly selected 25% of the interviews. Kappas were 0.81 for mood disorders, 0.72 for anxiety disorders, and 0.80 for behavior disorders. 5.2.1.1. Depressive symptoms. Two measures assessed adolescent depressive symptomatology. At the initial interview and at each annual follow-up interview, childrens depressive symptoms were assessed by the Childrens Depression Inventory (CDI), a widely used self-report measure of depressive symptoms in children (Kovacs, 1981). The CDI has adequate internal consistency, testretest reliability, and convergent validity with other self-report measures of depressive symptoms (Saylor et al., 1984; Smucker et al., 1986). The internal consistency of the CDI in this sample was 0.81 at the initial assessment. Additionally, weekly adolescent depressive symptomatology was ascertained based on the K-LIFE (Keller and Neilsen, 1988). Symptoms were dated and given a severity score that ranged from one (no symptoms of depression) to six (severe symptoms of depression). A score of ve or six on this scale denoted that the adolescent met DSM criteria for major depressive disorder, whereas lower scores denoted that the adolescent had not met criteria for major depressive disorder.
83
5.2.2. Socio-economic status Household SES was assessed with the four factor index of social status (Hollingshead, 1975), the most frequently used measure of SES (Cirino et al., 2002; Edwards-Hewitt and Gray, 1995). Possible scores on this index range from 8 (lowest SES) to 66 (highest SES). To calculate the SES score of a household, scale values for occupation (which range from one to nine) and for education (which range from one to seven) were multiplied by factor weights of ve and three, respectively. These two products were then summed. Then, adjustments were made for marital status and related factors (e.g., receipt of child-support or alimony payments from an absent spouse) as outlined in Hollingshead (1975). Adolescents were assigned the household SES score. 5.3. Procedure Participants were told that the purpose of the study was to look at brain wave activity in adolescents. After informed consent was obtained from both the adolescent and parent, electrodes were applied for the measurement of EEG. Participants were then informed that: (1) there would be eight 1 min resting baselines; (2) four baselines would be conducted with eyes-open and four would be conducted with eyes-closed; and (3) during the resting baselines, they should try to minimize eye blinks and movements, but should not be so concerned about doing so that they were distracted. In accord with previous work (e.g., Tomarken et al., 1990, 1992), participants were not given highly specic instructions concerning the resting baselines. Two randomly assigned, counterbalanced orders were used for the eyes-open and eyesclosed trials of the resting baselines (OCCOCOOC and COOCOCCO). Participants heard one tone denoting the beginning of each 60 s baseline and two tones denoting the end of each baseline. There was a 3 min interval between the fourth and fth baselines. A 45 s interval occurred between all other baselines. Following the eighth and nal resting baseline, electrodes were removed. 5.4. Electroencephalographic recording and quantication EEG recording followed standard guidelines (see Pivik et al., 1993). Recordings were made from tin scalp electrodes sewn into a Lycra stretchable cap from Electro-Cap International, Inc. (see Blom and Anneveldt, 1982). The cap was positioned on the head using the 1020 international system (American Electroencephalographic Society, 1994; Jasper, 1958). Fifteen standard scalp locations from the 10 to 20 system were used: F3, F4, F7, F8, T3, T4, T5, T6, P3, P4, C3, CZ, C4, Pz, and Fz. In addition, a forehead ground was used and tin drop electrodes from the cap were used to record from the left and right earlobes (A1A2). Nine millimeter tin cup electrodes were placed above and below the eyes to record blinks and vertical eye movements and on the outer canthi to record horizontal eye movements. The electrooculogram (EOG) was recorded using a bipolar reference, and EOG electrode impedances were under 15 k . Electrode impedances for EEG sites were under 5 k , and impedances for homologous sites were within 1 k of each other. Through pre- and post-recording checks we documented that impedances changed minimally during the course of the experiment.
84
Raw EEG and EOG signals were amplied and ltered using Grass Model 12 A5 AC pre-ampliers (bandpass at half-voltage cut-off points 1 and 100 Hz for EEG and 1 and 30 Hz for EOG; 60 Hz notch lter in, lter rolloff = 6 dB per octave). The gain was set at 30,000 for EEG channels and 5,000 for EOG channels. Data were digitized at 1024 Hz using an Analogue Devices RTI-815A analogue to digital converter interfaced to the signal acquisition package Snapstream (HEM Inc.). Eight 1 min resting baselines, four with eyes-open and four with eyes-closed, were collected. All placements were referenced to the vertex (Cz) during the initial recording. A set of 50 V sine waves at several different frequencies (e.g., 10 Hz) were used to calibrate the digitized EEG and to assess the technical integrity of the recording system. Calibration assessments were run both immediately before and immediately after each experimental session. Manual post-session artifact scoring with EEGEDIT software (James Long Company) was performed to edit the EEG signals. This procedure eliminated epochs that were confounded by artifacts such as movement, extensive muscle tension, and saccades. Following the artifact-reduction procedures, data were re-referenced ofine using James Long Company EEG Analysis System software. In particular, we performed linear transformations of the digitized EEG to derive a computer-averaged-ears reference and an computer-averaged reference (see Henriques and Davidson, 1990). Averaged-ears EEG power at a given site is the difference between activity at that site and the averaged power recorded across the two ears. Averaged reference EEG power at a given site is the difference between power at that site and the averaged power across all active sites. At least one of the three referencing schemes used here (Cz, averaged-ears, or average) has been used in previous EEG studies on the correlates of frontal asymmetry. All artifact-free chunks that were 2.00 s in duration were extracted through a Hanning window, used to prevent spurious estimates of spectral power. Chunks were overlapped by 50% to counteract the differential weighting of data points attributable to the use of a Hanning window. The EEG Analysis System software was then used to execute discrete Fourier transforms of the digitized EEG. This process derived estimates of spectral power (in V2 ) in different half-hertz frequency bins. These power values were then averaged across each of the artifact-free chunks of a given resting baseline trial. When a participant had fewer than eight artifact-free chunks for a given baseline, that baseline was not included in the computation; that is, it received a weight of zero. Power values were converted to power density ( V2 /Hz) in each of seven bands: delta (1.33.5 Hz), theta (4.07.0), alpha 1 (8.510.5 Hz), alpha 2 (11.012.5), alpha (8.512.5), beta 1 (13.519.5 Hz), and beta 2 (20.529.5 Hz). Power density was computed by summing power values across all the half-hertz bins within a band and then dividing by the number of summed bins. Consistent with the general procedures used in prior research on resting frontal asymmetry (e.g., Tomarken et al., 1992), a natural log transformation was used to normalize the distribution of power density values of a given baseline trial. We then computed weighted means separately for the eyes-open and eyes-closed baselines for each participant. We weighted by the number of chunks within each baseline. In the next step, we computed the average of the eyes-open and eyes-closed baselines to generate a composite measure of EEG power density. Such composite values were computed for each combination of site, band, and reference montage. Finally, asymmetry scores were computed for each combination of region, band, and reference montage by subtracting power density in the left-hemisphere
85
site from power density in the homologous right hemisphere site (i.e., log right minus log left). The high risk and low risk groups did not differ in the total number of artifact-free chunks per participant used in computations of the composite power and asymmetry measures (high risk Mean = 176.08, S.D. = 81.61; low risk Mean = 208.23, S.D. = 93.47, t (36) = 1.10, P > 0.25). To test hypotheses, we focused on measures of log power density in the alpha frequency band. Because decreased alpha power in a given region has been linked to increased cortical activity in that region (Davidson, 1988; Miller and Tomarken, 2001; Pfurtscheller, 1986; Pfurtscheller and Klimesch, 1991) higher values on the asymmetry metric denote greater relative left frontal activity. The selection of the 8.512.5 Hz band as the focus of hypotheses merits comment given the small number of prior asymmetry studies that have assessed adolescents. The clear majority of prior studies indicating linkages between frontal EEG asymmetry and emotion or psychopathology have used either adults or infants. Investigations of the relation between frontal brain asymmetry and depression in adults have dened alpha as 911 Hz (Schaffer et al., 1983), 912 Hz (Davidson et al., 1985), or 813 Hz (Henriques and Davidson, 1990, 1991). Whereas the dominant EEG frequency band in very young children is clearly lower (Marshall et al., 2002), the relative proportion of slow wave activity in children decreases with age (Benninger et al., 1984; Colon et al., 1979; Matousek and Petersen, 1973), until about 10 years of age (Benninger et al., 1984; Gasser et al., 1988; Matousek and Petersen, 1973). Concomitant with the decrease in slow wave power during childhood are increases in faster frequency activity (e.g., alpha 2 band activity, Gasser et al., 1988; John et al., 1980; Matousek and Petersen, 1973). Participants in the current study ranged in age from 12.2 to 14.0-year-old. Prior results would indicate that the majority of the transition from lower to higher frequency activity has been completed by this age (Gasser et al., 1988). For this reason, our primary focus of analyses was the adult alpha-band of 8.512.5 Hz. Our decision was consistent with the approach used by Kentgen et al. (2000), who assessed frontal EEG asymmetry among adolescents using an alpha-band equivalent of 7.812.5 Hz. We also report the results of exploratory analyses of other bands, with a particular interest in the theta band (47 Hz) containing frequencies immediately below the alpha range.
6. Results 6.1. Depressive symptoms We examined CDI scores from four of the yearly assessments to examine whether high and low risk participants differed on symptoms of depression prior to or after the EEG recording. Across the two yearly assessments preceding and the two yearly assessments following the EEG session, the mean low risk CDI was 3.9 (S.D. = 3.7) and the mean high risk CDI was 4.5 (S.D. = 3.4). Thus, both groups clearly scored in the non-depressed range. Separate t-tests on risk status conducted at each of the four time points revealed the absence of any differences with respect to CDI scores, all ts < 0.50, all Ps > 0.25. Consistent with these results, an omnibus risk status X time ANOVA failed to yield any signicant main effects or interactions, all Ps > 0.17.
86
During the eight weeks prior to the EEG session, the week of the EEG session, and the eight weeks following the EEG session, all participants averaged less than 2 (i.e., possible mild depressive symptoms) on the retrospective measure of weekly depressive symptoms (potential range = 16). All low risk participants received scores of one (no depressive symptoms) on this scale for every week during this time period. With the exception of four adolescents, all high risk participants received scores of two or below during this time period (i.e., mild symptoms of major depressive disorder). The remaining four received scores of three or four on this measure (indicating some symptoms of major depressive disorder and impairment, but falling short of the criteria for major depressive disorder). Thus, no low risk and high risk participants met criteria for major depressive disorder during the two-months prior to and the two months after the EEG recording. When the analyses reported below were redone with those four individuals who demonstrated symptoms of major depressive disorder removed, the results and conclusions were unchanged. 6.2. Lifetime criteria for mood disorders Based on the K-SADS-E administered at baseline and K-LIFE administered at the rst yearly follow-up interview, no low risk participant met lifetime criteria for any mood disorder. One high risk participant met lifetime criteria for dysthymia. When the analyses reported below were redone with this individual removed, the results and conclusions were unchanged. 6.3. Electroencephalographic data 6.3.1. Effects of risk status on mid-frontal alpha asymmetry As noted above, the primary focus of analyses was the alpha (8.512.5 Hz) frequency band. Because the majority of prior studies linking frontal brain asymmetry to depression have focused on the mid-frontal (F3/F4) recording sites, we focused on measures of alpha asymmetry in this region. We predicted that high risk participants would demonstrate greater relative left frontal hypo-activity (greater relative left versus right alpha-band power) than low risk participants. Table 1 shows mean log-transformed alpha power density (in V) values for the midfrontal sites (F4 and F3) and mean asymmetry scores [ln(F4) ln(F3)] derived from the three reference montages for males and females. Recall that more positive asymmetry scores indicate greater relative left frontal activity. Because of how the asymmetry metric is computed, a main effect of risk status on asymmetry values is equivalent to an interaction between risk status and hemisphere on log power density values. Analyses used to test hypotheses in the current study focused on asymmetry values rather than on log power density. When analyses were done on log power values from each of the three reference montages with hemisphere as a factor, no main effects of risk status were observed that would indicate between-group differences on overall power averaged across the two frontal sites (all Ps > 0.90). In addition, no between-group differences were yielded by separate analyses of F3 and F4 log power density (all Ps > 0.60). These results reect the likelihood that a high proportion of the between-subject variability in power is due to skull thickness and other factors that are not of substantive interest. For these reasons, to simplify the
A.J. Tomarken et al. / Biological Psychology 67 (2004) 77102 Table 1 Mid-frontal (F3/F4) alpha-band log power density (in V) and asymmetry for and low risk adolescents High Risk F3 power Averaged-ears Male 1.200 (0.579) Female 1.634 (0.540) All 1.442 (0.588) Average Male Female All 0.388 (0.589) 0.825 (0.576) 0.633 (0.611) F4 power F3/F4 asymmetry 0.009 (0.049) 0.004 (0.067) 0.001 (0.059) 0.009 (0.094) 0.001 (0.120) 0.004 (0.107) 0.021 (0.092) 0.005 (0.056) 0.007 (0.073) Low risk F3 power F4 power F3/F4 asymmetry
87
1.208 (0.572) 1.629 (0.536) 1.444 (0.581) 0.379 (0.563) 0.824 (0.574) 0.629 (0.601) 0.953 (0.598) 1.054 (0.580) 1.010 (0.578)
1.448 (0.353) 1.320 (0.572) 1.380 (0.451) 0.760 (0.558) 0.509 (0.604) 0.644 (0.570) 1.200 (0.553) 0.739 (0.721) 0.987 (0.653)
1.496 (0.372) 1.366 (0.551) 1.450 (0.448) 0.796 (0.579) 0.600 (0.612) 0.706 (0.578) 1.172 (0.543) 0.844 (0.709) 0.621 (0.172)
0.047 (0.041) 0.046 (0.057) 0.047 (0.047) 0.039 (0.126) 0.091 (0.086) 0.061 (0.114) 0.027 (0.082) 0.105 (0.035) 0.034 (0.093)
Vertex (Cz) Male 0.932 (0.637) Female 1.059 (0.576) All 1.002 (0.594)
Note: High risk male N = 11. High risk female N = 14. Low risk male N = 7. Low risk female N = 6. Standard deviations are indicated in parentheses.
presentation of results, we report below only the results of analyses performed on asymmetry values. Fig. 1 shows the mean log-transformed mid-frontal alpha asymmetry values, derived from three reference montages, for low risk and high risk adolescents. This gure indicates that across all three montages, there was greater relative left frontal activity in the low risk group compared to the high risk group. The results of risk status (low risk/high risk) X sex (male/female) ANOVAs performed on computer-averaged-ears referenced and average referenced mid-frontal EEG asymmetry values were consistent with these observations. The analysis of computer-averaged-ears referenced data revealed a signicant main effect of risk status, F(1,37) = 5.49, P < 0.05, but no signicant effects of sex, F(1,37) = 0.23, P > 0.50, or the risk status X sex interaction, F(1,37) = 0.09, P > 0.50. Similarly, the analysis of
0.10
Mid-frontal Alpha Asymmetry
High Risk Low Risk
0.05
0.00
-0.05
Ears Reference
Average Reference
Vertex (Cz) Reference
Fig. 1. Mid-frontal (F3/F4) alpha asymmetry [ln(left) ln(right)] across three reference montages for high risk (N = 25) and low risk (N = 13) adolescents. Error bars indicate one standard error of the mean.
88

0.15
Mid-frontal Alpha Asymmetry CZ Reference
High Risk 0.10 0.05 0.00 -0.05 -0.10 Low Risk
Males
Females
Fig. 2. Mid-frontal (F3/F4) vertex (Cz)-referenced alpha asymmetry [ln(left) ln(right)] for male and female high and low risk adolescents. Error bars indicate one standard error of the mean.
average referenced mid-frontal asymmetry revealed a signicant main effect of risk status, F(1,37) = 5.37, P < 0.05, but no signicant main effect of sex, F(1,37) = 0.50, P > 0.40, and no signicant interaction between risk status and sex, F(1,37) = 1.44, P > 0.20. Although the marginal means for vertex (Cz) referenced asymmetry values shown in Fig. 1 indicates greater relative left frontal activity in the low risk group, Fig. 2 indicates more complex, interactive relations with sex. As this gure indicates, high risk females, but not high risk males, demonstrated relative left frontal hypo-activity when compared to their low risk counterparts. The analysis of Cz asymmetry in the mid-frontal region revealed no main effect of risk status, F(1,37) = 1.21, P > 0.20, no main effect of sex, F(1,37) = 1.49, P > 0.20, but a signicant risk status X sex interaction, F(1,37) = 10.49, P < 0.01. Subsequent simple effects analyses indicated that, among females, high risk adolescents demonstrated signicantly greater left frontal hypo-activity when compared to low risk participants, F(1,19) = 19.43, P < 0.01. No signicant effects were yielded by the simple effects analysis performed on males asymmetry values, F(1,17) = 1.20, P > 0.25. 6.3.2. Effects of risk status on mid-frontal EEG Asymmetry in other bands We computed two-way risk status X sex ANOVAs on mid-frontal (F3/F4) EEG asymmetry in each of the six EEG bands extracted: delta (1.53.5 Hz), theta (4.07.0 Hz), alpha 1 (i.e., low alpha; 8.510.5), alpha 2 (i.e., high alpha; 11.012.5), beta 1 (13.519.5 Hz), and beta 2 (20.529.5 Hz). We assessed effects in alpha 1 and alpha 2 separately because: (1) in some previous studies, differential effects have been observed in these bands (e.g., Davidson et al., 2000); and, (2) the fact that our participants were young adolescents suggested that such effects might be particularly likely to occur (see the discussion of band selection in Section 5). Such analyses were computed for each of the three references. Because the number of ANOVAs was large, we used a step-down Bonferroni procedure to control for multiple signicance tests (e.g., Westfall et al., 1999; Westfall and Young, 1992). Three sets of corrections were used (i.e., one per reference montage). After step-down Bonferroni correction, there were highly signicant main effects for risk status in the theta band on both computer-averaged-ears referenced [corrected P < 0.001,
89
low risk M (S.D.) = 0.071 (0.077); high risk (M) = 0.032 (0.060)] and average referenced [corrected P < 0.001, low risk M (S.D.) = 0.120 (0.117); high risk (M) =0.037 (0.090)] mid-frontal asymmetry. We also found a signicant main effect of risk status on average referenced delta band asymmetry [corrected P < 0.025, low risk (M) S.D. = 0.110 (0.136); high risk M = 0.017 (0.121)] and on Cz-referenced asymmetry in the low alpha-band (corrected P < 0.02). These signicant effects all indicated that high risk participants showed greater relative power in the left relative to right frontal region for the target band. Thus, the direction of these effects parallels that of the alpha-band (8.512.5 Hz) effects reported above. No other signicant effects were observed. 6.3.3. Effects of risk status on EEG alpha asymmetry in other regions For each of the three references, we computed two-way risk status X sex ANOVAs on EEG asymmetry in each of ve regions: lateral frontal (F7/F8), parietal (P3/P4), anterior temporal (T3/T4), posterior temporal (T5/T6), and central (C3/C4). Because the clear focus of our predictions was alpha asymmetry in the mid-frontal sites and because the number of ANOVAs was large, we once again used a step-down Bonferroni procedure to control for multiple signicance tests. Across all three references, the only signicant effects that emerged were main effects of region (all Ps < 0.01). These effects reected topographic differences in the patterning of asymmetry. Most importantly, there were no signicant main effects or interactions involving risk status (all Ps > 0.05). Table 2 conveys the overall direction and strength of the relation between risk status and alpha-band asymmetry across regions. Presented are point biserial correlations between the dichotomous variable risk status (coded 0 for low risk and 1 for high risk) and measures of brain asymmetry [ln(right ln(left)] in a given band. Consistent with the main effects of risk status on mid-frontal asymmetry presented above, the correlations between risk status and ears-referenced (r = 0.37) and average-referenced (r = 0.36) mid-frontal asymmetry are both signicantly greater than 0. Although the correlation involving Cz-referenced frontal asymmetry is not signicant, recall that the ANOVAs revealed a more complex risk status X sex interaction on this measure. Although the correlations involving lateral
Table 2 Correlations between risk status and alpha-band asymmetry across regions Region Reference montage Averaged-ears Mid-frontal (F3F4) Lateral frontal (F7F8) Central (C3C4) Anterior temporal (T3T4) Posterior temporal (T5T6) Parietal (P3P4) 0.37 0.24 0.15 0.25 0.15 0.17 Average 0.36 0.25 0.16 0.06 0.19 0.19 Vertex (Cz) 0.16 0.09 0.08 0.08 0.14 0.35
Note: N = 38. Point-biserial correlations are shown. Risk status coding: 0 = low risk, 1 = high risk. Mid-frontal correlations were evaluated at a per-correlation alpha level = 0.05. Within each reference montage, step-down Bonferroni corrections were applied to correlations involving the ve other sites. Uncorrected P < 0.05, but step-down Bonferroni corrected P > 0.15. P < 0.05.
90
frontal asymmetry (F7/F8) are in the predicted direction, their magnitude is lower than that of the mid-frontal correlations and is not statistically signicant. Overall, other than the aforementioned mid-frontal correlations, only one of the correlations shown in Table 3 was statistically signicant when considered in isolation (Cz-referenced P3/P4; P = 0.03). Moreover, even this value was not statistically signicant when step-down Bonferroni corrections were used to account for the total number of correlations computed among the ve regions that were not of the focus of our initial hypotheses (P = 0.15). 6.3.4. Effects of socio-economic status 6.3.4.1. Zero-order correlations. One subsidiary goal was to assess the relation between SES and frontal asymmetry. This question is particularly salient in the present context because the high risk and low risk groups differed on SES. When participants in both risk status groups were pooled into one sample, we observed a signicant overall correlation between SES and alpha-band mid-frontal asymmetry for two of the three reference montages, computer-averaged-ears reference r = 0.57, P < 0.001, average reference r = 0.50, P < 0.001, Cz reference r = 0.16, P > 0.30. These correlations indicate that higher SES predicted greater relative left frontal activity. Fig. 3 shows scatter plots depicting the relation between frontal brain asymmetry (derived from three reference montages) and SES, with risk status symbolically indicated. Although this gure makes evident the small number of low SES participants in the low risk group, it also clearly illustrates the relation between relative left frontal activity and SES. To investigate further the relation between alpha asymmetry in the mid-frontal region and social class, correlations were computed separately for each risk group. Within the high risk group, frontal asymmetry measures derived using the computer-averaged-ears and average references both correlated signicantly with SES, computer-averaged-ears reference r = 0.56, P < 0.004, average reference r = 0.50, P < 0.01, Cz reference r = 0.14, P > 0.50. Within the low risk group, there were no signicant correlations between SES and frontal brain asymmetry, computer-averaged-ears reference r = 0.04, average reference r = 0.05, Cz reference r = 0.03, all Ps > 0.80. Clearly, however, caution is necessary here because: (1) differences between relevant pairs of correlations (e.g., high risk computer-averaged-ears versus low risk computer-averaged-ears) were not statistically signicant (all Ps > 0.10); (2) there was a restricted range of SES in the low risk group that could signicantly inuence the magnitude of the observed correlation (low risk M = 53.2, S.D. = 6.8; high risk M = 37.3, S.D. = 13.3) and, (3) due to the differences in variability, comparisons of unstandardized beta weights are likely more appropriate than comparisons of correlations (e.g., Tukey, 1954). Such comparisons constituted the SES X risk status interaction effects tested in the multiple regression analyses reported below. 6.3.4.2. Multiple regression results. To investigate further the relations among SES, frontal brain asymmetry, and risk status, we conducted multiple regression analyses in which risk status, sex, and SES were specied as predictors of mid-frontal asymmetry in the alpha-band. These analyses were designed to test the unique effects of each of the three predictors on frontal asymmetry and to test for moderator effects. For example, a signicant risk status X SES two-way interaction would suggest that the magnitude of the relation between SES and
91
Fig. 3. Scatter plots of the relation between SES and mid-frontal (F3/F4) alpha asymmetry for each of three reference montages. Top panel: averaged-ears reference; middle panel: average reference; bottom panel: vertex (Cz) reference; H: high risk adolescents; L: low risk adolescents.
92
Table 3 Multiple regression analyses predicting mid-frontal alpha asymmetry Predictors Mid-frontal asymmetry measure Ears reference First-order Risk status Sr2 Sex Sr2 SES Sr2 R2 Increment (set) Two-way interactions Risk status sex Sr2 Risk status SES Sr2 Sex SES Sr2 R2 Increment (set) Three-way interaction Risk status sex SES Sr2 R2 Increment (set) Average reference Vertex (Cz)
0.067 0.003 0.094 0.009 0.529 0.191 0.336
0.128 0.011 0.096 0.009 0.432 0.128 0.270
0.122 0.010 0.175 0.030 0.096 0.006 0.064
0.005 0.000 0.211 0.011 0.170 0.016 0.054
0.136 0.011 0.069 0.001 0.243 0.032 0.102
0.631 0.226 0.249 0.016 0.120 0.008 0.277
0.344 0.028 0.028
0.004 0.000 0.000
0.091 0.002 0.002
A hierarchical structure was used in which rst-order terms were entered in an initial step, followed by two-interaction terms, and the three-way interaction in subsequent steps. For rst-order terms, the s are standardized coefcients. For second-order terms, the s are the unstandardized coefcients for terms that are the product of standardized variables but are not standardized themselves. This procedure was followed in order to yield test statistics and probability values that are invariant with respect to the unstandardized analyses of the raw data values. Sr2 : squared semi-partial correlation. R2 Increment (set): the increment in proportion of variance accounted for by the set of predictors entered in a given step. P < 0.025. P < 0.005.
frontal asymmetry is conditional on risk status and, conversely, that the relation between risk status and frontal asymmetry is conditional on level of SES. A hierarchical structure (Cohen and Cohen, 1983) was used for the three analyses (one per reference). In the rst step, the three predictors were entered simultaneously (risk status, sex, and SES). This step was used to test for the main effects of each predictor on frontal
93
asymmetry removing shared variance with the other predictors. In the second step, the three two-way interaction terms (risk status X SES, risk status X sex, sex X SES) were entered as a set. In the third step, we entered the risk status X sex X SES three-way interaction term. At each step, we tested the statistical signicance of each of the individual regression coefcients in the set just entered and the signicance of the incremental variance attributable to the set. To facilitate the interpretation of coefcients, we centered SES (i.e., expressed it as deviations from its mean, thus resulting in a transformed mean of 0, e.g., Aiken and West, 1991). We used dummy codes for the categorical variables of risk status (0 = low risk, 1 = high risk) and sex (0 = females, 1 = males). It is important to note that identical results and conclusions were yielded when alternative model-testing strategies were used (e.g., for descriptions of alternative approaches, see, e.g., Aiken and West, 1991). Table 3 shows the results of the regressions on alpha-band asymmetry for each of the three references. To provide an interpretable metric, this table displays standardized regression coefcients s for each rst-order (i.e., main effect) term (risk status, sex, SES). The test statistics and P values for such rst-order coefcients are identical to those for the unstandardized coefcients that were yielded by an analysis of the raw data values in their original metric. In the case of two- and three-way interactions, the coefcients shown in Table 3 are actually the unstandardized coefcients yielded by an analysis of interaction terms that were the product of standardized rst-order terms. Such interaction terms were not, however, themselves standardized. By this means, we were able to present interaction coefcients that had both a reasonably interpretable metric and test statistics and signicance levels identical to those yielded by analyses of the raw data values in their original metric (Aiken and West, 1991; Friedrich, 1982). Also shown in Table 3 are squared semi-partial correlations denoting the proportion of the total variance in asymmetry scores uniquely attributable to each predictor (see, e.g., Cohen and Cohen, 1983) and R2 values indicating the increment in variance accounted for by each set of predictors. As indicated by Table 3, the results of the regressions of mid-frontal asymmetry on the rst-order terms (risk status, SES, and sex) were consistent across the ears referenced and average referenced montages. In each case, SES, but not risk status or sex, contributed signicant unique variance to the prediction of alpha-band asymmetry (ears referenced P < 0.005, average referenced P < 0.025). The squared semi-partial correlation coefcients shown in Table 3 indicate that SES accounted for a notably higher proportion of the variance in ears referenced and average referenced frontal asymmetry scores than the other two predictors (see Table 3). We should emphasize that the effects shown for each of the three predictors in Table 3 are adjusted for the effects of the other two predictors in the equation (i.e., the three predictors were entered simultaneously as a set). When Cz-referenced mid-frontal asymmetry was the outcome variable, no effects for rst-order terms were signicant. In subsequent steps testing for moderation, we entered the two-way and three-way interaction terms in the regression equations. As shown in Table 2, the only signicant effect across all three dependent measures was the risk status X sex interaction on Cz-referenced mid-frontal asymmetry (P < 0.005). This effect was also yielded by the risk status X sex ANOVA reported above and reects greater differences between the high and low risk groups among females relative to males. There were no signicant two- or three-way interactions involving SES (all Ps > 0.20).
94
6.3.4.3. Components of SES. We also examined the relations between the components of SES and alpha-band asymmetry. Our measure of SES (Hollingshead, 1975) is derived primarily from parental occupation, education, and marital status. We rst calculated zero-order correlations between risk status and these three components of SES. To orient readers to the scaling and direction of relations, occupation ranged from 0 (unemployed) to 9 (higher professional), education ranged from 1 (less that 6 years of schooling) to 7 (more than 18 years of schooling), and marital status was coded as either 0 (unmarried) or (1) married. The high and low risk groups differed on these components of SES in a manner that paralleled the differences on the composite SES index reported above (occupation P < 0.001; education P < 0.005; marital Status P < 0.02). Parents of low risk participants were more likely to have attained higher occupation and educational levels and were more likely to be married. When participants in both risk status groups were pooled into one sample, we observed a signicant overall correlation between occupation and alpha-band mid-frontal asymmetry for two of the three reference montages, ears reference r = 0.50, P < 0.002, average reference r = 0.45, P < 0.01, Cz reference r = 0.06, P > 0.50. We also observed a signicant overall correlation between education and alpha-band mid-frontal asymmetry for two of the three reference montages, ears reference r = 0.33, P < 0.05, average reference r = 0.38, P < 0.05, Cz reference r = 0.19, P > 0.20. Finally, we observed a signicant overall correlation between marital status and alpha-band mid-frontal asymmetry for one of the three reference montages, ears reference r = 0.42, P < 0.0.01, average reference r = 0.20, P > 0.20, Cz reference r = 0.12, P > 0.40. Thus, higher occupation, more years of education, and being married generally predicted greater relative left frontal activity.
7. Discussion 7.1. Primary hypotheses The primary goal of the present study was to assess whether children of mothers with a history of depression demonstrated relative left frontal hypo-activity when compared to low risk children. Across the three reference montages assessed, analyses supported predictions. The risk status X sex ANOVAs performed on ears referenced and average referenced mid-frontal asymmetry values indicated that high risk participants demonstrated relative left frontal hypo-activity when compared to low risk participants.1 The pattern of effects on the Cz-referenced mid-frontal asymmetry measures was more complex. The two risk status groups differed in frontal asymmetry among females but not among males. These latter ndings are intriguing and may link up meaningfully with the evidence for sex differences in depression that emerge during early adolescence (e.g., Heller, 1993; Hankin et al., 1998; Nolen-Hoeksema and Girgus, 1994).
1 As indicated by Fig. 1, the low risk group on average demonstrated relative left frontal hyper-activity (i.e., greater alpha suppression in the left relative to right hemisphere) whereas the high risk group demonstrated a more symmetrical pattern. In this regard, it is relevant to note that unselected adult participants typically demonstrate relative left frontal activity on these measures (Reid et al., 1998; Tomarken et al., 1992). Thus, although the high risk group did not demonstrate left frontal hypo-activity in an absolute sense, their pattern does appear to deviate from the norm.
95
From a broad perspective, each of the three references provided support for the notion that resting frontal asymmetry may indicate heightened vulnerability to depression. This is an important observation because, as noted above, questions have been raised about the degree to which mid-frontal EEG asymmetry ndings are consistent across references (e.g., Hagemann et al., 1998; Reid et al., 1998). The degree of convergence that we found suggests that we may have tapped into a robust phenomenon. The present results were also generalizable across several different EEG bands. For example, we found particularly notable effects of risk status on theta band measures of ears referenced and averaged referenced asymmetry and additional effects on delta band measures of average referenced asymmetry. This convergence across the lower-frequency EEG components may be related to the fact that participants were young adolescents. Our lower-band asymmetry effects merit replication and extension. Our alpha asymmetry effects were, however, not generalizable across EEG sites. The only signicant effects that emerged were for the mid-frontal region. 7.2. Implications of risk status ndings Our results signicantly extend previous ndings indicating that relative left frontal hypo-activity may indicate vulnerability to depression. Prior studies using adult participants have compared non-depressed samples to either currently depressed (e.g., Schaffer et al., 1983) or remitted samples (Henriques and Davidson, 1990). As noted above, both comparisons do not allow unambiguous inference concerning whether resting frontal asymmetry indicates vulnerability. More compelling evidence has been provided by studies comparing infants of depressed and non-depressed mothers (e.g., Dawson et al., 1997; Jones et al., 1997). Our ndings link up with these latter results and suggest that there may be continuity of risk that extends beyond infancy. If frontal brain asymmetry is an indicator of risk that demonstrates such continuity, it should be stable over time. Prior evidence suggests that resting asymmetry is moderately stable in both adults (Tomarken et al., 1992) and in infants and young children (e.g., Jones et al., 1997). Our ndings mandate an examination of the stability of resting frontal asymmetry among adolescents. Of course, if frontal brain asymmetry indicates differential risk for depression, it should predict the long-term onset and/or maintenance of depressive symptoms. In this regard, the 1214 year age range that we have studied is ideal because this is the point in development immediately prior to the increased onset of depressive symptoms that occurs in mid-adolescence (e.g., Hankin et al., 1998; Nolen-Hoeksema and Girgus, 1994). There is a need for longitudinal studies that address whether those adolescents characterized by relative left frontal hypo-activity are most likely to demonstrate an increase in depressive symptoms and a higher proportion of diagnosable episodes over time. We believe that such effects are most likely to be observed in interaction with other factors (e.g., stressful events). 7.3. Effects of socio-economic status We assessed the effects of SES because it has been largely neglected in prior studies of anterior brain asymmetry and because there is strong evidence for inverse relations between SES and rates of unipolar depression (e.g., Leventhal and Brooks-Gunn, 2000; Murphy
96
et al., 1991). Consistent with the latter results, in the present study high and low risk groups differed notably in mean household SES on the Hollingshead index. Our correlational results indicated reasonably strong relations between SES and mid-frontal alpha asymmetry for two of the three reference montages assessed. That is, higher SES predicted relative left frontal activity. We also found a signicant relation between SES and mid-frontal asymmetry among high risk participants alone. It is unclear precisely, why SES is correlated with mid-frontal asymmetry. The major theories concerning the linkage between SES and psychopathology are one basis for speculation (Johnson et al., 1999). According to social causation theory, the chronic stress and other adversities linked to low SES contribute to the onset of psychopathology. According to social selection theory, biological, and environmental factors contribute to the onset of psychopathology, which, in turn, induces downward social drift toward lower socio-economic classes. Although current research favors the social causation hypothesis in application to unipolar depression, there are ndings that support both views (for a review, see Johnson et al., 1999). Extensions of both these hypotheses might provide a basis for speculation about the nature of the relation between SES and frontal asymmetry. For example, a variant of the social causation hypothesis would be the proposal that the chronic stress and other adversities associated with low SES produce long-term changes in brain asymmetry. Although no human studies have addressed this issue, there is infrahuman evidence that stress can induce changes in lateralization of neurotransmitter functions that appears correlated with anxiety-related behaviors (e.g., Fride and Weinstock, 1988). In addition, there are a variety of factors other than chronic stress per se that are correlated with SES and could conceivably produce effects on brain asymmetry. Such factors include maternal warmth, peer group instability, social support, and cognitive stimulation (for a review, see, e.g., Dodge et al., 1994). 7.4. Joint effects of maternal history of depression and socio-economic status Our multiple regression analyses addressed the unique and interactive effects of maternal history (denoted risk status), sex, and SES on measures of mid-frontal brain asymmetry derived from the alpha-band. Somewhat surprisingly, SES but not risk status signicantly predicted asymmetry measures derived from the averaged-ears and average references. No other main effects or interactions were signicant. In contrast, we observed a signicant risk status X sex interaction on Cz-referenced frontal asymmetry values. From a causal modeling perspective (e.g., Baron and Kenny, 1986; Bollen, 1989), the overall pattern of correlational and multiple regression results that we observed on ears referenced and averaged referenced frontal asymmetry might be taken to indicate that SES has a direct causal effect on frontal brain asymmetry but that maternal history of depression has only a weak direct effect at best. One possible model consistent with the data would specify that SES largely mediates the effects of maternal history of depression on brain asymmetry among children. This model would stipulate that: (1) maternal episodes of depression or related vulnerability factors induce a downward drift in SES; and, (2) lowered SES, in turn, is a proximal cause of changes in brain asymmetry among the children of such mothers. This model has features of both the social causation and social selection perspectives on the relation between SES and psychopathology.
97
Although such an interpretation might initially appear plausible, there are a host of reasons why it is premature. First, this conclusion is based on data that if, not cross-sectional in a strict sense, embody many of the features and limitations of cross-sectional data. Clearly, a more systematic evaluation of the causal relations among these variables would require a longitudinal design involving repeated assessments over time and across generations (see, e.g., Johnson et al., 1999). Estimates of causal parameters can be seriously misleading when cross-sectional designs are used to model effects that, by their very nature, occur over time (e.g., Cole and Maxwell, 2003). A second issue that complicates interpretation of our regression analyses is the problem of omitted variables (e.g., Berk, 2004; Tomarken and Waller, 2003). Undoubtedly, there are a number of relevant variables that are correlated with risk status, SES, and/or frontal brain asymmetry and may be implicated in a complex, multi-factorial causal nexus. For example, we did not assess chronic stress or frontal brain asymmetry in the parents of adolescents. Both factors could well have causal linkages to the variables assessed in the present study (e.g., Field et al., 1995; Kessler, 1997). It is well known that the omission of important predictors from regression analyses can strongly bias estimates of coefcients and result in highly misleading causal inferences (e.g., Gollob, 1991; Reichardt and Gollob, 1986). There are additional factors limit that limit conclusions. Via effects on univariate and joint distributions, the general absence of low SES participants from the low risk group and the overall approach used to sample participants might have signicantly affected the results of the multiple regression analyses (e.g., McClelland and Judd, 1993; Sher and Trull, 1996). For example, these factors might have signicantly lowered the power of tests for moderation. In addition, risk status was treated as a dichotomous variable. A continuous measure of maternal history might have yielded more robust effects. Finally, some prior ndings in this area may be inconsistent with a model stipulating that maternal depression has only indirect effects on brain asymmetry that are mediated by SES. For example, this model cannot easily accommodate evidence that infants of depressed and non-depressed mothers can differ on frontal EEG asymmetry even when the two groups fail to differ on measures of SES (e.g., Field et al., 1995). In sum, although the results of our regression analyses are intriguing, they should be viewed with caution and certainly require replication and extension in the context of a longitudinal design that includes additional variables. We should also note that the ultimate goal of such a research program is not simply to clarify the effects of maternal history of depression and SES on frontal brain asymmetry. Rather, it is to examine the joint effects of all three variables on the onset and maintenance of depression.
8. Summary and conclusions In accord with predictions, we found that adolescent offspring of mothers with a history of depression demonstrated relative left frontal hypo-activity relative to low risk adolescents. At least some support for this hypothesis was found across all three reference montages assessed. Such effects were specic to the mid-frontal region. We also found that lower SES predicted greater relative left frontal hypo-activity. This linkage remained signicant even when we controlled for maternal history of depression. Further longitudinal studies
98
that include a broader array of variables are necessary to clarify the precise causal linkages among these variables. Acknowledgements This project was funded in part by NIMH grant MH49759 awarded to Andrew J. Tomarken, Ph.D., by a FIRST Award (R29-MH4545801A1) from the National Institute of Mental Health, a Faculty Scholar Award (88-1214-88), and grant (96173096) from the William T. Grant Foundation awarded to Judy Garber. Gabriel Dichter was supported in part from an NIMH training grant (T32-MH18921). We appreciate the cooperation of the Nashville Metropolitan School District, Drs. Binkley and Crouch, and we thank the parents and children who participated in the project.
References
Aiken, L.S., West, S.G., 1991. Multiple Regression: Testing and Interpreting Interactions. Sage Publications, Newbury Park, CA. Allen, J.J., Iacono, W.G., Depue, R.A., Arbisi, P., 1993. Regional electroencephalographic asymmetries in bipolar seasonal affective disorder before and after exposure to bright light. Biological Psychiatry 33 (89), 642646. Alloy, L.B., Abramson, L.Y., Raniere, D., Dyller, I.M., 1999. Research Methods in Adult Psychopathology. In: Holmbeck, G.N. (Ed.), Handbook of Research Methods in Clinical Psychology, second ed., John Wiley & Sons Inc., New York, pp. 466498 American Electroencephalographic Society, 1994. Guideline thirteen: guidelines for standard electrode position nomenclature. Journal of Clinical Neurophysiology 11, 111113. Baron, R.M., Kenny, D.A., 1986. The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology 51, 1173 1182. Baxter Jr., L.R., Schwartz, J.M., Phelps, M.E., Mazziotta, J.C., Guze, B.H., Selin, C.E., Gerner, R.H., Sumida, R.M., 1989. Reduction of prefrontal cortex glucose metabolism common to three types of depression. Archives of General Psychiatry 46 (3), 243250. Beck, A.T., Steer, R.A., 1987. BDI, Beck depression inventory: manual. Psychological Corp., Harcourt Brace Jovanovich, San Antonio, TX. Bench, C.J., Friston, K.J., Brown, R.G., Scott, L.C., Frackowiak, R.S., Dolan, R.J., 1992. The anatomy of melancholiafocal abnormalities of cerebral blood ow in major depression. Psychological Medicine 22 (3), 607615. Benninger, C., Matthis, P., Scheffner, D., 1984. EEG development of healthy boys and girls. Results of a longitudinal study. Electroencephalography and Clinical Neurophysiology 57 (1), 112. Berk, R.A., 2004. Regression analysis: a constructive critique. Thousand Oaks, Sage, CA. Blom, J.L., Anneveldt, M., 1982. An electrode cap tested. Electroencephalography and Clinical Neurophysiology 54 (5), 591594. Bollen, K.A., 1989. Structural Equations with Latent Variables. Wiley/Interscience, New York. Cirino, P.T., Chin, C.E., Sevcik, R.A., Wolf, M., Lovett, M., Morris, R.D., 2002. Measuring socioeconomic status: reliability and preliminary validity for different approaches. Assessment 9 (2), 145155. Cohen, J., 1960. A coefcient of agreement for nominal scales. Educational and Psychological Measurement 20, 3746. Cohen, J., Cohen, P., 1983. Applied Multiple Regression/Correlation Analyses for the Behavioral Sciences, second ed. Hillsdale, Lawrence Erlbaum, NJ. Cole, D.A., Maxwell, S.E., 2003. Testing mediational models with longitudinal data: Questions and tips in the use of structural equation modeling. Journal of Abnormal Psychology 112, 558577.
99
Colon, E.J., de Weerd, J.P., Notermans, S.L., de Graaf, R., 1979. EEG spectra in children aged 8, 9 and 10 years. Reference values. Journal of Neurology 221 (4), 263268. Dam, H., Pedersen, H.E., Ahlgren, P., 1989. Depression among patients with stroke. Acta Psychiatry Scandinavica 80 (2), 118124. Davidson, R.J., 1988. EEG measures of cerebral asymmetry: conceptual and methodological issues. The International Journal of Neuroscience 39 (12), 7189. Davidson, R.J., 1995. Brain asymmetry. In: Hugdahl, K. (Ed.), Cerebral Asymmetry, Emotion, and Affective Style. The MIT Press, Cambridge, MA, pp. 361387. Davidson, R.J., 1998a. Affective style and affective disorders: perspectives from affective neuroscience. Cognition and Emotion 12 (3), 307330. Davidson, R.J., 1998b. Anterior electrophysiological asymmetries, emotion, and depression: conceptual and methodological conundrums. Psychophysiology 35 (5), 607614. Davidson, R.J., Marshall, J.R., Tomarken, A.J., Henriques, J.B., 2000. While a phobic waits: regional brain electrical and autonomic activity in social phobics during anticipation of public speaking. Biological Psychiatry 47 (2), 8595. Davidson, R.J., Pizzagalli, D., Nitschke, J.B., Putnam, K., 2002. Depression: perspectives from affective neuroscience. Annual Review of Psychology 53, 545574. Davidson, R.J., Schaffer, C.E., Saron, C., 1985. Effects of lateralized presentations of faces on self-reports of emotion and EEG asymmetry in depressed and non-depressed subjects. Psychophysiology 22 (3), 353364. Davidson, R.J., Tomarken, A.J., 1989. Laterality and emotion: an electrophysiological approach. In: Boller, F., Grafman, J. (Eds.), Handbook of Neuropsychology. Amsterdam, Elsevier, pp. 419441. Dawson, G., Frey, K., Panagiotides, H., Osterling, J., Hessl, D., 1997. Infants of depressed mothers exhibit atypical frontal brain activity: a replication and extension of previous ndings. Journal of Child Psychology and Psychiatry 38 (2), 179186. Dawson, G., Frey, K., Panagiotides, H., Yamada, E., Hessl, D., Osterling, J., 1999a. Infants of depressed mothers exhibit atypical frontal electrical brain activity during interactions with mother and with a familiar, nondepressed adult. Child Development 70 (5), 10581066. Dawson, G., Frey, K., Self, J., Panagiotides, H., Hessl, D., Yamada, E., Rinaldi, J., 1999b. Frontal brain electrical activity in infants of depressed and nondepressed mothers: relation to variations in infant behavior. Development and Psychopathology 11 (3), 589605. Dawson, G., Klinger, L.G., Panagiotides, H., Hill, D., Spieker, S., 1992a. Frontal lobe activity and affective behavior of infants of mothers with depressive symptoms. Child Development 63 (3), 725737. Dawson, G., Panagiotides, H., Klinger, L.G., Hill, D., 1992b. The role of frontal lobe functioning in the development of infant self-regulatory behavior. Brain Cognition 20 (1), 152175. Dodge, K., Pettit, G.S., Bates, J.E., 1994. Socialization mediators of the relation between socioeconomic status and child conduct problems. Child Development 65 (2 Spec No), 649665. Dohrenwend, B.P., 2000. The role of adversity and stress in psychopathology: some evidence and its implications for theory and research. Journal of Health and Social Behavior 41 (1), 119. Downey, G., Coyne, J.C., 1990. Children of depressed parents: an integrative review. Psychological Bulletin 108 (1), 5076. Ebert, D., Feistel, H., Barocka, A., 1991. Effects of sleep deprivation on the limbic system and the frontal lobes in affective disorders: a study with Tc-99m-HMPAO SPECT. Psychiatry Research 40 (4), 247251. Edwards-Hewitt, T., Gray, J.J., 1995. Comparison of measures of socioeconomic status between ethnic groups. Psychological Reports 77 (2), 699702. Field, T., Fox, N.A., Pickens, J., Nawrocki, T., 1995. Relative frontal EEG activation in 3- to 6-month-old infants of depressed mothers. Development and Psychology 31 (3), 358363. Fox, J.W., 1990. Social class, mental illness, and social mobility: the social selection-drift hypothesis for serious mental illness. Journal of Health and Social Behavior 31 (4), 344353. Fox, N.A., 1991. If its not left, its right. Electroencephalograph asymmetry and the development of emotion. American Psychologist 46 (8), 863872. Fride, E., Weinstock, M., 1988. Prenatal stress increases anxiety related behavior and alters cerebral lateralization of dopamine activity. Life Science 42 (10), 10591065. Friedrich, R.J., 1982. In defense of multiplicative terms in multiple regression equations. American Journal of Political Science 26, 797833.
100
Gasser, T., Verleger, R., Bacher, P., Sroka, L., 1988. Development of the EEG of school-age children and adolescents. I. Analysis of band power. Electroencephalography and Clinical Neurophysiology 69 (2), 9199. Gelfand, D.M., Teti, D.M., 1990. The effects of maternal depression on children. Clinical Psychology Review 10 (3), 329353. George, M.S., Wassermann, E.M., Kimbrell, T.A., Little, J.T., Williams, W.E., Danielson, A.L., Greenberg, B.D., Hallett, M., Post, R.M., 1997. Mood improvement following daily left prefrontal repetitive transcranial magnetic stimulation in patients with depression: a placebo-controlled crossover trial. American Journal of Psychiatry 154 (12), 17521756. Gollob, H.F., 1991. Interpreting and estimating indirect effects assuming time lags matter. In: Collins, L., Horn, J. (Eds.), Best Methods for the Analysis of Change. Recent Advances, Unanswered Questions, Future Directions, American Psychological Association, Washington DC, pp. 243259. Goodman, E., 1999. The role of socioeconomic status gradients in explaining differences in US adolescents health. American Journal of Public Health 89 (10), 15221528. Hagemann, D., Naumann, E., Becker, G., Maier, S., Bartussek, D., 1998. Frontal brain asymmetry and affective style: a conceptual replication. Psychophysiology 35 (4), 372388. Hammen, C., 1991. Generation of stress in the course of unipolar depression. Journal of Abnormal Psychology 100 (4), 555561. Hankin, B.L., Abramson, L.Y., Moftt, T.E., Silva, P.A., McGee, R., Angell, K.E., 1998. Development of depression from preadolescence to young adulthood: emerging gender differences in a 10-year longitudinal study. Journal of Abnormal Psychology 107, 128140. Heller, W., 1993. Gender differences in depression: perspectives from neuropsychology. Journal of Affective Disorder 29 (23), 129143. Henriques, J.B., Davidson, R.J., 1990. Regional brain electrical asymmetries discriminate between previously depressed and healthy control subjects. Journal of Abnormal Psychology 99 (1), 2231. Henriques, J.B., Davidson, R.J., 1991. Left frontal hypoactivation in depression. Journal of Abnormal Psychology 100 (4), 535545. Hollingshead, A.B., 1975. Four factor index of social status. Working paper. Department of Sociology, Yale University, New Haven, CT. Hollingshead, A.d.B., Redlich, F.C., 1958. Social class and mental illness; a community study. Wiley, New York. House, A., Dennis, M., Warlow, C., Hawton, K., Molyneux, A., 1990. Mood disorders after stroke and their relation to lesion location. A CT scan study. Brain 113 (Pt 4), 11131129. Jasper, H.H., 1958. The 1020-electrode system of the International Federation. Electroencephalography and Clinical Neurophysiology 10, 370375. John, E.R., Ahn, H., Prichep, L., Trepetin, M., Brown, D., Kaye, H., 1980. Developmental equations for the electroencephalogram. Science 210 (4475), 12551258. Johnson, J.G., Cohen, P., Dohrenwend, B.P., Link, B.G., Brook, J.S., 1999. A longitudinal investigation of social causation and social selection processes involved in the association between socioeconomic status and psychiatric disorders. Journal of Abnormal Psychology 108 (3), 490499. Jones, N.A., Field, T., Davalos, M., Pickens, J., 1997. EEG stability in infants/children of depressed mothers. Child Psychiatry and Human Development 28 (2), 5970. Jones, N.A., Field, T., Fox, N.A., Lundy, B., Davalos, M., 1997. EEG activation in 1-month-old infants of depressed mothers. Development and Psychopathology 9 (3), 491505. Kaplan, G.A., Roberts, R.E., Camacho, T.C., Coyne, J.C., 1987. Psychosocial predictors of depression. Prospective evidence from the human population laboratory studies. American Journal of Epidemiology 125 (2), 206220. Keller, M.B., Neilsen, E., 1988. The Kiddie-LIFE: Longitudinal Course. The Massachusetts General Hospital, Boston, MA. Kentgen, L.M., Tenke, C.E., Pine, D.S., Fong, R., Klein, R.G., Bruder, G.E., 2000. Electroencephalographic asymmetries in adolescents with major depression: inuence of comorbidity with anxiety disorders. Journal of Abnormal Psychology 109 (4), 797802. Kessler, R.C., 1997. The effects of stressful life events on depression. Annual Review of Psychology 48, 191214. Kovacs, M., 1981. Rating scales to assess depression in school-aged children. Acta Paedopsychiatry 46 (56), 305315. Leventhal, T., Brooks-Gunn, J., 2000. The neighborhoods they live in: the effects of neighborhood residence on child and adolescent outcomes. Psychological Bulletin 126 (2), 309337.
101
MacHale, S.M., ORourke, S.J., Wardlaw, J.M., Dennis, M.S., 1998. Depression and its relation to lesion location after stroke. Journal of Neurosurgery and Psychiatry 64 (3), 371374. Marshall, P.J., Bar-Haim, Y., Fox, N.A., 2002. Development of the EEG from 5 months to 4 years of age. Clinical Neurophysiology 113 (8), 11991208. Martinot, J.L., Hardy, P., Feline, A., Huret, J.D., Mazoyer, B., Attar-Levy, D., Pappata, S., Syrota, A., 1990. Left prefrontal glucose hypometabolism in the depressed state: a conrmation. American Journal of Psychiatry 147 (10), 13131317. Matousek, M., Petersen, I., 1973. Frequency analysis of the EEG in normal children and adolescents. In: Petersen, I. (Ed.), Automation of Clinical Electroencephalography. Raven Press, New York, pp. 75102. McClelland, G.H., Judd, C.M., 1993. Statistical difculties of detecting interactions and moderator effects. Psychological Bulletin 114 (2), 376390. Miller, A., Tomarken, A.J., 2001. Task-dependent changes in frontal brain asymmetry: effects of incentive cues, outcome expectancies, and motor responses. Psychophysiology 38 (3), 500511. Morris, P.L., Robinson, R.G., Raphael, B., Hopwood, M.J., 1996. Lesion location and poststroke depression. The Journal of Neuropsychiatry and Clinical Neurosciences 8 (4), 399403. Murphy, J.M., Olivier, D.C., Monson, R.R., Sobol, A.M., Federman, E.B., Leighton, A.H., 1991. Depression and anxiety in relation to social status. A prospective epidemiologic study. Archives of General Psychiatry 48 (3), 223229. Nolen-Hoeksema, S., Girgus, J.S., 1994. The emergence of gender differences in depression during adolescence. Psychological Bulletin 115 (3), 424443. Oldeld, R.C., 1971. The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia 9 (1), 97113. Orvaschel, H., Puig-Antich, J., Chambers, W., Tabrizi, M.A., Johnson, R., 1982. Retrospective assessment of prepubertal major depression with the Kiddie-SADS-e. Journal of the American Academy of Child Psychiatry 21 (4), 392397. Pascual-Leone, A., Rubio, B., Pallardo, F., Catala, M.D., 1996. Rapid-rate transcranial magnetic stimulation of left dorsolateral prefrontal cortex in drug-resistant depression. Lancet 348 (9022), 233237. Pearlin, L.I., Johnson, J.S., 1977. Marital status, life-strains and depression. American Sociology Review 42 (5), 704715. Pfurtscheller, G., 1986. Event-related desynchronization mapping: visualization of cortical activation patterns. In: Duffy, F.H. (Ed.), Topographic Mapping of Brain Electrical Activities, Stoneham, MA, Butterworth, pp. 99111. Pfurtscheller, G., Klimesch, W., 1991. Event-related desynchronization during motor behavior and visual information processing. In: Verbaten, M.N. (Ed.), Event-Related Brain Research, vol. 42. Amsterdam, Elsevier, pp. 5865. Pivik, R.T., Broughton, R.J., Coppola, R., Davidson, R.J., Fox, N., Nuwer, M.R., 1993. Guidelines for the recording and quantitative analysis of electroencephalographic activity in research contexts. Psychophysiology 30 (6), 547558. Pohjasvaara, T., Leskela, M., Vataja, R., Kalska, H., Ylikoski, R., Hietanen, M., Leppavuori, A., Kaste, M., Erkinjuntti, T., 2002. Post-stroke depression, executive dysfunction and functional outcome. European Journal of Neurology 9 (3), 269275. Reichardt, C.S., Gollob, H.F., 1986. Satisfying the constraints of causal modelling. In: Trochim, W. (Ed.), Advances in quasi-experimental design and analysis. New Directions for Program Evaluation Series Number 31. San Francisco, Jossey-Bass. Reid, S.A., Duke, L.M., Allen, J.J., 1998. Resting frontal electroencephalographic asymmetry in depression: inconsistencies suggest the need to identify mediating factors. Psychophysiology 35 (4), 389404. Robinson, R.G., Downhill, J.E., 1995. Lateralization of psychopathology in response to focal brain injury. In: Davidson, R.J., Hugdahl, K. (Eds.), Brain Asymmetry, MIT Press, Cambridge, MA, pp. 693711. Robinson, R.G., Kubos, K.L., Starr, L.B., Rao, K., Price, T.R., 1984. Mood disorders in stroke patients. Imporance of location of lesion. Brain 107 (Pt 1), 8193. Robinson, R.G., Price, T.R., 1982. Post-stroke depressive disorders: a follow-up study of 103 patients. Stroke 13 (5), 635641. Robinson, R.G., Szetela, B., 1981. Mood change following left hemispheric brain injury. Annals of Neurology 9 (5), 447453.
102
Rodgers, B., Mann, S.L., 1993. Re-thinking the analysis of intergenerational social mobility: a comment on John W. Foxs Social class, mental illness, and social mobility. Journal of Health and Social Behavior 34 (2), 165172 (Discussion 173167). Rushing, W.A., Ortega, S.T., 1979. Socioeconomic status and mental disorder: new evidence and a sociomedical formulation. AJS 84 (5), 11751200. Saylor, C.F., Finch Jr., A.J., Spirito, A., Bennett, B., 1984. The childrens depression inventory: a systematic evaluation of psychometric properties. Journal of Consulting and Clinical Psychology 52 (6), 955967. Schaffer, C.E., Davidson, R.J., Saron, C., 1983. Frontal and parietal electroencephalogram asymmetry in depressed and nondepressed subjects. Biological Psychiatry 18 (7), 753762. Sher, K.J., Trull, T.J., 1996. Methodological issues in psychopathology research. Annual Review of Psychology 47, 371400. Shimoda, K., Robinson, R.G., 1999. The relationship between poststroke depression and lesion location in long-term follow-up. Biological Psychiatry 45 (2), 187192. Smucker, M.R., Craighead, W.E., Craighead, L.W., Green, B.J., 1986. Normative and reliability data for the childrens depression inventory. Journal of Abnormal Child Psychology 14 (1), 2539. Spitzer, R.L., Williams, J.B.W., Gibbon, M., Frist, M.B., 1990. Users guide for the Structured Clinical Interview for DSM-III-R. American Psychiatric Press, Washington, DC. Srole, L., 1962. Mental health in the metropolis: the midtown Manhattan study. Blakiston Division, McGraw-Hill, New York. Starkstein, S.E., Robinson, R.G., Price, T.R., 1987. Comparison of cortical and subcortical lesions in the production of poststroke mood disorders. Brain 110 (Pt 4), 10451059. Tomarken, A.J., Davidson, R.J., Henriques, J.B., 1990. Resting frontal brain asymmetry predicts affective responses to lms. Journal of Personality and Social Psychology 59 (4), 791801. Tomarken, A.J., Davidson, R.J., Wheeler, R.E., Doss, R.C., 1992. Individual differences in anterior brain asymmetry and fundamental dimensions of emotion. Journal of Personality and Social Psychology 62 (4), 676687. Tomarken, A.J., Davidson, R.J., Wheeler, R.E., Kinney, L., 1992. Psychometric properties of resting anterior EEG asymmetry: temporal stability and internal consistency. Psychophysiology 29 (5), 576592. Tomarken, A.J., Keener, A.D., 1998. Frontal brain asymmetry and depression: a self-regulatory perspective. Cognition and Emotion 12 (3), 387420. Tomarken, A.J., Waller, N.G., 2003. Potential problems with well tting models. Journal of Abnormal Psychology 112, 578598. Tukey, J.W., 1954. Causation, regression, and path analysis. In: Kempthorne, O.K., Bancroft, T.A., Gowen, J.W., Lush, J.L. (Eds.), Statistics and Math Biology. Iowa State University Press, Ames, pp. 3566. Warner, V., Weissman, M.M., Fendrich, M., Wickramaratne, P., Moreau, D., 1992. The course of major depression in the offspring of depressed parents. Incidence, recurrence, and recovery. Archives of General Psychiatry 49 (10), 795801. Weissman, M.M., Bruce, M.L., Leaf, P.J., Florio, L.P., Holzer, C., Jr., 1991. Affective disorders. In: Regier, D.A. (Ed.), Psychiatric Disorders in America : The Epidemiologic Catchment Area Study. Free Press, New York. Weissman, M.M., Fendrich, M., Warner, V., Wickramaratne, P., 1992. Incidence of psychiatric disorder in offspring at high and low risk for depression. Journal of the American Academy of Child Adolescent Psychiatry 31 (4), 640648. Weissman, M.M., Myers, J.K., 1978. Affective disorders in a US urban community: the use of research diagnostic criteria in an epidemiological survey. Archives of General Psychiatry 35 (11), 13041311. Weissman, M.M., Warner, V., Wickramaratne, P., Moreau, D., Olfson, M., 1997. Offspring of depressed parents. 10 Years later. Archives of General Psychiatry 54 (10), 932940. Westfall, P.H., Tobias, R.D., Rom, D., Wolnger, R.D., Hochberg, Y., 1999. Multiple Comparisons and Multiple Tests Using the SAS System. SAS Publishing, Cary, NC. Westfall, P.H., Young, S.S., 1992. Resampling-Based Multiple Testing: Examples and Methods for P-Value Adjustment (Wiley Series in Probability and Mathematical Statistics. Applied Pro) Wiley/Interscience. Williams, D.R., 1990. Socioeconomic differentials in health: a review and redirection. Social Psychology Quarterly 53 (2), 8199.
Patterns of brain electrical activity in infants of depressed mothers who breastfeed and bottle feed: the mediating role of infant temperament
Nancy Aaron Jones , Barbara A. McFall, Miguel A. Diego
Florida Atlantic University at Jupiter, 5353 Parkside Drive, Jupiter, FL 33458, USA
Abstract Successful breastfeeding involves a dyadic interaction between a mother and her infant. The present study was designed to examine the association between breastfeeding and temperament in infants of depressed mothers. Seventy-eight mothers, 31 who were depressed, and their infants participated. Depressed mothers who had stable breastfeeding patterns were less likely to have infants with highly reactive temperaments. Multivariate analyses of variances (MANOVAs) showed that infants of depressed mothers who breastfed did not show the frontal asymmetry patterns, i.e., left frontal hypoactivity, previously reported. Moreover, breastfeeding stability, even in depressed mothers, was related to more positive dyadic interactions. Finally, a model was supported, in which the effects of maternal depression on infant feeding are mediated by infant frontal EEG asymmetry and infant temperament. These ndings could provide a foundation for developing intervention techniques, employing breastfeeding promotion and support, directed toward attenuating the affective and physiological dysregulation already noted in infants of depressed mothers. 2004 Elsevier B.V. All rights reserved.
Keywords: Frontal EEG asymmetry; Emotion; Breastfeeding; Temperament; Infancy
1. Introduction In 1997, the American Academy of Pediatrics recommended that women breastfeed their infants for the rst year of life (Gartner, 1998) increasing interest and rates of breastfeeding. However, a substantial portion of women, especially depressed women, do not choose to
Corresponding author. Tel.: +1-561-799-8632; fax: +1-561-799-8535. E-mail address: njones@fau.edu (N.A. Jones).
104
N.A. Jones et al. / Biological Psychology 67 (2004) 103124
breastfeed or do so for relatively short periods of time (Field et al., 2002a; Galler et al., 1999). Investigators are now compelled to examine the factors that are associated with the benets of sustained breastfeeding and to examine whether breastfeeding can benet the physiological and affective development of infants. Moreover, it is important to investigate whether breastfeeding has positive or negative effects on the affective development of infants of depressed mothers. Previous research has consistently demonstrated that infants of depressed mothers are at increased risk for physiological and affective dysregulation (Field, 1995; Jones et al., 1998) necessitating further inquiry into the benets of breastfeeding for these infants. In this study, we attempted to examine whether the dysregulated affective and physiological development, noted in infants of depressed mothers, was also evident in infants of depressed mothers with sustained breastfeeding patterns. Little is know about patterns of breastfeeding in depressed mothers, and it is likely that what may be true of other adults who breastfeed does not hold for breastfeeding depressed mothers. Depressed mothers have been less likely than other adults to choose to breastfeed (Galler et al., 1999; Milligan et al., 1990), making it unclear whether breastfeeding can benet depressed mothers and their infants. Until recently, inquiry into the benets of breastfeeding to infants of depressed mothers was largely precluded by a scarcity of research participants since breastfeeding has been more prevalent among psychologically non-symptomatic, college-educated, middle-class women whose infants are at low risk for adverse outcomes. However, due to the resurgence of cultural acceptance of breastfeeding and the enhanced understanding of the physiological importance (Newman, 1995), depressed mothers are now choosing to breastfeed. Besides the wealth of research that has touted the physiological benets of breastfeeding on infant health, a number of studies have also assessed the emotional benets of breastfed compared to bottle fed motherinfant pairs. In parallel with earlier ndings highlighting greater reciprocity and affection during breastfeeding (Bernal and Richards, 1970; Dunn and Richards, 1977), recent research has also demonstrated (Lavelli and Poli, 1998) that breastfeeding mothers provide their neonates with less auditory stimulation, but more visual gaze and tactile stimulation. In addition, investigations probing whether the experiences entailed by breastfeeding extend beyond the feeding context have been addressed only rarely. One study compared dyadic interactions of breastfeeding versus bottle-feeding adult mothers in play situation (Kuzela et al., 1990) and reported that breastfeeding mothers touched their infants more frequently, suggesting the benets of breastfeeding extend to other contexts. Previous research has typically focused on the characteristics of the mother and her likelihood for continued breastfeeding (Mezzacappa and Katkin, 2002; Mezzacappa et al., 2002; Pugh, 1998) while relative few studies have examined the effects of infant characteristics on breastfeeding. Of those studies that have measured infant behavior, researchers have reported that breastfed newborns are more irritable but show more optimal physiological organization (Dipietro et al., 1987; Zeskind et al., 1992). However, later in infancy, breastfed infants are reported to be more active, temperamentally easier, less irritable, more positive, and more sociable (Field et al., 2002a; VanDiver, 1997; Worobey, 1992, 1998). These data suggest that breastfeeding may be benecial to the infant, yet the newborn may not manifest overt behavioral signs that promote continued breastfeeding. This issue is even more important for depressed mothers because these mothers are less likely to feel
105
competent in caring for their infants (Field et al., 2002a). Therefore, if their newborn responds with elevated distress to breastfeeding, depressed mothers are more likely to discontinue breastfeeding. Examination of the factors that promote breastfeeding and those that act as barriers to sustained breastfeeding should also focus on the role of infant temperament. We adhere to a denition of temperament that involves behavioral reactivity and regulation as a function of tonic biological and/or physiological patterns. Moreover, we contend that we are able to understand variations in temperament by analyzing the manner in which emotions are expressed and regulated during infancy (Calkins et al., 1996; Fox, 1994; Fox et al., 1994; Goldsmith et al., 1987). Our own studies, and those of others (Dawson et al., 1997), have suggested that newborns and infants of depressed mothers demonstrate greater dysregulated affective and physiological patterns (Jones et al., 1997a, 1998; Lundy et al., 1999). Findings from these studies have consistently found that infants of depressed mothers show greater right frontal EEG asymmetry, lower vagal tone, more dysregulated biochemical patterns, more disorganized behavioral, and sleep patterns. Collectively, these data suggest that infant bio-behavioral regulation patterns, associated with maternal depression, may lead to a negative temperamental bias that is likely to inuence later development and may predispose the child for later affective disorders (Field, 1995). Early in infancy, parental behaviors are primarily focused on care taking. Opportunities for breastfeeding benets are likely warranted to ameliorate the negative behavioral and physiological patterns already identied in infants of depressed mothers. However, we know of no other studies that have examined infants of depressed mothers and the potentially positive outcomes that may be garnered by exposing them to stable breastfeeding patterns during infancy. Within the present study, we examined whether stable breastfeeding patterns in depressed mothers were associated with attenuated negative developmental patterns in their infants. We also examined additional factors that may be related to the positive outcomes, whether affective bonding was increased as a result of stable breastfeeding, whether infants were physiologically beneted by breastfeeding, and whether mothers perceptions of their infants temperament and their behaviors were associated with feeding stability and depression status. The purpose of the present study was to examine breastfeeding in infants of depressed mothers. Specically, we examined individual differences in infant temperament (positive and negative) that may be related to stable breastfeeding patterns in depressed mothers, including infants physiological and behavioral reactivity, and the dyads socio-emotional interaction patterns. Our goal was to determine whether each factor was associated with breastfeeding stability and whether stable breastfeeding in depressed dyads was associated with attenuated negative outcomes. We hypothesized that maternal depression and more negatively reactive infant temperament would be associated with a shorter duration of breastfeeding. Second, we expected that depressed mothers who discontinued breastfeeding by 3 months of age (or never breastfed) would have infants who show the greatest dysregulation, in affective regulation and physiological activity. Third, we expected that depressed mothers who continued breastfeeding would report more positive perceptions of their infants temperament and would demonstrate greater mutual interactive behaviors with their infants. Finally, we tested a model that examined the direct and indirect effects of maternal depression on infant temperament, infant EEG activity, and feeding behaviors.
106
Infant EEG asymmetry
Infant Temperament
(positive responsive-ness during interaction)
Maternal Depression Score

Fig. 1. Hypothesized model.
Duration Breastfeeding
We hypothesized that the effects of maternal depression on feeding patterns are mediated by infant frontal EEG activity and infant temperament (Fig. 1).
2. Method 2.1. Prescreening Initially, hospital intake information (documenting demographic, feeding, and infant health information) was used to obtain a sample of new mothers who t the predened criteria. For the present study, we concentrated our efforts on obtaining middle class (mean of 23 on Hollingshead scale (Hollingshead, 1975), adult (age: 2039 years) women with healthy (>9 on 5 min APGAR), full-term (>37 weeks gestation) infants. We choose to recruit a middle-class, healthy infant sample to rule out other possible confounds (Buxton et al., 1991; Cooper et al., 1993). In order to expedite recruitment of depressed participants, only mothers who received high scores (>16) or low scores (<12) on the Center for Epidemiological Studies Depression Scale (CES-D; Radloff, 1977) were asked to participate in laboratory assessments. Mothers who were eligible at the newborn period were interviewed over the phone at 1 month on the CES-D to conrm their depression status and to schedule a laboratory session. The mothers and infants who participated in this study were drawn from 232 hospital and telephone questionnaires conducted. 2.2. Participants Seventy-eight mothers and their infants participated in the laboratory assessments at 1 month and 62 (77.5%) returned for the 3-month visit. Families were predominantly White (four African Americans, two Hispanic and one Asian American), middle to upper-middle class (Hollingshead mean = 2.32, S.D. = 0.80, range = 13). Mothers were adult (mean age = 32.20 years, S.D. = 5.04 years) and married (93.5%), with one to three children. Approximately half of the mothers were employed outside the home (44.2%). Infants (39 females and 39 males) were assessed at 1 month (mean = 35.99 days, S.D. = 5.70 days)
N.A. Jones et al. / Biological Psychology 67 (2004) 103124 Table 1 Mother and infant demographics Measure Depressed Breastfeedinga Infant variables Female (%) Age (days) One month Three months Maternal variables Age (years) SES (Hollingshead) Marital status (percentage of married) Parity (percentage of multiparous) CES-D One month Three months 26.7 36.80 (5.87) 93.78 (8.32) 33.60 (5.99) 2.67 (0.79) 93.3 66.7 Bottle 68.8 32.62 (4.63) 94.00 (2.24) 31.06 (5.43) 2.43 (0.81) 93.8 75.0 feedingb Non-depressed Breastfeedingc 51.7 36.69 (6.23) 91.87(10.64) 32.37 (4.97) 2.14 (0.84) 96.4 66.3
107
Bottle feedingd 55.6 36.83 (4.82) 95.30(11.05) 31.78 (7.78) 2.55 (0.70) 94.4 66.6 6.50 (3.45) 3.73 (3.39)
19.00 (2.53) 7.50 (4.32)
19.27 (6.11) 16.90 (6.48)
5.07 (3.28) 4.54 (3.44)
Note. Values represent mean scores. Standard deviations are in the parentheses. a n = 15. b n = 16. c n = 29. d n = 18. P < 0.001.
and at 3 months (28 females and 34 males, mean = 92.66 days, S.D. = 9.19 days) of age. There were no differences between the depression and feeding groups on their demographic characteristics (Table 1). 2.3. Measures The CES-D is a 20-item questionnaire used to assess depressive symptoms in a community sample. Scores range from 0 to 60 and a score of greater than 16 has been shown to differentiate clinical depression from non-depressed status. The CES-D has a 6.1% false-positive rate and a 36% false-negative rate (Myers and Weissman, 1980), however, we obtained additional information on maternal depression status from the Diagnostic Interview Schedule (DIS; Robins et al., 1981). The DIS was used to screen out participants who were mildly depressed. Extensive testing has shown that the DIS has good psychometric properties (Costello et al., 1984, 1985) and it can be used by lay persons to assess depressive symptoms. Participants who received high scores on the CES-D and who had elevated symptoms (consistent with dysthymia or major depressive disorder) on the DIS were assigned to the depressed group. Approximately 20 mothers (22% of those contacted) had elevated CES-D scores but were not assigned a diagnosis on the DIS. These mothers were not included in the data analyses for the present study. Non-depressed mothers were chosen based on their low CES-D scores (<12) and a negative personal and family history of depression.
108
Feeding status was also assessed at the newborn, 1- and 3-month assessments. Mothers were queried about their plans to use a specic feeding technique (and for how long), their current manner of feeding their infant, and their rational for using that feeding technique. 2.4. Laboratory procedures at 1 and 3 months Upon arrival at the lab, mothers signed a consent form, and completed a demographic and feeding inventory, a CES-D, the DIS (if appropriate), and an infant temperament inventory (Infant Behavior Questionnaire (IBQ); Rothbart, 1981). During this time, a research assistant administered the Brazelton Neonatal Behavioral Assessment Scale (BNBAS; Brazelton and Nugent, 1995) to examine the infants behavioral abilities during environmental interaction. The Brazelton examiners were trained to 0.90 reliability. Seven infants did not complete the entire Brazelton assessment. Following this assessment, mothers were asked to play with her infant as she would at home. This face-to-face interaction lasted for 3 min and both mothers and infants were videotaped during this procedure. Baseline EEG activity was recorded from infant participants during a quiet, alert state. The EEG activity was recorded while the infant was held by his/her mother or sitting in an infant seat. EEG recordings were for 56 min in duration. Five infants (three at 1 month and two at 3 months) would not tolerate the cap placement. In addition, ve infants at 1 month and four infants at 3 months had unusable data. Thus, a nal sample of 70 infants had 1 month EEG data and 53 infants had EEG data at both 1 and 3 months. Identical assessments were obtained at the 3-month visit, except the Infant Neurological International Battery (Innib; Ellison et al., 1985) was used as the developmental assessment and an arm restraint procedure was performed at 3 months. The Innib is designed to assess the neurological integrity of infants based on 20 items with 5 factors (spasticity, vestibular function, head and trunk ability, French angles, and leg function). This test is easy to administer and has sufcient reliability for clinical and research purposes (Ellison et al., 1985). For this study, the total score was computed and then standardized. This score used to verify normal neurological function across groups. All groups were similar in their scores (P > 0.05) and all infants were in above the above the abnormal cut-off score of 48. The arm restraint task is modeled after the one described by Stifter and Braungart (1995) and Stifter and Jain (1996). Stifter and her colleagues have used this task to examine temperamental reactivity and the infants ability to regulate during a mildly frustrating task. Infants are seated in an infant seat. Mothers sit in a chair facing her infant. Mothers are instructed restrain their infants arms by the sides of their body for approximately 2 min. The arm restraint is discontinued if the infant was distressed for 20 s. Mothers are also instructed to maintain a neutral facial expression (or to look away from their infant if they cannot remain neutral) and to remain silent during the procedure. Afterward the mother is instructed to hold or sooth her infant. Three mothers and their infants did not complete this task during the lab session. 2.5. Coding of mother and infant behaviors Video recordings of 3 min, motherinfant interactions were coded on a second-by-second basis. We modeled our coding schemes for the play interaction after ones described by
109
Kuzela et al. (1990). During the play interaction, both mothers and infants, were coded separately for affect, attention, and vocalization. The scales were scored on a seven-point graduated scale, with lower scores indicating more non-optimal affect, attention, and vocalizations and higher scores indicating more optimal affect, attention, and vocalizations. Percentage of time within each scaled score was weighted and summed to obtain an index of the infants and the mothers interactive behaviors, with higher scores indicating more positive facial affect, more directed attention, and more positive vocalizations. An index of motherinfant mutuality was also coded on a seven-point graded scale and coded on a second-to-second basis. The purpose of this coding was to obtain measures of mutually responsive play behaviors, employing similar coding systems described by Field et al. (1990). A low score on this scale indicated fewer and more negative mutually responsive behaviors and a high score indicated more optimal/positive interactive behaviors. The total percentage of time spent in mutual gaze and affective state (combining facial and vocal expressions) was obtained by observing the mother and infant interactive behaviors simultaneously. Descriptions of the behaviors used to index the behaviors during the play and mutuality coding can be found in Table 2. Finally, infant behavioral responses during the arm restraint were coded to assess infant reactivity using similar scales as those described by Stifter and Braungart (1995) and Stifter and Jain (1996). This task was also scored on second-by-second basis using scales to measure facial affect, attention and vocalization. Three, ve-point scales, with low scores associated with low behavioral responses and high scores indicating maximum behavioral responses, were coded. Percentage scores were used to compute an index of negative reactivity, using higher scores on vocal and facial affect. Using descriptive statistics, we separated the infants into low, middle, and high reactive groups, using the equal percentages of 33.33% to form the groups. Descriptions of the behaviors coded for this task are also found in Table 2. 2.6. Physiological recordings at 1 and 3 months A stretch lycra cap (Electro Cap Inc.) with the international 1020 system was positioned on the infants head to obtain a measure of EEG activity at each age. Electrode gel, used to conduct, and Omni Prep gel, used to gently abrade, was inserted into eight sites. The mid-frontal (F3 and F4), central (C3 and C4), parietal (P3 and P4), and occipital (O1 and O2) sites were chosen in order to remain consistent with previous research in this area (Jones et al., 1997a). Each site was referenced to the vertex (Cz). Although there has been controversy concerning the appropriate reference location, the vertex reference was used for three reasons. First in one study, this site has been shown to be comparable to other reference sites (Tomarken et al., 1992). Although other studies question the use of this reference with adult participants (Hagemann et al., 1998; Reid et al., 1998) no study has resolved this issue with infant participants (Pivik et al., 1993). Second recordings, using this site as a reference for infant participants, are more feasible given the ease of preparing the Cz site and the possible attrition of infant participants when employing more invasive and/or more numerous recording sites (Pivik et al., 1993). Third, the literature on infant EEG activity has used the vertex as the reference site almost exclusively (Field et al., 2002b; Fox et al., 1992; Jones et al., 1998). Thus, we choose to
110
Table 2 Mother and infant codes, descriptors, and inter-rater reliability Coded behavior Description
Mother and infant affect, attention and vocalization ( = 0.86 for infant and 0.89 for mother) 7 Participant shows strong interest, smiling, positive intonation of vocalizations, and positive touch 6 Participant moderate shows interest, pleasant positive attitude, positive intonation of vocalizations, and positive touch 5 Participant shows some interest but no variability in intonation of vocalizations, variable stimulation, little touch or heightened physical activity or physical contact. Participant is reactive but with little facial affect 4 Participant shows varied positive, no, and/or negative affect, no consistency in responsiveness or attention. Mothers touch is for grooming rather than stimulating. Infant is non-reactive or neutral 3 Mother speaks to child but her face and voice are at in affect, directs conversation without being responsive. Infant displays escape motor activities and may attend but briey, occasionally frets 2 Mother losing interest, shows some self-interest (i.e., self-grooming). Infant is not attentive, vocalizations are fussy in tone. Infant may achieve full cry temporarily but reverts back to a fret or neutral affect 1 Mother shows varied interest in infant, possibly negative facial expressions or vocalizations. Infant is crying or is negatively responsive Motherinfant mutuality ( = 0.91) 7 Positive mutuality: mother and child attending. Mother attending and allowing for infant response, positive intonation in vocalization in both mother and infant smiling, mother possibly singing 6 Intermittent positive mutuality: mother and child attending, varied intonation of vocalization, infants eyes are attending but smiling is varied 5 Varied mutuality: mother attends most of the time with varied positive and some neutral affect. Infant attends more than half of the time but with neutral affect 4 Neutral mutuality: mother and child attend but with little or no affect by either. Some grooming behaviors may be present but little stimulus/response interaction 3 Mother attending/infant not responsive or negatively responsive: mother stimulating but infant shows little interest. Mother possibly over-stimulating, infant my show negative response, infant clearly not enjoying the interaction and mother not changing her behavior in response 2 Mother attending/infant not attending or escaping: mother attends but no response from infant. Infant showing intermittent negative vocalization or full cry. Infant is showing signs that clearly indicate distress at interacting and mother not responding to infant distress 1 No mutuality: neither attending to each other nor interacting negatively Infant reactivity during arm restraint ( = 0.92) 5 Infant in full, continuous cry and twisting, stretching and kicking is observed 4 Infant showing no positive vocalizations, steady cry, with escape-type behaviors 3 Infant showing escalated fussing, may have brief neutral tone but infant reverts to fretting 2 Infant showing brief fussing and facial negativity (frown, pout) 1 No negative vocalizations or facial expressions are observed
use the vertex as the reference site in order to compare our results to similar ndings in the literature. Electrode impedances were brought down to less than 5 k or the sites were re-abraded. EOG was also be obtained, to aid artifact scoring, using two mini-electrodes, one at the
111
outer canthus and one at the supra orbit position of one eye. The electrical signal was amplied at each site using SA Instrumentation Bioamps and bandpassed from 1 to 100 Hz. EEG activity from each electrode lead was displayed on a computer acquisition monitor. The EEG was digitized on-line at a rate of 512 samples/s and saved to the computer hard disk using data acquisition software (Snapstream, v. 3.21, HEM Data Corp., 1991). EEG data were examined and scored for eye and motor movement artifact using the EOG channel as cues. Using software developed by James Long Inc., artifact was simultaneously eliminated from all channels. Then data were then submitted to a discrete Fourier transform using a Hanning window with 50% overlap. This analysis produced power for the specied frequency band in pW (1 V squared) for each channel. Previous research suggested that the commonly used (adult) frequency bands are not appropriate for infant EEG activity analyses (Bell, 2002). Moreover, research has shown that there is a shift in the alpha frequency band across development (Matousek and Petersen, 1973). EEG were analyzed in two ways: (1) using single hertz frequency bands in order to examine the spectral characteristics of the data; and (2) using the frequency bands commonly used for infants (36 Hz for 1-month olds and 69 Hz for the 3-month olds; Jones et al., 1997a, 1998). To normalize the distribution, power scores for each region were submitted to a natural log transformation (ln). Finally, EEG asymmetry scores were computed to determine the relative contribution of each hemisphere (ln(right) ln(left)), with negative scores reecting greater relative right hemisphere EEG activity and positive scores reecting greater relative left hemisphere EEG activity. 2.7. Data reduction and analyses Chi-square analyses were used to examine the incidence of stable feeding patterns across development for infants of depressed and non-depressed mothers. In addition, infants were assigned to a highly reactive, mid-level reactive, and low reactive groups based on their responses to the 3-month arm restraint procedure. Infants who were in the upper 33.33% of negative facial and vocal responses were classied as high reactive whereas infants in the lower 33.33% of facial and vocal responses were classied as low reactive. To examine the association between mother and infant behaviors and physiology related to depression status and feeding patterns, four groups were examined. These groups were composed of the following: (1) depressed mothers with stable breastfeeding patterns from newborn to 3 months of age; (2) depressed mothers with little breastfeeding and mostly bottle feeding from newborn to 3 months; (3) non-depressed mothers with stable breastfeed patterns across age; and (4) non-depressed mothers with little breastfeeding and mostly bottle feeding across age. Infant data were examined using multivariate analyses of variances (MANOVAs) for EEG activity and dyad data were examined for their behavioral responses during interactions (positive, neutral, negative, and mutual/synchronous responses). Follow-up analyses were conducted using separate univariate ANOVAs for depressed versus non-depressed groups and/or for stable breastfeeding versus bottle-feeding (and non-stable breastfeeding) groups.
112
3. Results 3.1. Maternal depression and infant temperament In an attempt to determine whether reactive infant temperament and maternal depression were related to the earlier cessation of breastfeeding, we conducted several chi-square analyses. Infants classied as high in reactive during the arm restraint procedure were compared to those low in reactive on the stability of their breastfeeding patterns to 3 months of age. Results showed that 77.8% of the infants who were high reactive and had a depressed mother were predominately bottle fed whereas only 22.2% were breastfed (2 (3, N = 59) = 13.45, P < 0.05). Moreover, 70% of infants of non-depressed mothers who were reactive were stable in their breastfeeding whereas only 30% of these infants were bottle fed, suggesting that infant temperament is associated with breastfeeding stability in infants of depressed mothers. These results were not found when examining infants with low or mid-level reactive groups. Feeding patterns were related to depressed mothers plans to breastfeed at the newborn period and their actual feeding duration during the rst 3 months. Depressed mothers planned to breastfed for shorter periods of time, with 94.2% planning to breastfed for less than 6 months than non-depressed mothers, 27.6% of whom were planning to breastfed for less than 6 months (2 (3, N = 78) = 45.92, P < 0.05). Further, a one-way ANOVA comparing depressed to non-depressed mothers on their feeding duration during the course of the study yielded a signicant effect (F(1, 76) = 5.68, P < 0.05), with depressed mothers breastfeeding their infants a shorter duration of time (mean = 43.81 days, S.D. = 38.20 days) compared to non-depressed mothers (mean = 64.13 days, S.D. = 35.95 days). 3.2. EEG patterns for infants of depressed who breastfeed A group (four levels; depressed/breastfed, depressed/bottle fed, non-depressed/breastfed, non-depressed/bottle fed) region (four levels; frontal, central, parietal, occipital) age (two levels; 1 and 3 months) repeated-measures MANOVA was conducted using the infants EEG asymmetry score as the dependent variable. Results yielded a signicant three-way interaction (F(9, 147) = 2.06, P < 0.05). Analyses of each region separately showed group differences in frontal EEG asymmetry (F(3, 49) = 3.14, P < 0.05) across age but no differences in central, parietal and occipital regions (all P > 0.05) (Fig. 2). Univariate ANOVAs showed that frontal EEG asymmetry differed between groups at 1 month of age (F(3, 66) = 6.48, P < 0.05) and were only marginally signicant at 3 months (F(3, 55) = 2.90, P = 0.09). In addition, an overall group (four levels) region (four levels) hemisphere (two levels; right versus left) age (two levels) MANOVA was conducted using the infants ln-transformed EEG power values as the dependent variable. Results yielded several multivariate main effects and interactions, with an overall signicant four-way interaction (F(9, 147) = 2.34, P < 0.05). In order to interpret the interactions and due to the specic hypotheses about frontal region EEG activity, we conned our subsequent analyses to examining frontal EEG activity.
113
Fig. 2. EEG asymmetries.
A subsequent group hemisphere age analysis for the infants frontal region, lntransformed power scores yielded a signicant main effect for group (F(3, 49) = 2.88, P < 0.05) and a signicant group hemisphere interaction (F(3, 49) = 3.26, P < 0.05), with less left hemisphere activity in the frontal region in the depressed bottle-feeding group than in the other three groups, post-hoc signicance (P = 0.04) (Fig. 3). The depressed/breastfeeding group and the non-depressed groups did not show differences in hemispheric EEG activity in the frontal region (P > 0.05). 3.3. Infant developmental assessments Infant neurobehavioral responses were assessed on the Brazelton exam at 1 month and on the Innib at 3 months of age. Multivariate analyses were conducted to determine whether Brazelton scores differed for feeding and depression groups. Results showed a multivariate interaction for Brazelton items between groups (F(21, 189) = 3.60, P < 0.05). Follow-up univariate analyses between groups showed differences between groups for habituation (F(3, 67) = 7.66, P < 0.05), orientation (F(3, 67) = 2.70, P < 0.05), motor maturity (F(3, 67) = 3.98, P < 0.05), regulation of state (F(3, 67) = 2.82, P < 0.05), and abnormal reexes (F(3, 67) = 6.32, P < 0.05). Mean scores for each measure can be found in Table 3. Overall, results showed early benets for infants of non-depressed mothers and
114
Fig. 3. EEG power (ln-transformed) scores for the frontal region.
infants of depressed mothers who were breastfed, with this latter group exhibiting increased habituation abilities and decreased abnormal reexes. An ANOVA comparing scores on the Innib for depression and feeding groups did not reach signicance, with both all groups showing similar developmental capabilities at 3 months of age (P > 0.05).
Table 3 Infant Brazelton scores Measure Depressed Breastfeedinga Habituation Orientation Motor maturity Range of state Regulation of state Autonomic stability Abnormal reexes 7.73 (0.57) 6.44 (1.06)+ 5.60 (0.31) 3.07 (1.18) 5.45 (1.24) 6.19 (1.17) 1.00 (1.04) Bottle feedingb Non-Depressed Breastfeedingc 7.76 (0.30) 7.17 (0.58) 5.44 (0.70) 3.38 (1.16) 6.38 (1.38) 6.70 (0.96) 0.92 (1.15) Bottle feedingd 7.10 (0.74) 6.78 (0.74) 5.61 (0.40) 3.34 (0.96) 6.20 (0.78) 6.57 (1.18) 1.00 (0.84)
6.76 (1.20)+ 6.69 (0.93) 6.07 (0.66)+ 3.86 (0.82) 5.58 (0.90) 6.29 (1.01) 2.25 (1.06)+
Note. Values represent mean scores. Standard deviations are in the parentheses. Different superscripts denote signicantly different post-hoc comparisons, P < 0.05. a n = 12. b n = 16. c n = 25. d n = 18. P < 0.05.
115
3.4. Maternal reports of infant temperament MANOVAs were conducted on the mothers reports of positive and negative (activity level, smiling and laughter, distress latency, distress to limitations, soothability and duration of orienting) behaviors on the six scales of the Infant Temperament Questionnaire (IBQ) for the four groups. Results showed that all mothers reported differences in infant temperament on the six scales (F(5, 53) = 86.50, P < 0.05), with more reports of positive behaviors and fewer reports of negative behaviors. In addition, maternal reports of infant behaviors on the temperament scales increased across age (F(1, 57) = 116.52, P < 0.05). However, no meaningful group differences in the types of behaviors reported by depressed and non-depressed mothers nor for breastfeeding and bottle-feeding groups were apparent from ratings of maternal perceived infant temperament. 3.5. Mother and infant behaviors during play A group affect (three levels; positive, neutral, negative) age MANOVA showed that infants were exhibiting more positive and less negative and neutral interactive behaviors at 3 months than at 1 month (F(2, 110) = 13.88, P < 0.05) (Fig. 4 illustrates the data). Follow-up univariate ANOVAs revealed that the infants of depressed, bottle-feeding mothers showed more negative affect at 1 month (F(3, 72) = 2.80, P < 0.05) and these infants also showed the least positive affect at 3 months (F(3, 55) = 3.22, P < 0.05), than the other three groups. A repeated-measures group age MANOVA was conducted on the percentage of time in mutually responsive interactions as the dependent variables. Result showed no signicant interaction effects. However, a signicant main effect for age was evident (F(1, 55) = 27.80, P < 0.05). As expected, the dyads demonstrated more mutually interactive behaviors at 3 months than at 1 month. Examination of mutually interactive behaviors separately for the maternal depression groups versus the feeding groups were conducted at 3 months of age. This univariate ANOVA yielded a signicant between-subjects effect for feeding group (F(1, 57) = 4.24, P < 0.05), with mean scores indicating that dyads who did not have stable breastfeeding patterns had less mutually interactive behaviors at the 3-month visit (mean = 61.23, S.D. = 7.04) than infants who had stable breastfeeding patterns (mean = 66.14, S.D. = 7.04). 3.6. Path analysis Finally, a path analysis was conducted using EQS (EQS v. 5.7b, Multivariate Software, Inc, 1998) in order to evaluate the role of infant frontal EEG asymmetry and temperament in mediating maternal depression effects on feeding behaviors. Missing data for nine subjects were replaced with mean values. Based on previous ndings, we hypothesized that maternal depression would negatively affect feeding behavior. Similarly, we hypothesized that infant temperament would be related to feeding behavior. Furthermore, we predicted that maternal depression would be related to infant temperament via its effects on the development of the underlying structures of emotion and emotion regulation, reected by frontal EEG asymmetry patterns. As such the hypothesized model examined the direct and
116 N.A. Jones et al. / Biological Psychology 67 (2004) 103124
Fig. 4. Infant behavioral responses during motherinfant interactions at 1 and 3 months.
N.A. Jones et al. / Biological Psychology 67 (2004) 103124 Table 4 Correlations Infant feeding Depression score Frontal EEG asymmetry
117
Infant temperament (positive during interaction) 78 149.491 0.366 0.001 78 22.102 0.302 0.007 78 0.801 0.352 0.002 78 102.093 1
Infant feeding N Covariance Pearson correlation Signicance (two-tailed) Depression score N Covariance Pearson correlation Signicance (two-tailed) Frontal EEG asymmetry N Covariance Pearson correlation Signicance (two-tailed)
78 1637.722 1
78 64.786 0.221 0.052 78 52.529 1
78 3.074 0.338 0.003 78 0.487 0.299 0.008 78 0.051 1
78 64.786 0.221 0.052 78 3.074 0.338 0.003
78 0.487 0.299 0.008 78 22.102 .302 0.007
Infant temperament (positive during interaction) N 78 Covariance 149.491 Pearson correlation 0.366 Signicance (two-tailed) 0.001 Correlation is signicant at the 0.01 level (two-tailed).
78 0.801 0.352 0.002
indirect effects of maternal depression on both infant temperament and feeding behaviors (Fig. 1). Analyses were conducted using the maximum likelihood estimation procedure on the variance/covariance matrix (Table 4). The hypothesized model was tested revealing an adequate t for the model (2 (1, N = 78) = 3.81, P < 0.05), comparative t index (CFI) of 0.91, RMSEA of 0.19 (Fig. 5). Post-hoc modications were then performed in an
R2 =0.09

-.30*
.29*
Infant Temperament
(positive responsive-ness during interaction) R2 =0.17
-.22 .33*
Maternal Depression Score
.12
R2 =0.15
Chi Square=3.81,p=0.05, CFI=0.91, RMSEA=0.19

Fig. 5. Initial model.
118

R2 =0.09

-.24* -.30*
-.29*
Maternal Depression Score Infant Temperament
R2 =0.18
-.28*
-.22*
(positive responsiveness during interaction) R2 =0.17
Chi Square=0.46, p=0.50, CFI=1.00, RMSEA=0.00

Fig. 6. Final model.
attempt to develop a better tting model. The Wald test indicated that eliminating the path between maternal depression and feeding behaviors would improve the t of the model. A path between infant frontal EEG asymmetry and feeding behavior was added following results of the Lagrange multiplier test. The model was re-estimated revealing an excellent t (2 (1, N = 78) = 0.46, P = 0.50, CFI = 1.0, RMSEA = 0.0), suggesting an improvement from the hypothesized model (Fig. 6). This nal model suggests that the effects of maternal depression on feeding are mediated by infant frontal EEG asymmetry and temperament. Furthermore, this model suggests that maternal depression was strongly predictive of infant frontal EEG asymmetry (standardized coefcient for direct effect = 0.30, t = 2.73, P < 0.05) and infant temperament (standardized coefcient for indirect effect = 0.30, t = 2.76, P < 0.001). Furthermore, maternal depression effects on infant temperament (standardized coefcient for indirect effect = 0.09, t = 1.89, P < 0.1) were marginally mediated by frontal EEG asymmetry and the effects of frontal EEG asymmetry on feeding (standardized coefcient for indirect effect = 0.08, t = 1.83, P < 0.1) were marginally mediated by infant temperament.
4. Discussion The goal the present investigation was to examine the factors that promote breastfeeding stability and to examine the physiological and affective development in infants of depressed mothers who breastfed compared to those who bottle fed. Previous research has suggested that depressed mothers are less likely to breastfeed (Galler et al., 1999; Milligan et al., 1990) and infants of depressed mothers demonstrate dysregulated physiological and behavioral patterns as early as the newborn period (Jones et al., 1997a, 1998; Lundy et al., 1996,
119
1999). Within the present study, we documented that maternal depression is associated with less stable breastfeeding, that greater negative infant temperament is associated with less stable breastfeeding in depressed dyads, and that more positive dyadic interaction is associated with infants of non-depressed mothers and infants of depressed mothers with a stable breastfeeding relationship than for infants of depressed mothers who bottle feed. Our nal model showed that infant temperament and frontal EEG asymmetry mediated the association between maternal depression and feeding patterns. As a caveat to these results we must note that this study is correlational in nature, as we could not assign dyads to depression and feeding groups, yet these data point to potentially important future intervention studies. 4.1. Maternal depression and breastfeeding patterns That depressed mothers in this study were less likely to intend to maintain an extended duration of breastfeeding at the newborn period is discouraging given that numerous health care agencies are working to increase breastfeeding rates across infancy (Department of Health and Human Services, Blueprint for Action on Breastfeeding, 2000). However, these data are consistent with previous research showing lower rates of breastfeeding in depressed women (Field et al., 2002a; Galler et al., 1999). Typically studies have examined the characteristics of the parents and the likelihood of breastfeeding continuity across infancy. This is one of the few studies to show an association between infant temperament and breastfeeding stability across the early months of development. Specically, we showed that greater negative infant reactivity was associated with more variable feeding patterns in depressed mothers compared to non-depressed mothers. While our data are confounded, as we cannot separate situational variables (mothers plans for breastfeeding or bottle feeding and mothers depression status) and individual difference variables (infant temperament and EEG activity patterns), this study points several important future studies that may be benecial to conduct. For example, breastfeeding support and promotion could be employed, using random assignment, as a clinical intervention for depressed mothers during the prenatal period. The ndings of this study could determine more directly whether breastfeeding can benet the affective and physiological functioning of the dyad even though exposed to maternal depression. Moreover, there are many reasons that the depressed mothers may fail to establish stable breastfeeding patterns with their infants. The more simplistic explanation is that depressed mothers are more concerned about their own emotional state than the feeding status of their infants. While this explanation is possible, we are suggesting that depressed mothers may show lower rates of breastfeeding due to their lack of understanding of normative temperamental changes across infancy. As noted by previous studies, newborns who breastfed are more temperamentally irritable and more difcult to sooth (Dipietro et al., 1987). Bottle feeding, on the other hand, has a depressive effect on infant behavior (Dipietro et al., 1987). Therefore, newborn behavioral responses, as a result of breastfeeding, may confuse and tax the already depleted resources of the depressed mother. However, several studies have noted that breastfed newborns demonstrate more optimal physiological organization (e.g., Zeskind et al., 1992) and, later in development, mothers who maintain stable breastfeeding patterns report that their infants exhibit easier temperaments (VanDiver, 1997) and more socially responsive behaviors (Kuzela et al., 1990; Worobey, 1992, 1998), suggesting posi-
120
tive outcomes later in development for breastfed dyads. Therefore, depressed mothers must be educated about the numerous benets of breastfeeding stability and must be encouraged to continue breastfeeding, despite the seemingly more challenging behaviors displayed by their newborns. Moreover, these parents should be counseled that the seemingly more taxing newborn behaviors are normative and not due to inadequate parenting skills. Currently depressed mothers are discontinuing breastfeeding within the rst 2 months and therefore fail to experience the potential benets associated with the more positive infant physiological and behavioral responses that are the result of stable breastfeeding patterns. Although we did demonstrate that infant temperament is related to breastfeeding stability in the depressed group, the sample size for this study is small and is limited to the rst 3 months of development. The ndings presented here, however, remain an important area for future work, primarily due to the increasing evidence that breastfeeding stability benets overall infant health and development. Ultimately supporting mothers during the breastfeeding months will likely have long lasting benets for depressed mothers and their infants. 4.2. Physiological patterns in infants of depressed mothers who breastfeed Our study also suggests that infant EEG activity is associated with maternal depression and breastfeeding stability. The nal path model showed a strong association between maternal depression and infant EEG patterns and infant temperament. Specically, infants of depressed mothers who had been breastfed until their third month of life were less likely to show the right frontal EEG asymmetry patterns previously associated with maternal depression (Dawson et al., 1999, 1997; Jones et al., 1998) and linked to risk for depression in adults (Davidson, 1994; Henriques and Davidson, 1990). Alternatively, infants of depressed mothers who were bottle fed exhibited a bilateral decrease in frontal EEG activity that was especially pronounced on the left side. Although the difference in EEG asymmetry scores decreased with age, this may be due, in part to the declining depressive symptoms in both depression groups or it may be due to the resiliency of these infants. Nonetheless, future studies should examine the stability of EEG patterns across development and the situational factors that may attenuate these seemingly dysregulated EEG patterns in infants of depressed mothers. The EEG ndings related to feeding patterns are intriguing in light of the fact that previous investigations have only rarely assessed neurophysiological patterns associated with infant feedings. One recent study by Lehtonen et al. (2002) examined EEG activity in 3- and 6-month-old infants during feeding. While their goal was to examine EEG patterns during normative infant feeding session, they found undifferentiated theta (similar frequency band to the band we called alpha) responses to feeding and did not establish organized EEG activity patterns associated with feeding until 6 months of age. The authors speculate that the development of the frontal regions and the emotional arousal inherent in feeding was directly related to their ndings. While their ndings are later in development than our own, the ndings do parallel our associations between emotional responses and EEG activity. Of interest, they did not nd distinct patterns of EEG activity for bottle fed and breastfed groups, most likely due to the small number of bottle fed participants in their study. Collectively this study and our own ndings suggest that further research should
121
focus on feeding patterns in the infant and the association between emotional processing and frontal lobe development. 4.3. Neurobehavioral assessments of infants of depressed mothers who breastfed Consistent with previous studies, infants of depressed mothers also showed fewer optimal behaviors on the Brazelton Scales (Abrams et al., 1995; Jones et al., 1997a, 1998; Lundy et al., 1996). Specically infants of depressed mothers who bottle fed showed less habituation and more abnormal reexes. Although these results may be transitory, as these infants did not differ on developmental assessments at 3 months of age, the ndings suggest that subtle differences in early neurobehavioral abilities are affected by maternal depression and feeding status. 4.4. Motherinfant interactive behavior for depression and feeding groups In separate analyses, we also demonstrated that motherinfant interactions were less negative at 1 month of age and more positive by 3 months of age in the depressed group with stable breastfeeding patterns but not in the depressed group who were bottle fed. While previously we have established that infants of depressed mothers displayed greater negative affect during interaction (Jones et al., 1997b) and Field et al. (1990) demonstrated more negatively matched and fewer positively matched interactive patterns in depressed groups, in the present study, infants of breastfeed, depressed mothers seemed to display reduced dysregulated interaction patterns than the infants of depressed mothers who bottle fed. Although these results could be due to other factors that motivate a depressed mother to breastfeed, the fact that the depressed-breastfeeding dyads appeared to interact with their infants like the non-depressed groups suggests that breastfeeding may benet the socio-emotional interactions of depressed dyads. Breastfeeding patterns have also been examined retrospectively as a potential factor that differentiates depressed and non-depressed groups. For example, Allen and his colleagues (Allen et al., 1998) have also investigated the prenatal and perinatal risk factors for psychopathologies in children and adolescents. Their study concluded that major depression in childhood and adolescence was associated with not having been breastfed and having a mother with an affective disorder during pregnancy. Given these data and the data presented here, it may be benecial to support breastfeeding patterns in depressed mothers to possibly reduce the negative outcomes and the increased risk factors for affective problems noted in previous studies (Allen et al., 1998; Field, 1995). In conclusion, our previous research has demonstrated that infants of depressed mothers exhibit risk factors that are associated with physiological and affective dysregulation (Field, 1995; Jones et al., 1998). However, within this study we were able to demonstrate that infants of depressed mothers who are stable in their breastfeeding patterns showed a reduced association between physiological and affective dysregulation resulting from exposure to maternal depression. Specically, infants of depressed mothers who breastfed did not demonstrate the greater relative right frontal EEG asymmetry (nor the left frontal hypoactivity) compared to the bottle fed group. Moreover, increased positive affect was apparent in 3-month-old infants of breastfed compared to the bottle-feeding/depressed group,
122
suggesting that breastfeeding should be examined further as a potential intervention factor for depressed mother and their infants.
Acknowledgements We would like to thank the mothers and infants who participated in this study. We would also like to thank all the students at Florida Atlantic University at Jupiter who helped with data collection. This research was supported by a NIMH grant (MH61888) and a FAU Research Initiation Award to Nancy Aaron Jones, Ph.D.
References
Abrams, S.M., Field, T., Scadi, F., Prodromidis, M., 1995. Maternal depression effects on infants Brazelton Scale performance. Infant Mental Health Journal 16 (3), 231235. Allen, N.B., Lewinsohn, P.M., Seeley, J.R., 1998. Prenatal and perinatal inuences on risk for psychopathology in childhood and adolescence. Development and Psychopathology 10, 513529. Bell, M.A., 2002. Power changes in infant EEG frequency bands during a spatial working memory task. Psychophysiology 39, 19. Bernal, J., Richards, M., 1970. The effects of bottle and breastfeeding on infant development. Journal of Psychosomatic Research 14, 247252. Brazelton, T.B., Nugent, J.K., 1995. Neonatal Behavioral Assessment Scale, third ed. Cambridge University Press. Buxton, K.E., Carlson-Gielen, A., Faden, R.R., Hendricks-Brown, C., Paige, D.M., Chwalow, A.J., 1991. Women intending to breastfeed: predictors of early infant feeding experiences. American Journal of Preventive Medicine 7, 101106. Calkins, S.D., Fox, N.A., Marshall, T.R., 1996. Behavioral and physiological antecedents of inhibited and uninhibited behavior. Child Development 67, 523540. Cooper, P.J., Murray, L., Stein, A., 1993. Psychosocial factors associated with the early termination of breastfeeding. Journal of Psychosomatic Research 37, 171176. Costello, A.J., Edelbrock, C.S., Costello, A.J., 1985. Validity of the NIMH Diagnostic Interview Schedule for Children: a comparison between psychiatric and pediatric referrals. Journal of Abnormal Psychology 13, 579595. Costello, A.J., Edelbrock, C.S., Dulcan, M.K., Kalas, R., 1984. Testing of the NIMH Diagnostic Interview Schedule for Children (DISC) in a clinical population (contract no. DB-81-0027). Final Report to the Center for Epidemiological Studies, National Institutes for Mental Health, University of Pittsburgh, Pittsburgh. Davidson, R.J., 1994. Asymmetric brain function, affective style and psychopathology: the role of early experience and plasticity. Development and Psychopathology 6, 741758. Dawson, G., Frey, K., Panagiodies, H., Yamada, E., Hessel, D., Osterling, J., 1999. Infants of depressed mothers exhibit atypical frontal electrical brain activity during interactions with mother and with a familiar, nondepressed adult. Child Development 70, 10581066. Dawson, G., Panagiotides, H., Klinger, L.G., Spieker, S., 1997. Infants of depressed and nondepressed mothers exhibit differences in frontal brain electrical activity during the expression of negative emotions. Developmental Psychology 33, 650656. Department of Health and Human Services, Ofce on Womens Health, 2000. HHS Blueprint for Action on Breastfeeding. Washington, DC. DiPietro, J.A., Larson, S.K., Porges, S.W., 1987. Behavioral and heart rate pattern differences between breast-fed and bottle-fed neonates. Developmental Psychology 23, 467474. Dunn, J., Richards, M., 1977. Observations on the developing relationship between mother and baby in the neonatal period. In: Schaffer, R. (Ed.), Studies in MotherInfant Interaction. Academic Press, New York.
123
Ellison, P.H., Horn, J.L., Brown, C.A., 1985. Construction of an Infant Neurological International Battery (Innib) for the assessment of neurological integrity in infancy. Physical Therapy 65, 1631. Field, T., 1995. Infants of depressed mothers. Infant Behavior and Development 18, 113. Field, T., Diego, M., Hernandez-Reif, M., Schanberg, S., Kuhn, C., 2002b. Relative right versus left frontal EEG in neonates. Developmental Psychobiology 41, 147155. Field, T., Healy, B., Goldstein, S., Guthertz, M., 1990. Behavior-state matching and synchrony in motherinfant interactions of nondepressed versus depressed dyads. Developmental Psychology 26, 714. Field, T., Hernandez-Reif, M., Feijo, L., 2002a. Breastfeeding in depressed motherinfant dyads. Early Child Development and Care 172, 539545. Fox, N.A., 1994. Dynamic cerebral processes underlying emotion regulation. In: Fox, N.A. (Ed.), The Development of Emotion Regulation: Biological and Behavioral Considerations, vol. 59. Monographs of the Society for Research in Child Development no. 240. Fox, N.A., Bell, M.A., Jones, N.A., 1992. Individual differences in response to stress and cerebral asymmetry. Developmental Neuropsychology 8, 161184. Fox, N.A., Calkins, S.D., Bell, M.A., 1994. Neural plasticity and development in the rst two years of life: evidence from cognitive and socioemotional domains of research. Development and Psychopathology 6, 677696. Galler, J.R., Harrison, R.H., Biggs, M.A., Ramsey, F., Forde, V., 1999. Maternal mood predicts breastfeeding in Barbados. Journal of Developmental and Behavioral Pediatrics 20, 8087. Gartner, L.M., 1998. Letters to the editorReply. Questions about AAP breastfeeding statement. Pediatrics 102, 14951497. Goldsmith, H.H., Buss, A., Plomin, R., Rothbart, M.K., Thomas, A., Chess, S., Hinde, R., McCall, R., 1987. Roundtable: what is temperament? Four approaches. Child Development 58, 505529. Hagemann, D., Naumann, E., Becker, G., Maier, S., Bartussek, D., 1998. Frontal brain asymmetry and affective style: a conceptual replication. Psychophysiology 35, 372388. Henriques, J.B., Davidson, R.J., 1990. Regional brain electrical asymmetries discriminate between previously depressed and healthy control subjects. Journal of Abnormal Psychology 99, 2231. Hollingshead, A.B., 1975. Four Factor Index of Social Status. Yale University, New Haven, CT. Jones, N.A., Field, T., Fox, N.A., Davalos, M., Lundy, B., Hart, S., 1998. Newborns of mothers with depressive symptoms are physiologically less developed. Infant Behavior and Development 21, 537541. Jones, N.A., Field, T., Fox, N.A., Davalos, M., Malphurs, J., Carraway, K., Schanberg, S., Kuhn, C., 1997b. Infants of intrusive and withdrawn mothers. Infant Behavior and Development 20, 177189. Jones, N.A., Field, T., Fox, N.A., Lundy, B., Davalos, M., 1997a. EEG activation in one-month-old infants of depressed mothers. Development and Psychopathology 9, 491505. Kuzela, A.L., Stifter, C.A., Worobey, J., 1990. Breastfeeding and motherinfant interactions. Journal of Reproductive and Infant Psychology 8, 185194. Lavelli, M., Poli, M., 1998. Early motherinfant interaction during breast- and bottle-feeding. Infant Behavior and Development 21, 667684. Lehtonen, J., Kononen, M., Purhonen, M., Partanen, J., Saarikoski, S., 2002. The effects of feeding on the electroencephalogram in 3- and 6-month-old infants. Psychophysiology 39, 7379. Lundy, B., Field, T., Pickens, J., 1996. Infants of mothers with depressive symptoms are less expressive. Infant Behavior and Development 19, 419424. Lundy, B., Jones, N.A., Field, T., Nearing, G., Davalos, M., Pietro, P., Schanberg, S., Kuhn, C., 1999. Prenatal depression effects on neonates. Infant Behavior and Development 22, 121137. Matousek, M., Petersen, I.A., 1973. Frequency analysis of the EEG in normal children and adolescents. In: Kellaway, P., Petersen, I. (Eds.), Automation of Clinical Electroencephalography. Raven, New York. Mezzacappa, E.S., Katkin, E.S., 2002. Breastfeeding is associated with reduced perceived stress and negative mood in mothers. Health Psychology 21, 187193. Mezzacappa, E.S., Kelsey, R.M., Myers, M.M., Katkin, E.S., 2002. Breastfeeding and maternal cardiovascular function. Psychophysiology 38, 988997. Myers, J.K., Weissman, M.M., 1980. Use of self-report symptom scale to detect depression in a community sample. American Journal of Psychiatry 137, 10811083. Milligan, R., Parks, P., Lenz, E., 1990. An analysis of postpartum fatigue over the rst three months of the postpartum period. In: Wang, J., Simoni, P., Nath, C. (Eds.), Vision of Excellence: The Decade of the Nineties. West Virginia Nursing Associates, Charleston, WV.
124
Newman, J., 1995. How breast milk protects newborns. Scientic American 273, 7679. Pivik, R.T., Broughton, R.J., Coppola, R., Davidson, R.J., Fox, N.A., Nuwer, M.R., 1993. Guidelines for the recording and quantitative analysis of electroencephalographic activity in research contexts. Psychophysiology 30, 547558. Pugh, L.C., 1998. Nursing intervention to increase the duration of breastfeeding. Applied Nursing Research 11, 190194. Radloff, L.S., 1977. The CES-D scale: a self-report symptoms scale to detect depression in a community sample. American Journal of Psychiatry 137, 10811083. Reid, S.A., Duke, L.M., Allen, J.J.B., 1998. Resting frontal electroencephalographic asymmetry in depression: inconsistencies suggest the need to identify mediating factors. Psychophysiology 35, 389404. Robins, L., Helzer, J., Croughan, J., Ratcliff, K., 1981. National Institute of Mental Health Diagnostic Interview Schedule. Archives of General Psychiatry 38, 381390. Rothbart, M.K., 1981. Measurement of temperament in infancy. Child Development 52, 569578. Stifter, C., Braungart, J., 1995. The regulation of negative reactivity: function and development. Developmental Psychology 38, 448455. Stifter, C., Jain, A., 1996. Psychophysiological correlates of infant temperament: stability of behavior and autonomic patterning from 5 to 18 months. Developmental Psychology 29, 379391. Tomarken, A.J., Davidson, R.J., Wheeler, R.E., Kinney, L., 1992. Psychometric properties of resting anterior EEG asymmetry: temporal stability and internal consistency. Psychophysiology 29, 576592. VanDiver, T.A., 1997. Relationship of mothers perceptions and behavior to the duration of breastfeeding. Psychological Reports 80, 13751384. Worobey, J., 1992. Development milestones related to feeding status: evidence from the Child Health Supplement to the 198l National Health Interview Survey. Journal of Human Nutrition and Dietetics 5, 363369. Worobey, J., 1998. Feeding method and motor activity in 3-month-old human infants. Perceptual and Motor Skills 86, 883895. Zeskind, P.S., Marshall, T.R., Goff, D.M., 1992. Rhythmic organization of heart rate in breast-fed and bottle-fed newborn infants. Early Development and Parenting 1, 7987.
Prefrontal cortex activity differentiates processes affecting memory in depression

Jack B. Nitschke a, , Wendy Heller b , Marci A. Etienne c , Gregory A. Miller b,d
a
W.M. Keck Laboratory for Functional Brain Imaging and Behavior, Departments of Psychiatry and Psychology, Waisman Center, Room T229, University of Wisconsin, 1500 Highland Avenue, Madison, WI 53705-2280, USA b Department of Psychology, Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, 603 East Daniel Street, Champaign, IL 61820, USA c Department of Psychology, Mental Health Service Line, Edward J. Hines Hospital, Hines, IL 60141-5000, USA d Department of Psychiatry, University of Illinois at Urbana-Champaign, USA
Abstract Decits in the initiation and utilization of strategies contribute importantly to memory impairments in depression. Other research on depression has documented memory biases toward negative and away from positive material. This study investigated brain mechanisms accompanying the initiative decit and negative bias processes affecting memory in depressed individuals. Electroencephalography was recorded prior to and during emotional narratives and correlated with subsequent memory recognition of narrative material. Hypothesized to reect strategy initiation, bilateral activity of the prefrontal cortex (PFC) preceding a sad narrative was associated with memory performance for that narrative in nondepressed controls only. Negative memory bias in depressed participants was inferred from their association between right prefrontal activity during the sad narrative and memory performance, consistent with research implicating that region in withdrawal-related unpleasant emotions. These results highlight the importance of distinguishing processes that inuence memory performance when investigating the neural mechanisms of cognitive decit and bias in depression. 2004 Elsevier B.V. All rights reserved.
Keywords: Frontal EEG asymmetry; Emotion; Depression; Prefrontal cortex; Memory
Corresponding author. Tel.: +1-608-262-8600; fax: +1-608-262-9440. E-mail address: jnitschke@wisc.edu (J.B. Nitschke).
126
J.B. Nitschke et al. / Biological Psychology 67 (2004) 125143
1. Introduction Depression has been associated with a variety of cognitive correlates, including memory impairments and memory biases. In general, decrements have been reported for explicit memory tasks that require effortful processing (for reviews, see Burt et al., 1995; Hartlage et al., 1993). In an elegant set of studies, Hertel and her colleagues have shown that memory decits in depressed participants are eliminated when strategies are provided prior to the start of a task (for reviews, see Hertel, 1994, 1997, 2000). These data are not consistent with a capacity framework arguing for reduced attentional resources in depression. Instead, her cognitiveinitiative account posits that depressed individuals show diminished initiative in allocating available attentional resources for utilizing strategies that improve memory performance. In other studies, biases have been reported for valenced information, with depression accompanied by better recall for negative information and worse recall for positive material (for reviews, see Blaney, 1986; Watkins, 2002). Such biases have been observed for explicit memory tasks and some conceptually driven implicit memory tasks. The relationship between the initiative decit and the negative memory bias accompanying depression has not been systematically investigated, despite some evidence that they may be distinguished by their temporal and neuroanatomical patterns. Neuroimaging and neuropsychological research has shown that the prefrontal cortex (PFC), especially dorsolateral sectors of the PFC (DLPFC), mediates the type of information processing highlighted by Hertel (1994, 1997, 2000) as critical in depression (for reviews, see Davidson et al., 2002; Heller and Nitschke, 1997; Miller and Cohen, 2001). A central function of the DLPFC is thought to be the representation of goals and the maintenance of context information that promotes the achievement of these goals. Context information might include task demands, information regarding the results of previous behavior, emotional state, or any aspect of the internal or external environment that would inuence the accomplishment of the represented goals. Recruitment of the DLPFC in cognitive control has been found to be central to performance on a variety of memory tasks, including working memory (e.g., Constantinidis et al., 2001; DEsposito et al., 1998; Levy and Goldman-Rakic, 1999; Postle et al., 1999; Smith and Jonides, 1998, 1999) and episodic long-term memory (e.g., Brewer et al., 1998; Buckner et al., 1999; Fletcher et al., 1998; Petrides, 1996; Schacter, 1997), with two studies directly comparing both forms of memory (Braver et al., 2001; Ranganath et al., 2003; for reviews, see Barch and Buckner, 2003; Simons and Spiers, 2003). Of further relevance to the research conducted by Hertel (1994, 2000) on strategy utilization in depression, patients with DLPFC lesions show impairments in using organization strategies during episodic memory tasks and benet from instruction in the use of such strategies (Gershberg and Shimamura, 1995; Incisa della Rocchetta and Milner, 1993). Similar impairments in strategy utilization have been observed for Parkinsons disease (Pillon et al., 1998), schizophrenia (Iddon et al., 1998; Sengel and Lovallo, 1983), and obsessive-compulsive disorder (Savage et al., 1999). Research examining the neural circuitry of depression has identied abnormal activity patterns in the DLPFC during resting states. A number of positron emission tomography (PET) and single photon emission computerized tomography (SPECT) studies have reported bilateral DLPFC decreases in blood ow and glucose metabolism (for reviews, see
127
Davidson and Henriques, 2000; Davidson et al., 2002; Dougherty and Rauch, 1997; Heller and Nitschke, 1997, 1998). Decreased DLPFC activity has also been observed during task performance. Consistent with Hertels cognitiveinitiative framework (1994, 2000), depressed patients showed less DLPFC activity bilaterally on a complex planning task than nonpsychiatric controls in a PET study measuring blood ow (Elliott et al., 1997). Other evidence suggests that the right DLPFC may also gure importantly in the negative memory bias accompanying depression. An extensive literature has implicated right PFC mechanisms in withdrawal-related negative emotions and in threat perception. Activity in this region under such circumstances is likely to contribute to a negative memory bias (for reviews, see Davidson, 1998; Heller and Nitschke, 1997; Nitschke and Heller, 2002). Furthermore, numerous electroencephalography (EEG) studies have reported more right than left PFC activity in depression (for reviews, see Davidson and Henriques, 2000; Davidson et al., 2002; Heller and Nitschke, 1997, 1998). A similar pattern of asymmetry has also been observed in a number of the aforementioned PET studies showing bilateral DLPFC decreases, in which hypoactivity was more pronounced on the left. Although these ndings indicate an asymmetry in favor of the right PFC in depression, investigation into the brain mechanisms contributing specically to the negative explicit memory bias in depressed subjects has only recently begun. An event-related brain potential (ERP) study revealed no evidence of a negative memory bias in depressed patients, including no topographical differences for the P300 component between patients and nonpsychiatric controls during encoding or recognition of negative words or faces (Deldin et al., 2001b). Another ERP study examining the slow wave component for negative adjectives in a delayed-match-to-sample paradigm primarily implicated left parietal irregularities in depressed patients, with some right PFC abnormalities observed for correlations with depression severity (Deldin et al., 2001a). However, behavioral data indicated that the paradigm was not successful in eliciting an overt negative working memory bias in depressed patients. A recent functional magnetic resonance imaging (fMRI) study (Elliott et al., 2002) examined another type of processing bias via an emotional go/no-go task previously shown to result in a bias toward sad stimuli in depressed patients (Murphy et al., 1999). They reported greater right DLPFC activity for sad stimuli in patients than nonpsychiatric controls. Finally, an earlier blood ow study using PET by George et al. (1997) assessing attentional biases did not report DLPFC abnormalities in depressed patients for a standard Stroop paradigm or an emotional Stroop task using four words (grief, misery, sad, and bleak). Although results from studies examining the brain instantiation of negative memory biases in depression have thus far been mixed, other research on emotion, depression, and anxiety leads to predictions of right prefrontal engagement (Davidson, 1998; Davidson and Henriques, 2000; Davidson et al., 2002; Heller and Nitschke, 1997, 1998; Nitschke and Heller, 2002). A question arises, then, of whether this region will be recruited only under particular conditions or sets of conditions. In the present study, it was hypothesized that timing of the processes involved may be crucial to the degree to which different patterns of PFC activity are seen. The present study tested Hertels (1994, 2000) cognitiveinitiative framework for depression by assessing activity of the PFC during a narrative task using EEG. We predicted that depressed participants would fail to initiate PFC-mediated memory-relevant strategies shown by nondepressed participants prior to narrative presentation. More specically, an
128
association between bilateral PFC activity during preparation for listening to a sad narrative and subsequent memory for that material was expected for controls but not depressed participants. In addition, in accord with Heller and Nitschkes (1997) framework linking right PFC activity to the negative memory biases often observed in depression, depressed individuals should show an association between right PFC activity during a sad narrative and subsequent memory performance. A sad narrative was selected because of the relevance of sad material for the negative memory bias often observed in depression. For purposes of comparison, brain activity and memory performance for three other narratives of different emotional content (fearful, happy, and neutral) were also analyzed. Both hypotheses were evaluated at midfrontal sites (F3/4), which lie over the DLPFC regions implicated in PET, SPECT, and fMRI research on depression. Moreover, the frontal asymmetry ndings observed in EEG research on depression have been most consistent at these midfrontal sites.
2. Method 2.1. Participants Participants were selected based on their scores at an initial screening of 2203 undergraduate students enrolled in Introductory Psychology. Students with extreme scores on the Dysthymia scale of the General Behavior Inventory (GBI; Depue et al., 1989) and the Trait scale of the State-Trait Anxiety Inventory (STAI; Spielberger, 1968) were classied into four original groups. Individuals scoring below 0.5 S.D. above the mean on both the GBI-D and STAI met criteria for the control group, whereas those scoring above the 90th percentile on both were assigned to the mixed depression/anxiety group. The depression group was comprised of students scoring above the 90th percentile on the GBI-D and below 0.5 S.D. above the mean on the STAI. The anxiety group was comprised of students scoring above the 90th percentile on the STAI and below 0.5 S.D. above the mean on the GBI-D. Of the 68 participating in the EEG study, 8 were in the depression group, and 20 were in each of the other three groups. Participants were screened again prior to the EEG session and only those showing stable scores across both times on both selection instruments were included in analyses. For the GBI-D, low and high scorers had to maintain scores below and above 10, respectively, whereas the cut-off for low and high scorers on the STAI was 50. In order to maintain reasonable sample sizes, the following exceptions to these criteria were made. The STAI cut-off was 45 for the anxiety and mixed depression/anxiety groups. In the depression group, one of the three participants not meeting the GBI-D criterion endorsed multiple items on the GBI Bipolar scale. Therefore, a cut-off of 8 was applied when combining the Dysthymia and Bipolar scales, which resulted in the addition of one additional subject in the mixed depression/anxiety group as well. Missing EEG data for one control subject resulted in a nal sample size of 54 participants: 15 control (8 female), 6 depression (3 female), 16 anxiety (8 female), and 17 mixed depression/anxiety (10 female). Because the hypotheses concerned depression, the two original groups with minimal depressive symptoms (control and anxious groups) were combined into a low depression group (n = 31), whereas the other two groups reporting high levels of depression (depression and mixed depression/anxiety) comprised the high depression group (n = 23). To
129
examine effects of anxiety, a similar strategy was adopted. The two original groups with minimal anxious symptoms (control and depression groups) were combined into a low anxiety group (n = 21), whereas the other two groups reporting high levels of anxiety (anxiety and mixed depression/anxiety) comprised the high anxiety group (n = 33). The majority of participants were strongly right-handed (M = 80.24, S.D. = 20.99) as determined by the Edinburgh Handedness Inventory (EHI; Oldeld, 1971). Only two individuals scored below 45, with scores of 0 and 10 (maximum score of 100 indicates strong right-handedness, 100 corresponds to strong left-handedness, and 0 to ambidextrous). Analyses conducted without those two participants did not alter the pattern of results. Individuals reporting hypomanic symptoms with scores above the 90th percentile on the GBI Biphasic and Hypomania scales combined were excluded. 2.2. Self-report screening measures The GBI is a 73-item questionnaire that assesses subsyndromal levels of chronic depression and mania in psychiatric outpatient and nonclinical populations. Internal consistency in a nonclinical sample was 0.96 (Depue and Klein, 1988). The test-retest reliability with a nonclinical sample was 0.75 (Depue and Klein, 1988). The 46-item Dysthymia scale was used to select participants with low or high levels of depression for this study, and the Biphasic and Hypomania scales were used to exclude cyclothymic or manic students. The STAI is a 40-item questionnaire that assesses state and trait anxiety. Internal consistency is in the low 0.90s with college students for both the state and trait forms (Spielberger, 1983). Previous studies have reported test-retest reliability with college students in the 0.70s (Spielberger, 1983). The 20-item Trait scale was used to classify participants. The EHI is a 10-item questionnaire assessing hand preference as a continuum. Each item lists a task, and the participant responds with the degree of preference for performing the task with the right or left hand. Persons who are left-handed, ambidextrous, or ambilateral can show different patterns of brain organization than right-handed individuals. Because degree of right-handedness can vary among those who report being right-handed, this inventory was used to conrm the degree of right-handedness of participants (laterality quotients greater than zero). 2.3. Narrative script materials Because the development of the narratives has previously been described extensively (Heller et al., 1997), only a brief explanation is provided here. Four narrative scripts were selected from a pilot sample of 12 narratives, three for each of the emotion categories of sad, happy, fearful, and neutral. The narratives were designed to probe for both explicit and implicit memory. Hence, each pilot narrative was constructed with one exemplar (e.g., scooter) from each of six different object categories (e.g., motor vehicle). Exemplars were selected based on word frequency research (Battig and Montague, 1969). The range and word frequency for exemplars in each narrative category were closely matched. Pilot narratives were read by 19 raters (11 female) who completed an adjective checklist after reading each narrative. Participants were asked to rate the content of the narrative along a calm-aroused dimension and a pleasant-unpleasant dimension. In addition, raters indicated
130
the amount of discrete emotion present in the content of the narrative for the following emotion descriptors: happy, sad, fear, anger, surprise, disgust, interest, and amusement. All ratings were made by marking a slash on a 100 mm vertical bar for each dimension or emotion adjective. Narratives were selected based on their representativeness of the target emotion content for each of the four categories (for further details, see Heller et al., 1997; Etienne, 2002). To test the proposed hypotheses concerning cognitive abnormalities in depression, analyses concentrated on the sad narrative due to the relevance of sad material to depression and the negative memory biases observed in depression. All participants heard the same sad narrative: Steve just drove his motor scooter off of the edge. Still wearing your shorts from when you played tennis just hours earlier, you and Karen stood stunned, as you listened to the details of what would be called a suicide. You close your eyes and can picture him in his denims waving good-bye to you after he dropped you off. You feel a lump build in your throat as you touch the garnet ring he had just left with you. 2.4. Apparatus and physiological recording During the laboratory session, participants were seated in a comfortable chair in a quiet room connected to the adjacent equipment room via intercom. A computer-controlled tape recorder was used to present narratives over a centrally located loudspeaker. A 500 ms 1000 Hz tone generated by a Beckman Circuitmate function generator was presented at a comfortable listening level over the loudspeaker to signify the start and end of each narrative period as described below. EEG recording used the International 1020 system (Jasper, 1958) with electrodes at sites F3, F4, Cz, P3, P4, and right mastoid, each referenced to the left mastoid. The right and left mastoid were algebraically linked off-line to serve as reference for the other sites (Davidson, this issue; Miller et al., 1991; Nunez, 1981). Beckman miniature AgAgCl electrodes were used for EEG recording and were placed with Grass EC2 electrode paste. Horizontal and vertical EOG (electro-oculogram) were also recorded. Beckman miniature AgAgCl electrodes were placed above and below the left eye and near the outer canthus of each eye. Electrode impedances were below 10 k . Stimulus presentation and physiological data collection were controlled by an LSI-11/73 based PEARL II-B microcomputer (Hefey et al., 1985). Recordings were amplied by a Grass Model 12 Neurodata system with EOG channels set to a gain of 2 K and EEG channels set to a gain of 10 K. Half-amplitude frequency cutoffs were set at 0.01 and 100 Hz. Analog signals were digitized on-line at 250 Hz. 2.5. Procedure During an initial session, a general tour of the laboratory was conducted, a detailed description of the study was presented, and informed consent was obtained. Participants were screened for known health conditions that would affect data collection during physiological recording (e.g., closed-head injury with post-traumatic amnesia). Participants were asked
131
to refrain from the use of psychoactive drugs on the day of the laboratory session (including tobacco, alcohol, and caffeine). Participants were reassessed during this initial session for trait anxiety with the STAI and for chronic depression with the GBI. At the EEG session, 8 min of resting baseline EEG were collected using counter-balanced eyes-open (O), and eyes-closed (C) 1 min trials, ordered COOCOCCO for half the participants and OCCOCOOC for the other half. Participants received instructions on the terminal screen in front of them to either open (eyes open) or close (eyes closed) their eyes. The start of each 1 min period was marked by a single tone, and the end of each minute was marked by a double tone presented over the loudspeaker. These resting baseline EEG data have been published (Heller et al., 1997) and are not presented here. Following the collection of baseline data, the trial structure for the narrative presentation was explained. Important for Hertels proposed cognitiveinitiative decit in depression (Hertel, 1994, 2000), participants were not informed of the memory tasks that would follow nor provided with any strategies for attending to the narratives. Four narrative trials were presented. The order of narrative presentation was counterbalanced across participants. The trial structure was as follows: 30 s Preparation, 30 s Listen (tape-recorded narrative), 30 s Relax. Eyes-closed EEG was collected during all three periods of all four narrative presentations. Following each Relax period, the participant completed ratings about the narrative. After all four narrative scripts were presented, explicit memory recognition of the narratives was assessed. The recognition test consisted of a sheet of paper with 32 words typed in a 4 8 grid. Sixteen of the words were targets (four of the category targets selected from each of the four narratives presented), and the other sixteen were semantically related distractors (e.g., Steve was a target, Larry was a distractor). One distractor was selected from each of the 16 categories represented by the targets, and each had a word frequency count comparable to its corresponding target. Target words were randomly distributed throughout the grid. Participants were asked to circle any words that were in any of the four short stories listened to at the beginning of the study. Number of hits and false positives were scored. Because very few subjects had any false positives, analyses were conducted for hits only. Implicit memory was assessed by a primed exemplar production task, following the procedures outlined by Graf et al. (1985). This task was not successful in eliciting implicit memory; therefore, group differences and relations with EEG were not assessed. 2.6. EEG data reduction and analysis An eye-movement correction program was used off-line to remove EOG artifact from the EEG (Gratton et al., 1983; Miller et al., 1988), with raw data processed in contiguous 5 s segments. The complete data sample was then digitally ltered for alpha frequency (813 Hz) with a 501-weight lter (Cook, 1981; Cook and Miller, 1992; Nitschke et al., 1998). Filter gain at 8.3, 10.7, and 12.7 Hz was 0.85, 0.99, and 0.85, respectively. EEG alpha activity was quantied by rst computing a root mean square (RMS) score for each second of alpha-ltered EEG. An average RMS value was then computed separately for the eyes-open and eyes-closed baselines, and separately for each narrative period (Preparation, Listen, Relax). Distributions for the averaged RMS values at each EEG site were consistently skewed; therefore, log transformations were performed to reduce skew in these distributions.
132
All MANOVAs and regressions were conducted using alpha = 0.05, with the directional hypotheses following from Hertels (1994, 2000) cognitiveinitiative model and Heller and Nitschkes (1997) neuropsychological framework for cognitive biases evaluated via one-tailed distributions. Reported analyses do not include Sex as a factor because parallel MANOVAs and regressions with Sex as an additional factor revealed no sex differences.
3. Results 3.1. Recognition memory A Depression (Low, High) Anxiety (Low, High) Narrative (Sad, Fearful, Neutral, Happy) repeated-measures MANOVA with memory performance (number of hits) as the dependent variable revealed main effects for Depression, F(1, 50) = 5.89, P < 0.05, 2 = 0.105; Anxiety, F(1, 50) = 8.98, P < 0.01, 2 = 0.152; and Narrative, F(3, 48) = 4.62, P < 0.01, 2 = 0.224 (see Table 1). High depression participants performed more poorly on the recognition test than low depression participants. Conversely, high anxious participants showed better memory performance than low anxious participants. Post hoc analyses using the Least Signicant Difference test or a Bonferroni correction for all six possible pairwise comparisons indicated that recognition was better for the sad narrative than the fearful or neutral narratives and no signicant differences for the other four comparisons. Whereas none of the interactions involving Narrative was signicant, Depression Anxiety was F(1, 50) = 5.86, P < 0.05, 2 = 0.105. This interaction effect is reected in an analogous MANOVA using the four original groups, which revealed a main effect for Group, F(3, 50) = 4.43, P < 0.01, 2 = 0.210. The depression group (participants with high depression and low anxiety) recognized fewer words from the narratives than did each of the other groups, including the mixed depression/anxiety group (Tukeys HSD). There were no pairwise differences among the other three groups. Of relevance for the present report focusing on the sad narrative, the above between-subjects effects were present for a comparable MANOVA conducted on memory performance for the sad narrative alone: Depression, F(1, 50) = 9.05, P < 0.01, 2 = 0.153; Anxiety, F(1, 50) = 7.63, P < 0.01, 2 = 0.132; and Depression Anxiety, F(1, 50) = 7.71, P < 0.01, 2 = 0.134.
Table 1 Means and standard deviations for memory recognition of words used in the four narratives Narrative Low depression (n = 31) Low anxiety (n = 15) M Sad Fearful Neutral Happy 3.07 2.00 2.33 2.27 S.D. 1.10 0.76 1.18 1.10 High anxiety (n = 16) M 3.06 2.19 2.12 2.69 S.D. 0.85 1.22 1.26 0.87 High depression (n = 23) Low anxiety (n = 6) M 1.50 1.33 1.83 1.67 S.D. 0.84 0.82 1.47 1.37 High anxiety (n = 17) M 3.00 2.35 2.06 2.65 S.D. 0.79 0.79 1.03 1.06
133
3.2. EEG Patterns of brain activity were assessed using a repeated-measures MANOVA with Depression (Low, High) and Anxiety (Low, High) as between-subjects variables and four within-subjects variables: Narrative (Sad, Fearful, Neutral, Happy), Period (Preparation, Listen), Region (Frontal, Parietal), and Hemisphere (Left, Right). A main effect for Depression, F(1, 48) = 5.16, P < 0.05, 2 = 0.097, revealed that high depression participants had less activity across all frontal and parietal regions (more global alpha) than low depression participants. There were only two other signicant effects involving Depression or Anxiety: Depression Anxiety Narrative Period Region, F(3, 46) = 2.93, P < 0.05, 2 = 0.160; and Anxiety Narrative Period Hemisphere, F(3, 46) = 3.22, P < 0.05, 2 = 0.174. For a parallel MANOVA on the sad narrative alone, there were no signicant effects involving Depression or Anxiety. Following from the focus of the hypotheses on frontal regions, another MANOVA for the sad narrative was conducted for frontal sites only. Again, there was a main effect for Depression, F(1, 50) = 4.28, P < 0.05, 2 = 0.079, indicating less bilateral frontal activity (more alpha) in high depression than low depression participants. The only other signicant effect involving Depression or Anxiety was a Hemisphere Anxiety interaction, F(1, 50) = 7.24, P < 0.05, 2 = 0.126, indicating greater left than right activity in high than low anxiety participants (see Heller et al., 1997). Also of relevance to the current report was a marginally signicant Depression Period Hemisphere interaction, F(1, 48) = 3.73, P < 0.06, 2 = 0.069, due to greater left than right activity in low than high depression participants during preparation for the sad narrative, F(1, 50) = 4.01, P < 0.06, 2 = 0.074. 3.3. Recognition memory and EEG Hierarchical regression analyses evaluating the two hypotheses under investigation were conducted on the EEG and memory data for the sad narrative. The omnibus model predicting recognition memory performance for the sad narrative included Depression (Low, High), Anxiety (Low, High), frontal and parietal sites during the preparation and listen periods, and interactions between Depression and each EEG variable. The predicted effect based on Hertels (1994, 2000) model was an interaction between Depression and bilateral frontal activity during the preparation period. The predicted effect based on Heller and Nitschkes (1997) framework for negative memory bias was an interaction between Depression and right frontal activity during the listen period. In this and subsequent regression models, parietal sites were entered before frontal sites for two reasons: to test that neither right nor left parietal activity was associated with subsequent memory performance and as a means of removing variance associated with global EEG power prior to testing hypotheses for frontal regions. Results for parallel regressions omitting parietal sites replicated the effects reported below and are therefore not reported to avoid redundancy. When left and right sites were added on different steps, left were routinely entered rst for the results presented here. The pattern of results did not change if right sites were entered rst. Anxiety status was included in the model to account for variance associated with anxiety; however, interaction terms involving Anxiety were omitted from the model due to the high number of predictors (18) already included.
134
Depression and Anxiety entered on the rst step were signicant predictors of memory performance for the sad narrative, R2 = 0.12, F(2, 51) = 3.39, P < 0.05. As for the MANOVAs conducted on memory performance above, high depression was associated with poorer memory performance, t(51) = 2.11, P < 0.05, and high anxiety was associated with better memory performance, t(51) = 1.96, P < 0.06. Entering all eight frontal and parietal values for the preparation and listen periods on the second step did not result in a statistically signicant increment in variance explained, R2 = 0.15, F(8, 43) = 1.06, P > 0.40. The interaction of depression and each parietal site during the preparation period was not signicant when entered next, R2 = 0.00, F(2, 41) = 0.08, P > 0.92. As predicted, the interaction of depression and each frontal site during the preparation period was signicant, R2 = 0.10, F(2, 39) = 3.18, P < 0.05. The Depression Left Frontal interaction and the Depression Right Frontal interaction each were signicant predictors of memory performance if entered alone at this step, R2 = 0.10, t(40) = 2.49, P < 0.01, and R2 = 0.08, t(27) = 2.22, P < 0.05, respectively. Interaction terms involving Depression and EEG during the listen period were not signicant when entered on subsequent steps. Contrary to prediction, the Depression Right Frontal interaction when entered last did not signicantly predict memory performance for the sad narrative, R2 = 0.02, t(36) = 1.07, P < 0.15. Given that the inclusion of 18 variables in this model limits the statistical power to detect an effect for the last variables included in the model, an alternative model was tested by not including the eight EEG variables on the second step. For this model, the critical Depression Right Frontal interaction entered on the nal step was signicant, R2 = 0.05, t(43) = 1.70, P < 0.05, indicating that low and high depression are accompanied by different relationships between right frontal alpha during the sad narrative and subsequent memory performance. To examine the above signicant interactions involving depression status and EEG alpha, which are central to the hypotheses of this study, the low and high depression participants were assessed separately. A similar hierarchical regression approach was employed with Anxiety (Low, High) entered on the rst step. As a direct test of Hertels (1994, 2000) cognitiveinitiative model, left and right frontal alpha during the preparation period were added to the regression model after left and right parietal alpha during that period. The low depression group showed an association between less alpha (more activity) frontally and better memory performance, R2 = 0.23, F(2, 25) = 4.51, P < 0.01, whereas no signicant relationship was observed for participants reporting high levels of Depression, R2 = 0.05, F(2, 17) = 0.79, P > 0.23 (see Table 2, Fig. 1). For the low depression group, each frontal site contributed equivalently to the prediction of memory performance when entered alone: left frontal, R2 = 0.21, t(26) = 2.90, P < 0.01, and right frontal, R2 = 0.17, t(26) = 2.52, P < 0.01. Alpha during the listen period was then added to the model to assess Heller and Nitschkes (1997) neuropsychological framework for cognitive biases in depression. Right frontal alpha during the listen period added on the nal step (after left and right parietal and left frontal alpha during listen and all four sites during preparation) resulted in no change for the low depression participants, R2 = 0.00, t(21) = 0.12, P > 0.98, but produced a further increment in variance explained among high depression participants, R2 = 0.18, t(13) = 2.99, P < 0.01. As predicted for depression, less right frontal alpha (more ac-
Table 2 Hierarchical regressions for EEG prior to and during sad narrative predicting subsequent memory recognition for that narrative Variable Low depression (n = 31) R2 1. Anxiety 0.000 0.151 0.225 0.020 0.000 0.000 0.396 R2 Test F(1, 29) = 0.00 F(2, 27) = 2.40 F(2, 25) = 4.51 F(2, 23) = 0.37 F(1, 22) = 0.00 F(1, 21) = 0.00 F(9, 21) = 1.53 P <1.00 <0.12 <0.02 <0.70 <0.97 <1.00 <0.21 0.742 High depression (n = 23) R2 0.425 0.015 0.047 0.076 0.003 0.177 R2 Test F(1, 21) = 15.52 F(2, 19) = 0.25 F(2, 17) = 0.79 F(2, 15) = 1.30 F(1, 14) = 0.08 F(1, 13) = 8.91 F(9, 13) = 4.15 P <0.002 <0.79 <0.24 <0.31 <0.79 <0.01 <0.02
EEG alpha during preparation period 2. Left and right parietal 3. Left and right frontal EEG alpha during listen period 4. Left and right parietal 5. Left frontal 6. Right frontal Full model
Note: Numbered rows (15) indicate the order in which variables were entered into the regression equations. Rows 3 and 6 provide tests of hypotheses, and therefore corresponding probability levels are one-tailed. All other probability levels are two-tailed.
135
136
Fig. 1. Bilateral frontal activity (decreased alpha) during the preparation period preceding the sad narrative predicted improved memory performance for that narrative in low depression but not high depression participants. All values have parietal activity during the preparation period partialled out, consistent with the main regressions reported in text. Regression slopes and individual participants values are plotted, and corresponding partial correlations are shown. Filled squares are low anxiety participants. Open circles are high anxiety participants. Removing the low depression outlier with the lowest memory score resulted in very similar results for the low depression group, rp = 0.41, P < 0.02 (one-tailed).
tivity) while listening to the narrative was associated with better memory performance, as illustrated in Fig. 2. Supplemental analyses provided further support for the hypotheses. Similar hierarchical regressions were conducted for alpha values averaging across left and right EEG sites. When added to the model after bilateral parietal alpha, the averaged frontal alpha showed the same
Fig. 2. Right frontal activity (decreased alpha) during the sad narrative predicted improved memory performance for that narrative in high depression but not low depression participants. All values have bilateral parietal and frontal activity preceding the narrative and bilateral parietal and left frontal activity during the narrative partialled out, consistent with the main regressions reported in text. Regression slopes and individual participants values are plotted, and corresponding partial correlations are shown. Filled squares are low anxiety participants. Open circles are high anxiety participants.
137
relationship mentioned above for low depression participants, R2 = 0.13, t(27) = 2.08, P < 0.05, and again no signicant relationship for high depression, R2 = 0.02, t(19) = 0.75, P > 0.46. The averaged frontal alpha during the listen period added on the nal step did not predict memory performance for either group, both P > 0.58. Hierarchical regression models were also conducted on asymmetry scores predicting memory performance for the sad narrative. For the low depression participants, the only signicant effect to emerge was for frontal asymmetry was during the preparation period entered after anxiety status and parietal asymmetry during that period, R2 = 0.15, t(27) = 2.34, P < 0.05, indicating that the association between increased bilateral frontal activity and better memory recognition is particularly pronounced on the left. Conversely, for the high depression participants, the only effect observed for frontal asymmetry was during the listen period entered on the nal step of the model, R2 = 0.08, t(17) = 1.87, P < 0.05. The direction of this effect is consistent with the above ndings of more right frontal activity predicting better memory performance for the high depression participants. Taken together, these results indicate a bilateral frontal effect immediately prior to presentation of the narrative as expected for Hertels (1994, 2000) model and additionally a unilateral right-sided frontal effect while listening to the narrative as predicted by Heller and Nitschke (1997). Follow-up analyses revealed that the results of the above regressions for the low depression group were equally strong for both the control and anxiety original groups, and neither group showed a right frontal effect during the listen period. The effects for the high depression group were observed when the original mixed depression/anxiety group was analyzed alone, but the n of 8 for the depression group was insufcient for the regression analyses conducted. The specicity of the ndings to depression was further assessed by dividing the sample into a low and a high anxiety group. No frontal effects emerged for EEG during the preparation period for either of the two omnibus models testing interactions of anxiety status and EEG activity, P > 0.60. However, there was an effect for right frontal activity during the listen period entered on the nal step of the omnibus regression model including the eight EEG variables on the second step, R2 = 0.06, t(36) = 1.95, P < 0.03 (one-tailed). This effect was largely driven by an association between right frontal activity in response to the sad narrative and subsequent memory performance for that narrative in the high anxious participants, R2 = 0.08, t(23) = 1.71, P < 0.06 (one-tailed), as determined by separate regressions conducted for the low and high anxious participants as done above for low and high depression participants. The frontal effects observed for the sad narrative were not obtained for parietal sites or the other three narratives (fearful, happy, and neutral). Analogous omnibus regressions with parietal sites entered after the frontal sites resulted in no signicant parietal effects during the preparation or listen periods, all P > 0.26. Regression predicting memory performance for each of the other three narratives revealed no frontal effects during either period, all P > 0.23.
4. Discussion This study builds on prior research examining cognitive correlates of depression by investigating brain mechanisms associated with two distinct cognitive abnormalities often
138
accompanying depression. EEG was collected immediately prior to and during the auditory presentation of a sad narrative in a sample of college students with extreme levels (low and high) of depression and anxiety. Bilateral activity recorded over the PFC during a preparatory period immediately preceding the sad narrative was associated with better memory performance in participants reporting low levels of depression but not in those with high levels of depression. This nding is consistent with behavioral studies conducted by Hertel and colleagues (for reviews, see Hertel, 1994, 2000) suggesting that depressed individuals do not initiate task-relevant cognitive processes unless concrete strategies are provided. Corroborating other research implicating the right PFC in withdrawal-related negative emotions and threat perception, factors that may play a role in negative cognitive biases (Heller and Nitschke, 1997; Nitschke and Heller, 2002), high depression participants showed an association between right PFC activity when exposed to the sad narrative and improved subsequent recognition of words used in that narrative. Although the experimental design employed does not allow indisputable conclusions about the precise aspects of memory differentially associated with distinct patterns of brain activity in low and high depression, these data are consistent with previous research attributing these cognitive functions to the PFC (Heller and Nitschke, 1997; Miller and Cohen, 2001; Nitschke and Heller, 2002) and investigating the cognitive neuroscience of depression (Davidson et al., 2002; Heller and Nitschke, 1997). Previous research on asymmetries in emotion and memory are germane to these data. As alluded to above, the right PFC ndings are highly consistent with prior studies investigating various forms of withdrawal-related negative emotions (Davidson et al., 2002, 2003; Heller and Nitschke, 1997; Nitschke and Heller, 2002). On the other hand, the bilateral frontal effect supporting Hertels (1994, 2000) model was particularly pronounced on the left, as indicated by the analysis on asymmetry scores. Such a pattern would be expected to the extent that initiating cognitive strategies is approach-related (e.g., Davidson et al., 2003) or involves verbal processing. In addition, the analyses for EEG data (regardless of memory performance) replicated previous ndings in the literature. During preparation for the sad narrative, there was a trend for high depression subjects to show relatively less left than right frontal activity compared to the low depression subjects (for review, see Davidson et al., 2002). The nding of relatively more left than right activity in high than low anxiety participants has been reported previously for this same sample subdivided differently (Heller et al., 1997) and is consistent with ndings for anxious apprehension (for review, see Nitschke et al., 2000). One model of episodic memory has accorded different memory-related functions to the left and right PFC (Habib et al., 2003; Tulving et al., 1994), a position that has met with considerable controversy (Fletcher and Henson, 2001; Lee et al., 2000; Owen, 2003; Shallice, 2003; Simons and Spiers, 2003). Although it is feasible that the association between right frontal EEG and memory in the high depression participants might be due to right PFC involvement in episodic memory retrieval, as proposed in Tulvings model, it seems unlikely this would be more true for high than low depression participants. The differential involvement of the left PFC in retrieving information from semantic memory and encoding novel material is not a likely contributor to the nding of greater left than right activity since that was observed during the preparation period prior to the presentation of material to be encoded or retrieved. Furthermore, it is important to note that their model is largely based on earlier PET studies, none of which employed a rigorous test of asymmetry (Davidson
139
and Irwin, 1999). Taken together, the asymmetry ndings are most consistent with previous work on the neural correlates of emotion. The study design afforded the possibility of testing the specicity of the ndings in several respects. First, the predicted associations between brain activity and memory performance were restricted to the sad narrative. This was expected for the right PFC ndings indicative of a negative memory bias, because of the relevance of sadness to depression. We predicted that the bilateral PFC effect, however, would be seen in preparation for all narratives. The poorer memory performance for the other three narratives may explain why the effect was not also observed for them. Second, careful measurement of anxiety revealed that the results obtained for depression could not be explained by co-occurrence with anxiety. Although there was a right frontal effect observed across participants reporting high levels of anxiety, that effect was carried by the subset of participants also showing high depression. Third, with regard to the specicity of the effects to the sites over the PFC, analyses revealed no signicant results for parietal sites. Further claims about spatial localization are not possible due to the limited number of electrode sites sampled. For example, the EEG methods employed were not able to differentiate between sectors of the right PFC subserving cognitiveinitiative and memory bias functions. Finally, these ndings point to an important temporal distinction between the two cognitive abnormalities assessed, with the putative cognitiveinitiative decit associated with bilateral PFC hypoactivity immediately preceding the narrative and the negative memory bias related to right PFC hyperactivity during the narrative. The observed associations between brain activity and subsequent memory performance is consistent with two seemingly conicting literatures, one documenting memory decits in depression and the other reporting better memory performance for negative material. Those studies nding memory decits in depressed samples have either not used negative stimuli or have not systematically assessed the emotional content of the material tested. The present study provides a neurocognitive account for why memory performance in depression might be impaired in some circumstances and enhanced in others. Furthermore, the conuence of the cognitiveinitiative decit producing memory impairment and the negative memory bias producing selective memory enhancement may result in no apparent memory effects in depressed individuals under certain circumstances. In line with this position, there was some indication of a memory decit in depression across all four narratives, but only in the small sample of depressed participants reporting low anxiety. This initial study on the brain instantiation of cognitiveinitiative and memory bias abnormalities points to a number of future research directions. One limitation of the current study is that it did not include an experimental manipulation or behavioral measure of strategy initiation or utilization (Gershberg and Shimamura, 1995; Hertel, 1994, 1997, 2000; Incisa della Rocchetta and Milner, 1993). Second, selecting narratives or other stimuli eliciting comparable levels of memory performance across emotion categories will be important in assessing the relative contributions of the cognitiveinitiative decit and the negative memory bias to memory performance in depression. Finally, other physiological recording methods such as dense-array EEG using source localization tools or hemodynamic neuroimaging methodologies will be important in distinguishing the specic sectors of the PFC contributing to the cognitiveinitiative decit and to the negative memory bias. This study investigated neural mechanisms that may account for the behavioral data supporting Hertels (1994, 2000) model of memory performance in depression. Present
140
ndings support that model and suggest that the failure to recruit the PFC in preparation for information processing may result in poorer performance on memory tasks. In addition, these data support Heller and Nitschkes (1997) claim that recruiting the right PFC under conditions of negative emotion or threat may be associated with better memory performance. These results highlight the importance of distinguishing different processes inuencing memory performance when investigating the neural mechanisms of cognitive decit and bias in depression.
Acknowledgements The authors wish to thank E. Keolani Taitano and Tara Cecola for their help in collecting these data and Krystal Cleven, Kristen Mackiewicz, Alissa Possin, and Emily Seidler for assistance with manuscript preparation. This research was supported by NIH grants R21 DA14111, R01 MH39628, R03 MH52079, and R01 MH61358 and a University of Illinois Research Board Award. JBN was also supported by K08 MH63984, T32 MH14257 (Larry Jones, Director), T32 MH18931 (Richard J. Davidson, Director), and a HealthEmotions Research Institute Fellowship. WH was also supported by MH19664. A portion of this work contributed to the dissertation of MAE.
References
Barch, D.M., Buckner, R.L., 2003. Memory. In: Fogel, B.S., Schiffer, R.B., Rao, S.M. (Eds.), Neuropsychiatry, second ed. Lippincott, Williams & Wilkins, Baltimore, pp. 426442. Battig, W.F., Montague, W.E., 1969. Category norms for verbal items in 56 categories: a replication and extension of the Connecticut category norms. Journal of Experimental Psychology Monograph 80 (Pt. 2), 146. Blaney, P.H., 1986. Affect and memory: a review. Psychological Bulletin 99, 229246. Braver, T.S., Barch, D.M., Kelley, W.M., Buckner, R.L., Cohen, N.J., Miezin, F.M., Snyder, A.Z., Ollinger, J.M., Akbudak, E., Conturo, T.E., Petersen, S.E., 2001. Direct comparison of prefrontal cortex regions engaged by working and long-term memory tasks. NeuroImage 14, 4859. Brewer, J.B., Zhao, Z., Desmond, J.E., Glover, G.H., Gabrieli, J.D., 1998. Making memories: brain activity that predicts how well visual experience will be remembered. Science 281, 11851187. Buckner, R.L., Kelley, W.M., Petersen, S.E., 1999. Frontal cortex contributes to human memory formation. Nature Neuroscience 2, 311314. Burt, D.B., Zembar, M.J., Niederehe, G., 1995. Depression and memory impairment: a meta-analysis of the association, its pattern, and specicity. Psychological Bulletin 117, 285305. Constantinidis, C., Franowicz, M.N., Goldman-Rakic, P.S., 2001. The sensory nature of mnemonic representation in the primate prefrontal cortex. Nature Neuroscience 4, 311316. Cook, E.W., 1981. FWTGENan interactive FORTRAN II/IV program for calculating weights for a non-recursive digital lter. Psychophysiology 18, 489490. Cook, E.W., Miller, G.A., 1992. Digital ltering: background and tutorial for psychophysiologists. Psychophysiology 29, 350367. DEsposito, M., Aguirre, G.K., Zarahn, E., Ballard, D., Shin, R.K., Lease, J., 1998. Functional MRI studies of spatial and nonspatial working memory. Cognitive Brain Research 7, 113. Davidson, R.J., 1998. Affective style and affective disorders: perspectives from affective neuroscience. Cognition and Emotion 12, 307330. Davidson, R.J., Henriques, J.B., 2000. Regional brain function in sadness and depression. In: Borod, J.C. (Ed.), The Neuropsychology of Emotion. Oxford University Press, New York, pp. 269297.
141
Davidson, R.J., Irwin, W., 1999. The functional neuroanatomy of emotion and affective style. Trends in Cognitive Science 3, 1121. Davidson, R.J., Pizzagalli, D., Nitschke, J.B., Putnam, K., 2002. Depression: perspectives from affective neuroscience. Annual Review of Psychology 53, 545574. Davidson, R.J., Pizzagalli, D., Nitschke, J.B., Kalin, N.H., 2003. Parsing the subcomponents of emotion and disorders of emotion: perspectives from affective neuroscience. In: Davidson, R.J., Scherer, K.R., Goldsmith, H.H. (Eds.), Handbook of Affective Science. Oxford University Press, New York, pp. 824. Deldin, P.J., Deveney, C.M., Kim, A.S., Casas, B.R., Best, J.L., 2001a. A slow wave investigation of working memory biases in mood disorders. Journal of Abnormal Psychology 110, 267281. Deldin, P.J., Keller, J., Gergen, J.A., Miller, G.A., 2001b. Cognitive bias and emotion in neuropsychological models of depression. Cognition and Emotion 15, 787802. Depue, R.A., Klein, D.N., 1988. Identication of unipolar and bipolar affective conditions in nonclinical and clinical populations by the General Behavior Inventory. In: Dunner, D.L., Gershon, E.S., Barrett, J.E. (Eds.), Relatives at Risk for Mental Disorder. Raven Press Ltd., New York, pp. 179204 Depue, R.A., Krauss, S., Spoont, M.R., Arbisi, P., 1989. General Behavior Inventory identication of unipolar and bipolar affective conditions in a nonclinical university population. Journal of Abnormal Psychology 98, 117126. Dougherty, D., Rauch, S.L., 1997. Neuroimaging and neurobiological models of depression. Harvard Review of Psychiatry 5, 138159. Elliott, R., Baker, S.C., Rogers, R.D., OLeary, D.A., Paykel, E.S., Frith, C.D., Dolan, R.J., Sahakian, B.J., 1997. Prefrontal dysfunction in depressed patients performing a complex planning task: a study using positron emission tomography. Psychological Medicine 27, 931942. Elliott, R., Rubinsztein, J.S., Sahakian, B.J., Dolan, R.J., 2002. The neural basis of mood-congruent processing biases in depression. Archives of General Psychiatry 59, 597604. Etienne, M.A., 2002. Information processing and regional brain activity in anxiety and depression. Unpublished Doctoral Dissertation. University of Illinois at Urbana-Champaign. Fletcher, P.C., Shallice, T., Dolan, R.J., 1998. The functional roles of prefrontal cortex in episodic memory. I. Encoding. Brain 121, 12391248. Fletcher, P.C., Henson, R.N.A., 2001. Frontal lobes and human memory: insights from functional neuroimaging. Brain 124, 849881. George, M.S., Ketter, T.A., Parekh, P.I., Rosinsky, N., Ring, H.A., Pazzaglia, P.J., Marangell, L.B., Callahan, A.M., Post, R.M., 1997. Blunted left cingulate activation in mood disorder subjects during a response interference task (the Stroop). Journal of Neuropsychiatry & Clinical Neurosciences 9, 5563. Gershberg, F.B., Shimamura, A.P., 1995. Impaired use of organizational strategies in free recall following frontal lobe damage. Neuropsychologia 33, 13051333. Graf, P., Shimamura, A.P., Squire, L.R., 1985. Priming across modalities and priming across category levels: extending the domain of preserved function in amnesia. Journal of Experimental Psychology: Learning, Memory, and Cognition 11, 386396. Gratton, G., Coles, M.G.H., Donchin, E., 1983. A new method for off-line removal of ocular artifact. Electroencephalography and Clinical Neurophysiology 55, 468484. Habib, R., Nyberg, L., Tulving, E., 2003. Hemispheric symmetries of memory: the HERA model revisited. Trends in Cognitive Science 7, 241245. Hartlage, S., Alloy, L.B., Vazquez, C., Dykman, B., 1993. Automatic and effortful processing in depression. Psychological Bulletin 113, 247278. Hefey, E., Foote, B., Mui, T., Donchin, E., 1985. PEARL II: portable laboratory computer system for psychophysiological assessment using event related brain potentials. Neurobehavioral Toxicology and Teratology 7, 399407. Heller, W., Nitschke, J.B., 1997. Regional brain activity in emotion: a framework for understanding cognition in depression. Cognition and Emotion 11, 637661. Heller, W., Nitschke, J.B., 1998. The puzzle of regional brain activity in depression and anxiety: the importance of subtypes and comorbidity. Cognition and Emotion 12, 421447. Heller, W., Nitschke, J.B., Etienne, M.A., Miller, G.A., 1997. Patterns of regional brain activity differentiate types of anxiety. Journal of Abnormal Psychology 106, 376385.
142
Hertel, P.T., 1994. Depression and memory: are impairments remediable through attentional control? Current Directions in Psychological Science 3, 190193. Hertel, P.T., 1997. On the contributions of decient cognitive control to memory impairments in depression. Cognition and Emotion 11, 569583. Hertel, P.T., 2000. The cognitiveinitiative account of depression-related impairments in memory. In: Medin, D.L. (Ed.), The Psychology of Learning and Motivation: Advances in Research and Theory. Academic Press, San Diego, pp. 4771. Iddon, J.L., McKenna, P.J., Sahakian, B.J., Robbins, T.W., 1998. Impaired generation and use of strategy in schizophrenia: evidence from visuospatial and verbal tasks. Psychological Medicine 28, 10491062. Incisa della Rocchetta, A., Milner, B., 1993. Strategic search and retrieval inhibition: the role of the frontal lobes. Neuropsychologia 31, 503524. Jasper, H.H., 1958. The ten-twenty electrode system of the International Federation. Electroencephalography and Clinical Neurophysiology 10, 371375. Lee, A.C.H., Robbins, T.W., Owen, A.M., 2000. Episodic memory meets working memory in frontal lobe: functional-neuroimaging studies of encoding and retrieval. Critical Reviews in Neurobiology 14, 165198. Levy, R., Goldman-Rakic, P.S., 1999. Association of storage and processing functions in the dorsolateral prefrontal cortex of the nonhuman primate. Journal of Neuroscience 19, 51495158. Miller, E.K., Cohen, J.D., 2001. An integrative theory of prefrontal cortex function. Annual Review of Neuroscience 24, 167202. Miller, G.A., Gratton, G., Yee, C.M., 1988. Generalized implementation of an eye movement correction procedure. Psychophysiology 25, 241243. Miller, G.A., Lutzenberger, W., Elbert, T., 1991. The linked-reference issue in EEG and ERP recording. Journal of Psychophysiology 5, 273276. Murphy, F.C., Sahakian, B.J., Rubinsztein, J.S., Michael, A., Rogers, R.D., Robbins, T.W., Paykel, E.S., 1999. Emotional bias and inhibitory control processes in mania and depression. Psychological Medicine 29, 1307 1321. Nitschke, J.B., Heller, W., Miller, G.A., 2000. Anxiety, stress, and cortical brain function. In: Borod, J.C. (Ed.), The Neuropsychology of Emotion. Oxford University Press, New York, pp. 298319. Nitschke, J.B., Heller, W., 2002. The neuropsychology of anxiety disorders: affect, cognition, and neural circuitry. In: DHaenen, H., den Boer, J.A., Willner, P. (Eds.), Biological Psychiatry. Wiley, Chichester, pp. 975988. Nitschke, J.B., Miller, G.A., Cook III, E.W., 1998. Digital ltering in EEG/ERP analysis: some technical and empirical comparisons. Behavior Research Methods, Instruments, and Computers 30, 5467. Nunez, P.L., 1981. Electrical Fields of the Brain. Oxford University Press, Oxford. Oldeld, R.C., 1971. The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia 9, 97113. Owen, A.M., 2003. HERA today, gone tomorrow? Trends in Cognitive Science 7, 383384. Petrides, M., 1996. Specialized systems for the processing of mnemonic information within the primate frontal cortex. Philosophical Transactions: Biological Sciences 351, 14551461. Pillon, B., Deweer, B., Vidailhet, M., Bonnet, A.M., Hahn-Barma, V., Dubois, B., 1998. Is impaired memory for spatial location in Parkinsons disease domain specic or dependent on strategic processes? Neuropsychologia 36, 19. Postle, B.R., Berger, J.S., DEsposito, M., 1999. Functional neuroanatomical double dissociation of mnemonic and executive control processes contributing to working memory performance. Proceedings of the National Academy of Sciences 96, 1295912964. Ranganath, C., Johnson, M.K., DEsposito, M., 2003. Prefrontal activity associated with working memory and episodic long-term memory. Neuropsychologia 41, 378389. Savage, C.R., Baer, L., Keuthen, N.J., Brown, H.D., Rauch, S.L., Jenike, M.A., 1999. Organizational strategies mediate nonverbal memory impairment in obsessive-compulsive disorder. Biological Psychiatry 45, 905916. Schacter, D.L., 1997. The cognitive neuroscience of memory: perspectives from neuroimaging research. Philosophical Transactions: Biological Sciences 352, 16891695. Sengel, R.A., Lovallo, W.R., 1983. Effects of cueing on immediate and recent memory in schizophrenics. Journal of Nervous and Mental Disorders 171, 426430. Shallice, T., 2003. Functional imaging and neuropsychology ndings: how can they be linked? NeuroImage 20, S146S154.
143
Simons, J.S., Spiers, H.J., 2003. Prefrontal and medial temporal lobe interactions in long-term memory. Nature Reviews Neuroscience 4, 637648. Smith, E.E., Jonides, J., 1998. Neuroimaging analyses of human working memory. Proceedings of the National Academy of Sciences 95, 1206112068. Smith, E.E., Jonides, J., 1999. Storage and executive processes in the frontal lobes. Science 283, 16571661. Spielberger, C.D., 1968. Self-evaluation Questionnaire. STAI Form X-2. Consulting Psychologists Press, Palo Alto. Spielberger, C.D., 1983. Manual for the State-Trait Anxiety Inventory (Form Y). Consulting Psychologists Press, Palo Alto. Tulving, E., Kapur, S., Craik, F.I.M., Moscovitch, M., Houle, S., 1994. Hemispheric encoding/retrieval asymmetry in episodic memory: positron emission tomography ndings. Proceedings of the National Academy of Sciences 91, 20162020. Watkins, P.C., 2002. Implicit memory bias in depression. Cognition and Emotion 16, 381402.
Relative left-frontal activity is associated with increased depression in high reassurance-seekers

Jennifer A. Minnix , John P. Kline, Ginette C. Blackhart, Jeremy W. Pettit, Marisol Perez, Thomas E. Joiner
Department of Psychology, Florida State University, Tallahassee, FL 32306-1270, USA
Abstract Excessive reassurance-seeking, which has been associated with depression in many studies, can be dened as the relatively stable tendency to seek assurance perseveratively from others. We hypothesized that although depression has been associated with left-frontal EEG hypoactivity, reassuranceseekers may possess a unique diathesis that is more likely to be associated with increased left-frontal activity. Data were collected from 12 volunteers who were receiving therapeutic services from a University Clinic. EEG asymmetry scores were averaged over two measurement occasions at least 3 weeks apart. As predicted, stable relative right-frontal activity was associated with increased depression in those who were low on reassurance-seeking, while stable relative left-frontal activity was associated with increased depression among high reassurance-seekers. Perhaps those who seek reassurance excessively do so because of their inability to alter their behavior even when environmental cues are no longer reinforcing, which can maintain or exacerbate their depressive symptoms. 2004 Elsevier B.V. All rights reserved.
Keywords: Depression; EEG; Reassurance-seeking; Interpersonal style
1. Introduction Depression research has focused on a range of possible diatheses that run the gamut from the biological to the interpersonal. The present study provides a possible avenue toward a rapprochement between two heretofore-separate literatures on interpersonal and biological vulnerabilities namely, excessive reassurance-seeking and asymmetrical frontal lobe activity. Frijda (1986) contends that emotions are outcomes of the process of assessing the world
Corresponding author. Tel.: +1-850-321-3298. E-mail address: minnix@psy.fsu.edu (J.A. Minnix).
146
J.A. Minnix et al. / Biological Psychology 67 (2004) 145155
in terms of ones own goals that, in turn, modify behavioral tendencies. Because of the importance of the frontal lobes for the expression and regulation of both emotional and social behavior, the integration of these separate lines of research provides a possible means for the explanation of previous discrepancies in the literature on frontal asymmetry/depression relations. 1.1. Reassurance-seeking and depression With Coynes (1976) interactional description of depression as a starting point, Joiner and coworkers have theorized that depression-prone individuals display interpersonal behaviors that increase their vulnerability to future depression (Joiner, 2000; Joiner and Metalsky, 2001). Specically, these individuals are more likely to engage in excessive reassurance-seeking, or perseverative attempts to elicit feedback from others to assure them that they are cared for and loved. Although this information is affectively pleasing, it conicts with the individuals negative self-views and leads them to doubt the veracity of the information (Swann et al., 1987; Joiner et al., 1993; Joiner and Metalsky, 1995). Consequently, this doubt compels them again to seek reassurance from others and results in a downward interpersonal spiral, including interpersonal rejection (Joiner et al., 1992), contagious depression (Joiner, 1994), and exacerbated depression (Joiner and Metalsky, 2001; Joiner et al., 2001). Accordingly, excessive reassurance-seeking can be dened as the relatively stable tendency to seek assurance persistently from others that one is loved, regardless of how many times these assurances have already been provided (Joiner et al., 1999). Studies have shown reassurance-seeking to be a cohesive, valid, and reliable factor that is relatively specic to depression and relatively distinct from other aspects of interpersonal style (such as general dependency or doubt in others sincerity; Joiner and Metalsky, 2001). Multiple studies have demonstrated a reliable association (median value of 0.36) between reassurance-seeking and depression in a variety of samples, including college students, US Air Force cadets, various clinical samples, and women in heterosexual dating relationships (see Joiner et al., 1999; Katz and Joiner, 2001, for a review). At least six studies have demonstrated that excessive reassurance-seeking is a contributory cause to depression (Katz and Joiner, 2001). Studies based on the diathesis-stress model have shown that excessive reassurance-seeking and stress serve as interactive, contributory causes to the later development of depressive symptoms (Joiner and Metalsky, 1995; Katz et al., 1998). Habitual and excessive reassurance-seeking creates many problems in the interpersonal world of depressed individuals (Pothoff et al., 1995). Because reassurance-seeking initially elicits positive feedback from others, it is not inherently a maladaptive behavior. It is only because some depressed individuals continue to solicit this feedback when it is clearly creating difculties within their interpersonal relationships that it becomes problematic. It is possible that the negative consequences of excessive reassurance-seeking are secondary in importance to the immediate relief experienced by the depressed individual when he/she receives this reassurance, particularly if the depressed individual is convinced that this behavior is necessary to provide relief. In these situations, the negative consequences may be dismissed altogether (Schmidt et al., 1999). In essence, these persons neglect to regulate their own behavior in response to external social cues that typically guide social interactions
147
(Hamilton and Deemer, 1999). They fail to adapt their behaviors based on environmental contingencies, which may lead to a perseverative questioning of others that ultimately leads to rejection and may maintain their depressive symptoms (Joiner et al., 1999). Biological factors have also been related to an increased vulnerability for the presence and maintenance of depressive symptoms. 1.2. Frontal asymmetry and depression Asymmetrical activity in the frontal lobes of the brain has been related to emotional and motivational factors, as well as cognitive and behavioral ones. It has been hypothesized that relative left-frontal activity is related to behavioral approach motivation and positive emotion, whereas the right-frontal region is related to behavioral withdrawal motivation and negative emotion (Davidson, 1995; Davidson et al., 2000). Positive, approach-related emotion, is typically associated with relative left-frontal and anterior-temporal activity, while negative, withdrawal-related emotion has been associated with higher relative activity of the right-frontal and anterior-temporal regions (Ahern and Schwartz, 1985; Davidson, 1995; Harmon-Jones and Allen, 1997; Heller, 1990; Tomarken et al., 1990). However, evidence for these relations is not always consistent (Hagemann et al., 1998, 1999). Several studies have found that depression is associated with less left-frontal activity and/or more right-frontal activity (Bell et al., 1998; Gotlib et al., 1998; Henriques and Davidson, 1991; Roemer et al., 1992). Resting anterior asymmetry has been reported to distinguish depressed from non-depressed individuals during both episodes and remission (Allen et al., 1993; Bell et al., 1998; Henriques and Davidson, 1991), although some evidence for this nding is inconsistent. For example, Reid et al. (1998) largely failed to replicate these previous studies nding decreased left-frontal activity in depressed individuals relative to non-depressed controls across two samples, suggesting the need to identify intervening factors. The asymmetry model may have difculty accounting for emotions that potentially involve both withdrawal and approach-related strategies, such as anger or sadness. For example, sadness can evoke an approach strategy geared towards retrieving a lost object as well as a withdrawal strategy reecting the desire to escape from the noxious environment, depending on the context (see Shackman, 2000). As such, there may be factors involved that predict which kinds of strategies individuals may employ. 1.3. Frontal lobe damage and social behavior The frontal lobes of the brain seem to facilitate the ability to navigate a complex social landscape, and are responsible for action selection based on internal and external cues (Kolb and Whishaw, 1995; Kolb and Taylor, 1999). Studies have shown that the frontal cortex plays an important role in the regulation of arousal with respect to behavioral demands (Tucker et al., 1995; Stuss et al., 2000). Because emotions are directly tied into reinforcing and punishing events, a failure to respond to changing environmental cues may lead to inappropriate social and emotional behavior (Rolls et al., 1994). Patients with frontal lobe damage seem to have difculty using feedback from environmental cues to change or regulate their behavior, such that they are impaired at developing novel strategies to deal with problems (Kolb and Whishaw, 1995). Patients with frontal lobe lesions seem to be
148
unable to alter their behavior based on changes in previously learned associations between environmental stimuli and rewards (Rolls et al., 1994). Perseveration errors, or errors resulting from the failure to modify behavior based on changes in environmental cues, are relatively specic to frontal lobe damage. In addition, research suggests that these types of errors may be more common in patients with right-frontal lobe lesions than those with left-frontal lobe lesions (Haut et al., 1996; Stuss et al., 2000). Similar to patients with right-frontal lesions, those who seek reassurance excessively may do so because of an inability to alter their behavior even when environmental cues are no longer reinforcing. However, this is not to imply that those high in reassurance-seeking behaviors also evince frontal lobe damage. Rather, the research involving patients with frontal lobe damage serves to implicate a particular area in the cortex that might reect a tendency towards certain behaviors in normal individuals similar in nature, though not as severe, to those with lesions in the area. Specically, when their reassurance-seeking behaviors fail to provide desired positive feedback, and instead lead to interpersonal rejection, high reassurance-seekers continue to persist in seeking this feedback. This behavior may ultimately lead to rejection of the depressed individual and an increase in depressive symptoms. 1.4. Summary and hypotheses Based on research regarding reassurance-seeking and depression, one might expect that reassurance-seeking will moderate the relation between frontal brain asymmetry and depression. Individuals who seek reassurance excessively may behave in a perseverative manner, much like what is seen in patients with right-frontal lesions. This hypothesis is consistent with the motivational model in that proactively seeking feedback from the environment can be considered an approach behavior. It is possible that reassurance-seeking represents a unique diathesis for developing depressive symptoms, reected in frontal brain wave patterns, that is inherently different from the diathesis usually associated with relative left-frontal hypoactivity. Accordingly, we predicted that reassurance-seeking moderates the relation between depression and frontal asymmetry, such that, in low reassurance-seekers, the usual pattern is found (i.e. relative right-frontal activity is associated with increased depression), but that in high reassurance-seekers, the opposite pattern is found (i.e. relative left-frontal activity is associated with increased depression).
2. Method 2.1. Participants Data were collected from 12 (six males, six females) right-handed volunteers aged 1952 (M = 32, S.D. = 14.47) who were receiving therapeutic services from the Psychology Clinic at Florida State University. Upon application, patients agreed in writing to the research and training nature of the clinic; in addition, they signed a separate and additional informed consent form for this study. Diagnoses were formulated by a trained clinician
149
through a combination of symptom measures and a clinical interview, based on the Diagnostic and Statistical Manual of Mental Disorders (4th edition, 1994). This sample included individuals who experienced substantial depressive symptoms (mean Beck Depression Inventory (BDI) score was 23.5); diagnoses were Dysthymic Disorder (3), Major Depression (2), Social Phobia (3), and Bipolar I Disorder (1), depressed phase. Three individuals had no diagnosable Axis I mood or anxiety disorder. Co-morbid diagnoses included Alcohol Abuse (2) and Bulimia Nervosa (1). All indices were averaged across two measurement occasions approximately 4 weeks apart (M = 4.25, S.D. = 1.42). 2.2. Questionnaire administration and scoring Participants completed all questionnaires prior to EEG recording. Handedness was assessed with the Edinburgh Handedness Inventory (Oldeld, 1971). Level of depressive symptoms was assessed using the Beck Depression Inventory, a 21-item self-report inventory (Beck et al., 1961). Coefcient alphas (Cronbach, 1951) of 0.80 and above have been reported for this measure when used with a variety of clinical and other samples (Beck et al., 1988). A subscale of the Depressive Interpersonal Relationships Inventory (DIRI-RS; Joiner et al., 1992) was used to assess excessive reassurance-seeking. This four-item subscale (DIRI-RS) measures the tendency to seek reassurance excessively from others as to whether they truly care. Each item is rated on a seven-point Likert scale and averaged across items. Coefcient alpha of 0.88 has been reported for this subscale (Joiner et al., 1992). Scores from the BDI and DIRI-RS were averaged across both measurement sessions for each participant. 2.3. EEG recording and quantication Resting EEG was recorded with a stretch-lycra electrode cap (Electrocap International, Eaton, OH) from a standard International 1020 system of 19 channels referenced to linked ears. Data collection and reduction were accomplished with Neurolex software, and a Neurosearch 24C (Lexicor, Boulder, CO). Electrolytic gel was applied and each site gently abraded until impedances were below 10 k . Homologous leads (EEG and auricular leads) were within 1 k of each other. EEG records were visually examined for artifact by a trained technician, and epochs containing bioelectric artifact in any channel were rejected. Data were digitized on-line at 256 Hz (60 Hz notch lter enabled) during 60-s baselines: three with eyes open and three with eyes closed. High-pass lters were set at 2 Hz, and had a rolloff of 6 dB per octave (rst order), while low-pass lters set at 64 Hz had a rolloff of 48 dB per octave (eighth order). Artifact-free epochs of 256 samples (i.e. 1-s epochs for 256 Hz) were extracted through a Hanning window and submitted to fast Fourier transform. Average alpha (813 Hz) power (microvolts squared) was natural log transformed for normalization, as these power values tend to be positively skewed. Asymmetry scores were computed (log[right] log[left]), with higher values indicating relative left-frontal activity, for mid-frontal (F4-F3), frontal pole (Fp2-Fp1), lateral-frontal (F8-F7), central (C4-C3), anterior-temporal (T4-T3), posterior-temporal (T6-T5), parietal (P4-P3), and occipital (O2-O1) homologous pairs.
150
2.4. Design and analysis Reassurance-seeking scores (DIRI), BDI scores, and asymmetry scores were averaged across both sessions. A general linear model analysis was used to test the hypothesis that reassurance-seeking scores would moderate the relation between depression and anterior asymmetry, using continuous measures of DIRI, BDI, and asymmetry. As a follow-up procedure, a BDI DIRI analysis of variance (ANOVA) was conducted for each of the anterior sites found to be signicantly related to BDI and DIRI in the general linear model analysis. High and low BDI and DIRI groups were formed for the ANOVA analysis using a median split. To assess specicity to the frontal region, posterior sites were also tested using a general linear model analysis with continuous measures. Because of the small sample size and lack of power, a multivariate analysis of variance (MANOVA) would have provided an unacceptable Type II error rate.
3. Results Means and standard deviations for each of the dependent measures are as follows. F4-F3 time 1: 0.03 (0.14), F4-F3 time 2: 0.02 (0.20), F8-F7 time 1: 0.09 (0.23), and F8-F7 time 2: 0.19 (0.10). Descriptive statistics for the independent measures are presented in Table 1. BDI and DIRI were not signicantly correlated in this sample, although the magnitude and direction of the observed relation in the depressed individuals was comparable to that which has been observed in previous samples (r = 0.31), suggesting that the present sample was too small to have the power to detect this relation. Continuous analyses revealed that main effects of BDI, F(1, 8) = 7.55, P < 0.05, and DIRI, F(1, 8) = 12.47, P < 0.01 were found at F4-F3 as well as F8-F7: BDI, F(1, 8) = 10.42, P < 0.05; DIRI, F(1, 8) = 12.46, P = 0.01. A signicant BDI DIRI interaction emerged for the two frontal sites: F4-F3, F(1, 8) = 11.45, P = 0.01 and F8-F7, F(1, 8) = 16.83, P < 0.01. As predicted, there were no main or interaction effects for C4-C3, F(1, 8) = 1.65, P > 0.2; T4-T3, F(1, 8) = 3.65, P = 0.09; T6-T5, F(1, 8) = 0.196, P > 0.6; or P4-P3, F(1, 8) = 0.073, P > 0.7, suggesting that the effects found were specic to frontal sites. In addition, no signicant difference in asymmetry as a function of diagnostic group was found (i.e. Mood Disorder, Anxiety Disorder, or Other) at F4-F3, F(2, 9) = 0.89, P > 0.4; F8-F7, F(2, 9) = 1.54, P > 0.2; or T4-T3, F(2, 9) = 4.19, P < 0.06. In order to investigate the nature of the interaction, participants were divided into high and low BDI groups (high > 24, M = 27.8; low < 24, M = 12.8) and high and low
Table 1 Distributions of independent measures across both sessions M BDI DIRI 23.5 11.0 S.D. 13.3 6.2 Range 38 18 Skewness 0.023 0.60 Kurtosis 1.3 0.96
151
Fig. 1. Interactive effects of high/low depressive symptoms (BDI) and high/low reassurance-seeking (DIRI) on frontal asymmetry at mid-frontal and lateral-frontal sites.
DIRI groups (high > 9, M = 15.7; low > 9, M = 10) using median splits. Those individuals with high DIRI scores and high BDI scores were more likely to be relatively left-frontally active at F8-F7 and F4-F3, while those with high BDI scores and low DIRI scores were more likely to be relatively right-frontally active F(2, 7) = 16.20, P = 0.001) (see Fig. 1).
152
4. Discussion As was predicted, results showed that reassurance-seeking moderated the relation between depressive symptoms and anterior asymmetry at two anterior sites: mid-frontal and lateral-frontal. This relationship was such that depressive symptoms were related to less left relative to right-frontal activity in low reassurance-seekers, while depression was related to greater left relative to right-anterior activity in high reassurance-seekers. These results suggest that reassurance-seeking and relative left-frontal hypoactivation may constitute separate diatheses for developing depressive symptoms in a clinic population, as both have been reliably related to depression in the past (Gotlib et al., 1998; Joiner and Metalsky, 2001). Lane et al. (1999) (p. 3) suggest that emotion can be conceptualized as the outcome of an evaluation of the extent to which ones goals are being met in interaction with the environment. Individuals who engage in reassurance-seeking presumably take an active role in the maintenance of their depressive symptoms, such that they are motivated by their desire for assurance to interact with the environment in ways that eventually backre.Studies have shown that some depressed individuals generate distress within their environments (Davila et al., 1995; Hammen, 1991) that can exacerbate their symptoms. Specically, studies suggest that many of the interpersonal stressors in the lives of depressed individuals are not accidental events, but rather, self-generated consequences of a depressive lifestyle (Pothoff et al., 1995). Similar behavior patterns, though to a much greater degree, have been documented in patients who have lesions in their right-frontal lobes. These patients often fail to respond appropriately to environmental demands (Stuss et al., 2000). In addition, this behavior is consistent with the motivational model of frontal asymmetry, which relates approach behaviors to relative left-frontal activation and withdrawal behaviors to right-frontal activation (see Davidson, 1995). Reassurance-seeking can be considered an approach behavior, which is consistent with the current ndings that found a relation between depression and left-frontal activation in these individuals. These individuals may represent a potential caveat of the Davidson (1995) model, namely, that left-frontal activity may not be benecial in every context. Here, if relative left-frontal activation reects the approach motivation associated with reassurance-seeking, it can actually enable folks to facilitate the maintenance of their own depressive symptomatology. Though these results suggest an interesting relation between frontal asymmetry, depression, and reassurance-seeking, the nature of the interactions between these variables needs further exploration. In this study, baseline EEG was recorded using a linked-ears montage, though some research suggests that this strategy may be problematic (Hagemann et al., 2001). However, other studies have reported that linked-ear references made no signicant difference in measured asymmetries (Senulis and Davidson, 1989). In addition, the measurement of electrical activity from the scalp does not allow for the kind of spatial resolution that is necessary to make distinctions between frontal subregions. The current sample was taken from a clinic population and measured across two separate occasions separated by at least 3 weeks. However, the sample size was small and included multiple diagnostic categories (although the average depressive symptom score was substantial). Accordingly, the participants also differed in their individual medication status, both in type and amount prescribed. In addition, the participants varied in the amount of therapy they had received, both from the clinic and in the past. Because of the small sample size,
153
we lacked the statistical power needed to conduct a more comprehensive analysis of the boundary conditions of the observed effect. Because there was no comparison group, the results do not provide enough information to conclude that the effects found in the study are specic to individuals experiencing relatively serious psychopathological symptoms. This study represents a modest test of a potentially important concept. Though the results should be considered with appropriate caution, they serve as a starting point for interesting future research and a better understanding of the complex relationships between biological function, social behavior, and psychopathology. Despite these points of consideration, this study provides a potential extension to current theories of frontal asymmetry and depression by framing these relations from within an interpersonal context. Future research might examine such interpersonal moderators of frontal asymmetry/emotion relations in non-clinical samples, as well as within more homogeneous diagnostic categories within clinical populations. Dismantling the nature of these relations may help to explain some past inconsistencies in the frontal asymmetry/depression literature (Reid et al., 1998), as well as elucidate further some of the multiple factors that contribute to depression and its harmful effects on the lives of many.
References
Ahern, G.L., Schwartz, G.E., 1985. Differential lateralization for positive and negative emotion in the human brain: EEG spectral analysis. Neuropsychologia 23, 745755. Allen, J.J., Iacono, W.G., Depue, R.A., Arbisi, P., 1993. Regional electroencephalographic asymmetries in bipolar seasonal affective disorder before and after exposure to bright light. Biological Psychiatry 33, 642646. American Psychiatric Asscociation, 1994. Diagnostic and Statistical Manual of Mental Disorders, fourth ed. Washington, D.C. Beck, A.T., Steer, R., Garbin, M., 1988. Psychometric properties of the Beck Depression Inventory: 25 years of evaluation. Clinical Psychology Review 8, 77100. Beck, A.T., Ward, C.H., Mendelson, M., Mock, J., Erbaugh, J., 1961. An inventory for measuring depression. Archives of General Psychiatry 4, 561571. Bell, I.R., Schwartz, G.E., Hardin, E.E., Baldwin, C.M., Kline, J.P., 1998. Differential resting quantitative electroencephalographic alpha patterns in women with environmental chemical intolerance, depressives, and normals. Biological Psychiatry 43, 376388. Coyne, J.C., 1976. Toward an interactional description of depression. Psychiatry 39, 2840. Cronbach, L., 1951. Coefcient alpha and the internal structure of tests. Psychometrika 16, 297334. Davidson, R.J., 1995. Cerebral asymmetry, emotion, and affective style. In: Davidson, R.J., Hugdahl, K. (Eds.), Brain Asymmetry. MIT Press, Cambridge, MA, pp. 361387. Davidson, R.J., Jackson, D.C., Kalin, N.H., 2000. Emotion, plasticity, context, and regulation: perspectives from affective neuroscience. Psychological Bulletin 126, 890909. Davila, J., Hammen, C., Burge, D., Paley, B., Daley, S., 1995. Poor interpersonal problem-solving as a mechanism of stress generation in depression among adolescent women. Journal of Abnormal Psychology 104, 592600. Frijda, N.H., 1986. The Emotions. Cambridge, New York. Gotlib, I.H., Ranganath, C., Rosenfeld, J.P., 1998. Frontal EEG alpha asymmetry, depression, and cognitive functioning. Cognition and Emotion 12, 449478. Hagemann, D., Naumann, E., Becker, G., Maier, S., Bartussek, D., 1998. Frontal brain asymmetry and affective style: a conceptual replication. Psychophysiology 35, 372388. Hagemann, D., Naumann, E., Lurken, A., Becker, G., Maier, S., Bartussek, D., 1999. EEG asymmetry, dispositional mood, and personality. Personality and Individual Differences 27, 541568. Hagemann, D., Naumann, E., Thayer, J.F., 2001. The quest for the EEG reference revisited: a glance from brain asymmetry research. Psychophysiology 38, 847857.
154
Hamilton, J.C., Deemer, H.N., 1999. Excessive reassurance-seeking as self-regulatory preservation: implications for explaining the relation between depression and illness behavior. Psychological Inquiry 10, 293 297. Hammen, C., 1991. The generation of stress in the course of unipolar depression. Journal of Abnormal Psychology 100, 555561. Harmon-Jones, E., Allen, J.J.B., 1997. Behavioral activation sensitivity and resting frontal EEG asymmetry: covariation of putative indicators related to risk for mood disorders. Journal of Abnormal Psychology 106, 159163. Haut, M.W., Cahill, J., Cutlip, W.D., Stevenson, J.M., Makela, E.H., Bloomeld, S.M., 1996. On the nature of the Wisconsin Card Sorting Test performance in schizophrenia. Psychiatry Research 65, 1522. Heller, W., 1990. The neuropsychology of emotion: developmental patterns and implications for psychopathology. In: Stein, N., Leventhal, B.L., Trabasso, T. (Eds.), Psychological and Biological Approaches to Emotion. Erlbaum, Hillsdale, NJ, pp. 67211. Henriques, J.B., Davidson, R.J., 1991. Regional brain electrical asymmetries discriminate between previously depressed and healthy control subjects. Journal of Abnormal Psychology 99, 2231. Joiner Jr., T.E., 1994. Contagious depression: existence, specicity to depressive symptoms, and the role of reassurance-seeking. Journal of Personality and Social Psychology 67, 287296. Joiner Jr., T.E., 2000. A test of the hopelessness theory of depression among youth psychiatric inpatients. Journal of Abnormal Psychology 29, 167176. Joiner Jr., T.E., Alfano, M.S., Metalsky, G.I., 1992. When depression breeds contempt: reassurance-seeking, self-esteem, and rejection of depressed college students by their roommates. Journal of Abnormal Psychology 101, 165173. Joiner Jr., T.E., Alfano, M.S., Metalsky, G.I., 1993. Caught in the crossre: depression, self-verication, self-enhancement, and the response of others. Journal of Social and Clinical Psychology 12, 113 134. Joiner Jr., T.E., Metalsky, G.I., 1995. A prospective test of an integrative interpersonal theory of depression: a naturalistic study of college roommates. Journal of Personality and Social Psychology 69, 778788. Joiner Jr., T.E., Metalsky, G.I., 2001. Excessive reassurance-seeking: delineating a risk factor involved in the development of depressive symptoms. Psychological Science 12, 371378. Joiner Jr., T.E., Metalsky, G.I., Gencoz, F., Gencoz, T., 2001. The relative specicity of excessive reassurance-seeking to depressive symptoms and diagnoses among clinical samples of adults and youth. Journal of Psychopathology and Behavioral Assessment 23, 3541. Joiner Jr., T.E., Metalsky, G.I., Katz, J., Beach, S.R.H., 1999. Depression and excessive reassurance-seeking. Psychological Inquiry 10, 269278. Katz, J., Beach, S.R.H., Joiner, T.E., 1998. When does partner devaluation predict depression? Prospective moderating effects of reassurance-seeking and self esteem. Personal Relationships 5, 409421. Katz, J., Joiner, T.E., 2001. The aversive interpersonal context of depression: Emerging perspectives on depressotypic behavior. In: R.M. Kowalski (Ed.), Behaving Badly: Aversive behaviors in Interpersonal Relationships. American Psychological Association, Washington D.C., pp. 117147. Kolb, B., Taylor, L., 1999. Facial expression, emotion, and hemispheric organization. In: Lane, R.D., Nadel, L. (Eds.), Cognitive Neuroscience of Emotion. Oxford University Press, New York, pp. 6383. Kolb, B., Whishaw, I.Q., 1995. Fundamentals of Human Neuropsychology, fourth ed. W.H. Freeman and Company, New York. Lane, R.D., Nadel, L., Allen, J.J.B., Kaszniak, A.W., 1999. The study of emotion from the perspective of cognitive neuroscience. In: Lane, R.D., Nadel, L. (Eds.), Cognitive Neuroscience of Emotion. Oxford University Press, New York, pp. 361. Pothoff, J.G., Holahan, C.J., Joiner Jr., T.E., 1995. Journal of Personality and Social Psychology 68, 664670. Reid, S.A., Duke, L.M., Allen, J.J.B., 1998. Resting frontal electroencephalographic asymmetry in depression: inconsistencies suggest the need to identify mediating factors. Psychophysiology 35, 389404. Roemer, R.A., Shagass, C., Dubin, W., Jaffe, R., Siegal, L., 1992. Quantitative EEG in elderly depressives. Brain Topography 4, 285290. Rolls, E.T., Hornak, J., Wade, D., McGrath, J., 1994. Emotion-related learning in patients with social and emotional changes associated with frontal lobe damage. Journal of Neurology, Neurosurgery, and Psychiatry 57, 1518 1534.
155
Schmidt, N.B., Schmidt, K.L., Young, J.E., 1999. Schematic and interpersonal conceptualizations of depression: an integration. In: Joiner, T., Coyne, J.C. (Eds.), The Interactional Nature of Depression. American Psychological Association, Washington, D.C., pp. 127148. Senulis, J.A., Davidson, R.J., 1989. The effects of linking the ears on hemispheric asymmetry of the EEG. Psychophysiology 26, S54. Shackman, A.J., 2000. Anterior cerebral asymmetry, affect, and psychopathology: commentary on the withdrawal-approach model. In: Davidson, R.J. (Ed.), Anxiety, Depression, and Emotion. Oxford University Press, New York, NY, pp. 109132. Swann, W.B., Grifn, J.J., Predmore, S., Gaines, B., 1987. The cognitive-affective crossre: when self-consistency confronts self-enhancement. Journal of Personality and Social Psychology 52, 881889. Stuss, D.T., Levine, B., Alexander, M.P., Hong, J., Palumbo, C., Hamer, L., Murphy, K.J., Izukawa, D., 2000. Wisconsin Card Sorting Test performance in patients with focal frontal and posterior brain damage: effects of lesion location and test structure on separable cognitive processes. Neuropsychologia 38, 388402. Tomarken, A.J., Davidson, R.J., Henriques, J.B., 1990. Resting frontal brain asymmetry predicts affective responses to lms. Journal of Personality and Social Psychology 59, 791801. Tucker, D.M., Luu, P., Pribram, K.H., 1995. Social and emotional self-regulation. In: Grafman, J., Holyoak, K.J. (Eds.), Structure and Functions of the Human Prefrontal Cortex. Annals of the New York Academy of Sciences, vol. 769. New York, pp. 213239.
Methodology
Individual differences in anterior EEG asymmetry: methodological problems and solutions

Dirk Hagemann
Department of Psychology, Fachbereich IPsychologie, Universitt Trier, Universittsring 15, 54286 Trier, Germany
Abstract In the last two decades, a substantive body of empirical work investigated the association between individual differences of electroencephalogram (EEG) alpha asymmetry and affective/motivational dispositions. Recent work indicated that several methodological problems persist in this eld. The present paper reviews these problems with a focus on the reliability and validity of measures of anterior resting EEG asymmetry, which serve as a proxy for trait-like asymmetries of cortical activity. These issues include the treatment of ocular and muscle artifacts, the choice of the EEG reference, the use of current source density (CSD) measures, the statetrait nature of resting asymmetry, and the treatment of state-like uctuations of the measures. In addition, the statistical problem of inated type-I error due to multiple testing is also considered and different approaches to counteract this problem are suggested. 2004 Elsevier B.V. All rights reserved.
Keywords: Frontal EEG asymmetry; Emotion; Methodology
1. Introduction More than two decades ago, several researchers started to investigate electroencephalogram (EEG) asymmetry during emotional states (Davidson and Fox, 1982; Davidson and Schwartz, 1976; Harman and Ray, 1977; Tucker et al., 1981). A few years later, some studies began to investigate the relationship between individual differences in resting asymmetry and affective dispositions (Davidson and Fox, 1989; Schaffer et al., 1983). This research led to the proposal that frontal EEG asymmetry in a resting state reects a trait-like asymmetry of anterior cortical activity, which acts as a diathesis for emotion-elicitors and is a contrib
Tel.: +49-651-201-2896; fax: +49-651-201-3956. E-mail address: hagemann@uni-trier.de (D. Hagemann).
158
D. Hagemann / Biological Psychology 67 (2004) 157182
utory factor for individual differences in affective/motivational behaviors (e.g. Davidson, 1992, 1998b). This hypothesis stimulated empirical research in several laboratories. In the typical study, the spontaneous EEG in a resting state is recorded, the difference of alpha activity at contralateral homologous sites serves as a measure of anterior asymmetry of cortical activity, facets of affective style are measured with experimental procedures or questionnaires, and the statistical analysis is focused on the associations between asymmetry and affect (e.g. Hagemann et al., 1999; Harmon-Jones and Allen, 1998; Kline et al., 2001; Nitschke et al., 1999; Tomarken et al., 1990). Many ndings supported the hypothesis (for recent reviews, see Coan and Allen, 2003b; Davidson, 1998a; Hagemann et al., 2002), but several authors noted substantial inconsistencies in the evidence (Hagemann et al., 1998; Heller and Nitschke, 1998; Reid et al., 1998). Recently, this situation stimulated a search for methodological factors that may contribute to the divergence of ndings (Blackhart et al., 2002; Davidson, 1998b; Hagemann et al., 1998; Reid et al., 1998). The present review gives an overview of some of the methodological obstacles that persist in this eld of research and will suggest some prospective solutions.
2. EEG recording The rst step in each study is the recording of the resting EEG. In addition to the general issues like appropriate electrode placement, lter settings, AD rate, and the like (for recommendations, see Pivik et al., 1993), one important factor may be the length of the recording in order to obtain reliable asymmetry measures. Some studies showed that alpha power at single sites shows acceptable reliability estimates even for EEG segments as short as 20 or 30 s (Gasser et al., 1985, reported annual retest correlations of 0.70/0.80 for EEG segments of 20 s length; Hagemann et al., 1998, reported parallel test correlations of 0.80/0.90 for EEG segments of 30 s length). However, this observation may not generalize to asymmetry measures, because difference scores are less reliable than their constituent scores (especially if the constituent scores are correlated; see Cohen and Cohen, 1983). Several studies investigated the reliability of EEG asymmetry in a resting state. Hagemann et al. (1998) analyzed the effect of resting EEG length on the reliability of asymmetry measures. They reported that the reliability was low for a database of 30 s (with parallel test correlations of about 0.50), good for a database of 4 min (with Cronbachs alpha in a range of 0.80/0.90), and excellent for a database of 8 min (with most Cronbachs alphas in the 0.90 s). Other studies reported on similar reliability estimates for a database of 8 min (Reid et al., 1998; Sutton and Davidson, 1997; Tomarken et al., 1992). In a recent study, Hagemann et al. (2002) reported on an excellent reliability for a database of 12 min (estimated with a combined parallel and retest approach via structural equation modeling (SEM), reliability coefcients were close to 1). To sum up, good reliability of the asymmetry measure in a resting state can only be achieved if the EEG is recorded for a sufcient length in time (e.g. at least 4 min). However, it is not known at present if these reliability estimates, and the respective recommendation for EEG recording, can be generalized to affective and motivational states. Another factor that might be suspected to affect the asymmetry measures is the instruction to keep the eyes open or closed while the resting EEG is recorded. However, several studies
159
could consistently demonstrate excellent internal consistencies of EEG baselines that were recorded with subjects keeping their eyes open and closed, irrespective of EEG references such as Cz, A1 + A2, and average reference (AR) (Hagemann et al., 1998; Reid et al., 1998; Sutton and Davidson, 1997; Tomarken et al., 1992). Thus, this procedural factor is not likely to affect individual differences of resting asymmetry in a substantive size.
3. EEG artifacts After the recording of the EEG, the next analysis step is usually the rejection of/correction for artifacts in the EEG to ensure a compelling interpretation of the data (e.g. Gevins, 1987). Several technical artifacts (e.g. the electrode-pop artifact, lead sway artifact, electro smog) may be effectively prevented with adequate care in the EEG laboratory (e.g. use of appropriate electrode/electrolyte combinations, xation of the leads, shielding of the subject; see Barlow, 1986).1 Biological artifacts and in particular ocular and muscle artifacts, however, may pose a serious problem. Spontaneous movements of the eyes and blinks generate potentials that are conducted to the scalp where they may produce large distortions in the EEG, which are the ocular artifacts (see Elbert et al., 1985 for details). Unfortunately, these distorting effects are greatest at the anterior sites, which is the region of interest in frontal asymmetry research. In a recent study, Hagemann and Naumann (2001) analyzed the effects of ocular activity on the EOG and the (lateralized) EEG alpha activity in a resting state while subjects were allowed for spontaneous eye movements. One crucial nding was that in the EOG, blinks and eye movements may generate substantial power in the alpha and even the beta range. However, if no blinks and eye movements are present, then alpha power in the EOG appears to be primarily of neural origin. Moreover, there were signicant general effects of ocular artifacts on anterior EEG alpha activity (i.e. mean shifts due to artifacts). Ocular activity produced a bilateral increase of mean anterior EEG alpha, which was accompanied by asymmetrical distortions, in particular at frontopolar and midfrontal sites where the effect sizes were largest. However, there were almost no differential effects of ocular artifacts (i.e. no shifts of individuals in the sample distribution of asymmetry due to artifacts, with the exception of frontopolar asymmetry). For example, midfrontal asymmetry measures that were based on data portions that were selected for the absence of ocular artifacts (artifact-free EEG) or that were not screened for ocular artifacts (uncorrected EEG) showed correlations of 0.82 and 0.95 (for eyes closed and eyes open, respectively). Taken together, ocular activity may generate substantial power in the alpha range, which may distort mean anterior EEG alpha asymmetry, and thus, may mimic general asymmetries of cortical activity; however, ocular artifacts hardly effect measures of individual differences of alpha asymmetry.
1 Electrode-pop artifacts result from sudden changes in the DC potential between electrode and skin, and have a typical waveform with a sharp rise and an exponential decay (similar to a calibration waveform). Lead sway artifacts result from movements of the leads that connect the electrodes and the (pre-) amplier, and show slow waves. Electro smog refers to electromagnetic pollution due to AC power cables, lights, transformers, radio and television transmitters, and cellular phones, and may result in peak-like artifacts in the power spectrum. For details on technical artifacts, see Barlow (1986), and the literature cited there.
160
These ndings may have some implications for the treatment of ocular artifacts in asymmetry research. One common procedure to control for ocular artifacts is the rejection of EEG segments that show contamination with ocular activity (for an overview, see Barlow, 1986). If the identication of the artifacts is based on visual inspection, the method has the disadvantages that the criteria to classify the artifacts are somewhat subjective and that the procedure is time consuming. However, one advantage of this approach is that other types of artifacts (muscle, drift, technical artifacts) may be readily rejected as well. If the identication of the artifacts is done in an automatic fashion based on quantitative thresholds (such as amplitude or variance criteria of the EOG), however, the setting of thresholds is somewhat arbitrary, different subjects may show different artifact amplitudes, and thus, may need individually matched thresholds, and other types of artifacts may be missed. On the other side, this method can be very time effective. Taken together, the problem of identifying artifacts with subjective criteria or quantitative thresholds may constitute one major disadvantage of the rejection approach, and both methods have the additional disadvantage that contaminated data are lost, small artifacts may be missed, and that the accepted data portions are biased towards a brain state that is associated with low visual activity. However, the ndings of Hagemann and Naumann (2001) suggest that these limitations may be of lesser concern for resting asymmetry research. As noted above, there were hardly differential effects of ocular artifacts on the alpha asymmetry in a resting state, which suggests that even considerable differences in the criteria used to identify the artifacts would result in psychometrically parallel asymmetry measures. Thus, the specication of rejection criteria and even minor ocular artifacts would hardly affect the outcome of a study. In another approach, ocular activity is removed from the EEG with a regression procedure (for a general overview, see Gratton, 1998). In principle, the actual EEG time series is regressed on the EOG time series, and the resulting residual time series represent a new EEG from which the inuence of the ocular activity is statistically removed (for a review of several types of this approach, see Brunia et al., 1989). This method has the great advantages that is does not rely on subjective criteria or arbitrarily selected thresholds for rejection, it does not discard EEG data in particular if ocular activity is induced by the very task of interest (such as in visual tasks), and it can be performed automatically, and is thus, time efcient. However, one major disadvantage of the method is that its validity depends crucially on the assumption that there is no brain activity in the EOG data, or otherwise these procedures would give biased results, and thus, introduce a secondary artifact (e.g. Berg, 1989; Berg and Scherg, 1994; Gratton and Coles, 1989; Pham, 1989). Unfortunately, there is ample evidence for the presence of neural alpha activity in the EOG, in particular when gross eye movements and blinks are absent. For example when analyzing data portions that were selected for the absence of ocular activity, Hagemann and Naumann (2001) observed a protuberance in the EOG alpha band that corresponded to the EEG alpha peak (this nding corroborates results of Iacono and Lykken, 1981, who demonstrated that alpha activity of neural origin can be seen in the EOG). Moreover, there was a correlation between EOG alpha power and occipital alpha power of r = 0.42, which also suggests a neural origin of the EOG alpha protuberance (this nding corroborates results of Gasser et al., 1985; Mcks et al., 1989; van Driel et al., 1989, and others; see Hagemann and Naumann, 2001). Therefore, the regression approach of artifact removal is
161
in the danger of removing also anterior alpha activity, which is of course the very activity of interest.2 In a very recent study, Barz (2003) investigated whether different methods for the removal of ocular artifacts would yield different topographies of alpha activity in a resting state. In particular, she compared the rejection procedure and the Gratton and Coles (1989) regression method, and observed a signicant reduction of anterior alpha due to the regression. This nding is consistent with the presumed disadvantage of the latter method. However, this decrease of alpha did not effect individual differences in frontal resting asymmetry. The correlation of midfrontal asymmetry measures between both methods was r = 0.97. This preliminary nding suggests that both methods might be used interchangeably in research of frontal brain asymmetry, and indicates that the specic disadvantages of both methods might be unlikely to have implications of practical signicance. A third method for the removal of ocular activity from the EEG is the multiple source eye correction (MSEC) method of Berg and Scherg (1994). This approach rests on a modeling of the EEG in terms of multiple overlapping electric sources within the head. In particular, the method decomposes the EEG activity into two distinct source classes. First, ocular sources describe the topography of eye activity; these sources may be estimated empirically from calibration trials. Second, brain sources account for the EEG components which are due to neural activity; these sources may be estimated with a dipole modeling approach. Since the cortex is anatomically organized as a highly convoluted layer of billions of neurons, the activity of cortical neurons that are distributed over wide areas results in a dipole layer of complex geometry. This dipole layer induces electric currents that nally give rise to the surface EEG (Nunez, 1981). The dipole modeling approach aims to explain the surface EEG in terms of brain dipole sources. Because the true dipole layer is too complex in geometry, it must be replaced by a manageable set of point dipoles such that these dipoles would give rise to a surface EEG that is similar to the observed data. Thus, this approach explains the EEG in terms of hypothetical dipole sources, similar to factor analysis which explains the covariances of an item set in terms of factor components, and similar to structural equation modeling which explains the covariance structure of measured variables in terms of latent variables. If previous knowledge allows for an a priori location of the dipoles in specic brain areas, then these dipoles may be denoted as anatomically correct (e.g. it is known that the auditory N100 is generated in the temporal lobes; see Scherg and von Cramon, 1985). However, if such prior knowledge does not exist or if the signal of interest is not generated by circumscribed brain areas (as it may be the case for spontaneous resting EEG), then these dipoles may be placed in strategic brain regions (not necessarily the ones that generate the EEG signal) and thus, serve only as surrogates for the real dipole layer. The practical implementation of this method for the correction of resting EEG would consist of several steps. First, EOG and EEG are recorded while the subject performs ocular
2 It may be noted that Gasser et al. (1992) suggested a modied regression approach to counteract this problem. They proposed a ltering of the EOG with a low-pass of 7.5 Hz before the application of the regression procedure. Although this ltering may prevent the distortion of EEG alpha due to neural activity in the EOG, such a ltering would also remove all ocular activity in the alpha band from the EOG, which in turn may prevent an effective removal of ocular activity in this frequency band from the EEG.
162
calibration movements. These data allows to estimate the ocular source of electric activity on the head in an empirical (i.e. data-driven) fashion. In a second step, EOG and EEG are recorded in the resting situation. These data are decomposed into the previously identied ocular sources plus a set of surrogate dipoles within the head that account for brain activity. In a typical application of this method, four surrogate dipoles are introduced (each one for the anterior and posterior brain compartment of each hemisphere; see Berg and Scherg, 1994). Finally, a forward calculation of the brain surrogate dipoles (excluding the ocular sources) yields an EEG with all ocular activity removed. Although this method overcomes the limitations of the rejection and the regression approach, its accuracy is limited by several factors. For example the optimal number and location of brain surrogate dipoles is unknown, there is a dissimilarity between voluntary ocular calibration movements and involuntary ocular activity, and the inaccuracy of electrode placement may increase modeling errors. Moreover, inherent in this method is that any EEG modeling is only approximate. Another problem is that the application of this method is not only time-consuming but also demanding in terms of dipole modeling expertise and the availability of the required software. At the time of this writing, the MSEC method appears to be a viable alternative to the more traditional approaches, but its advantage for the measurement of anterior EEG asymmetry is not established empirically yet.3 A nal note should address another approach to minimize ocular artifacts, which is to instruct the subjects not to blink and not to move their eyes but to xate a point. This approach may be problematic, because blinks and spontaneous eye movements are controlled by several brain systems in a highly automatic fashion (e.g. Brodal, 1992), and the instruction to suppress these systems may act as a secondary task (see Verleger, 1991, for a discussion). For example there is evidence that the instruction to refrain from blinking affects the auditory P3 and N1 amplitude (Verleger, 1991), and that a xation of the eyes affects the contingent negative variation (Weerts and Lang, 1973). With respect to frontal asymmetry research, it is important to note that the frontal eye eld is of particular importance for voluntary saccades (e.g. Brodal, 1992), and has been implied as part of a neural subsystem for visuospatial selective attention and attention shifting (Chelazzi and Corbetta, 2000; Kosslyn and Thompson, 2000; see also Robertson and Rafael, 2000, for the effects of frontal eye eld lesions on visual attention). Therefore, the instruction not to move the eyes may result in unwanted change of activity in the frontal eye elds, which may interfere with or override the anterior resting asymmetry. Another type of biological artifact is due to pericranial muscle activity. The largest part of power of the surface EMG lies in a range between 10 and 200 Hz and shows a maximal energy between 20 and 100 Hz (Hayes, 1960; van Boxtel, 2001; van Boxtel et al., 1984), although ring rates of motor units of the pericranial muscles may produce considerable activity in the lower frequency range down to 10 Hz (Basmajian and De Luca, 1985). Accordingly, contraction of pericranial muscles may result in an appreciable increase of EEG power in the alpha range for sites that are directly located over the contracting muscle
3 In an unpublished diploma thesis of my group, Klingmann (2003) compared the event-related potentials of affective slides after correcting the data with a regression procedure (Gratton and Coles, 1989) and the MSEC (Berg and Scherg, 1994). There was virtually no difference in the evoked potentials between both methods, suggesting that the regression approach might be an economic but valid alternative to the costly MSEC at least in this paradigm.
163
(ODonnell et al., 1974; see also Gotman et al., 1981; Friedman and Thayer, 1991). Because facial muscle activity tends to be asymmetric (e.g. Borod et al., 1997, 1998), such muscle activity (e.g. of the frontalis or master/temporalis muscles) may mimic anterior EEG alpha asymmetry (e.g. at frontopolar or anterior temporal sites). A low-pass ltering of the EEG would naturally not remove the contamination in the lower frequency range (such as alpha) and cannot be recommended as a tool for muscle artifact reduction. However, two approaches may counteract this contamination. First, EEG segments that show contamination with muscle artifacts may be rejected from analysis. Like the rejection procedure for ocular artifacts, this approach has the disadvantage that dening the subjective criteria or numerical thresholds for the identication of artifacts is difcult, that data portions are lost, small artifacts may be missed, and that the accepted data portions are biased towards a brain state that coincides with low muscular activity (and thus, might be biased towards periods of lesser emotion). A second approach is based on a regression procedure (Davidson, 1988; Henriques and Davidson, 1990, 1991). In a rst step, the power of a high-frequency band between 70 and 80 Hz is extracted from the data of each channel, which presumably reects muscular but not neural activity. In a second step, the EMG power measure may be introduced into the statistical analysis as a covariate. In its most simple form, all statistical analyses of alpha activity may be rerun with the EMG measure; a null nding of the latter analysis would eliminate the possibility that signicant ndings of the former analysis are due to muscle activity. Alternatively, the EMG measure may be introduced as a covariate in an analysis of covariance (ANCOVA) of non-residualized alpha activity. More generally, simple regressions of EEG alpha on the EMG measure would yield a residual that represents alpha activity from which systematic muscle activity is statistically removed. This residual may be introduced into any further analysis, or used for the computation of asymmetry metrics. Although this approach avoids all disadvantages of the rejection procedure, it rests on the assumptions that (1) the EMG measure does not reect neural activity but indicates muscle activity and (2) that muscle activity in the alpha band is a linear function of the muscle activity in the high-frequency EMG band. To satisfy the rst condition, the EMG band must be selected carefully. Because the frequency range of the EEG has a fuzzy upper limit that lies somewhere between 70 and 100 Hz (Niedermeyer, 1998) but the EMG band should also be set in a range where pericranial muscles produce maximal spectral power (i.e. between 20 and 100 Hz), a EMG frequency band around 100 Hz (e.g. 90100 Hz) may be an appropriate compromise. The second assumption that supposes linearity is not strictly met, because the lower and upper range of the EMG spectrum reects different properties of muscle activity. In particular, the power between 10 and 30 Hz is mostly due to the ring rates of motor units and beyond 30 Hz it is due to the shape of the aggregated motor unit action potentials (Basmajian and De Luca, 1985). Recent experimental work has suggested that an individuals lower end of the envelope of the EMG spectrum (i.e. the presence or absence of a ring rate peak) cannot be predicted in a specic situation (van Boxtel, 2001). At present, it is not known how severely the regression approach is limited by this problem. Taken together, several methods allow a treatment of ocular artifacts, and in theory, all have their own advantages and limitations. The rejection procedure is the most commonly used method, but it has the limitation that some criteria for the presence of artifacts must be
164
specied and that small artifacts may remain in the data. Fortunately, empirical evidence suggests that the method is rather robust because even great variations of these criteria do not substantially affect individual differences in measures of anterior EEG asymmetry. The regression procedure has the disadvantage that it may remove brain activity from the data, but preliminary evidence suggests that this problem does not cause a distortion of frontal EEG asymmetry. The MSEC method is limited by its approximate nature, and its value for the measurement of anterior EEG asymmetry has not been established yet. Therefore, it appears that the rejection approach (and possibly the regression approach) may be an appropriate choice for the purpose of ocular artifact treatment. In contrast, the treatment of muscle artifacts poses a more serious difculty. Here, the rejection approach has the same limitations but its robustness has not been established yet. Moreover, the regression approach rests on the assumption that a high-frequency EMG band is indicative for low frequency muscle activity but the extend to which this assumption holds is not known in practice. It appears that both methods may be recommended on comparable grounds.
4. EEG reference Another classical issue in quantitative EEG research is the EEG reference problem (Katznelson, 1981; Lehmann, 1987). This problem is based on the dilemma that the quantication of asymmetries of cortical activity requires the measurement of electrical activity at single sites, but each EEG channel only yields information about the difference of electrical activity between two sitesusually between a cephalic target site and a reference site. Thus, the recording reects the activity at the target and the reference site (Nunez, 1981), and as a consequence, the derivation of one particular target site to different reference sites may result in quite distinct EEG data (e.g. Katznelson, 1981; Lehmann and Skrandies, 1984; Lehmann, 1984). This problem is of particular importance for the measurement of EEG asymmetry, because such a measurement requireslike all topographical analysisa non-ambiguous registration of single site activity. The effects of typically used reference schemes for the measurement of frontal EEG asymmetry have been analyzed in three recent studies (Hagemann et al., 1998, 2001; Reid et al., 1998), and the ndings were rather troublesome. Across four samples, the mean correlation between the common vertex and the mathematically linked ears/mastoids reference (which are the two most commonly used montages in asymmetry research) was r = 0.004 for the midfrontal asymmetry measures (range between 0.30 and 0.33). Because the reliability of these measures was good (reliability estimates were typically in the 0.80 and 0.90 s), this nding indicates that the common variance of the true scores of the two references was virtually zero. Therefore, the different reference schemes appear to measure psychometrically distinct properties of brain activity, i.e. there is no convergent validity for these measures. Thus, treating them as interchangeable might be inappropriate, and several consequences may arise from this problem with regard to the construct validity of EEG asymmetry measures. According to Cronbach and Meehl (1955), construct validity is faced with the problem of what constructs account for variance in test performance. Only a nomological network, which is a system of laws that relate different constructs to one another and which relates
165
constructs with observables, can imply construct validity for a given measure. Because there may be multiple constructs that account for variance in a particular EEG asymmetry measure (such as genetic, neurochemical, neurophysiological, psychological/behavioral, and social factors), a complete account of the construct validity of a particular EEG asymmetry measure must ultimately link the measure to both the biological and also the behavioral constructs. In this sense, a particular EEG asymmetry measure may have a neurophysiological construct validity (if there is a link between the measure and an underlying asymmetry in brain activity) and it may have a psychological/behavioral construct validity (if the measure is related to behavioral activation, approach motivation, and affective style). Most important, one kind of construct validity does not imply the other; therefore, an EEG asymmetry measure that is based on a particular reference may have good neurophysiological construct validity but insufcient psychological/behavioral construct validity, or vice versa. A somewhat different account on construct validity was fostered by Cook and Campbell (1979), who based inference about constructs more on the t between operations and conceptual denitions than on the t between obtained data patterns and theoretical predictions about such data patterns. In this sense, construct validity can already bee implied by parts of the nomological network, which relate constructs with observables (rather than considering the links between a multitude of constructs). Thus, an EEG measure of the construct asymmetry of cortical activity has good construct validity sensu Cook and Campbell (1979) if there is a link between the measure and an underlying asymmetry in brain activity, which corresponds to the idea of neurophysiological construct validity noted above. Within the framework of Cook and Campbell (1979), construct validity may be investigated with the multitrait-multimethod matrix (Campbell and Fiske, 1959). This methodology requires that different traits are measured with different methods. If different measurement methods of the same trait show convergence and the same measurement methods of different traits show divergence, then good construct validity is established. Thus, a low convergence between different reference schemes for measures of frontal alpha asymmetry constitutes a problem of construct validity. As noted above, the correlation between the common vertex and the mathematically linked ears/mastoids reference is about zero for the midfrontal asymmetry measure, thus, implying that different constructs are measured with these montages. However, other reference schemes may yield measures with a more convergent validity. Across three samples, the mean correlation between the mathematically linked ears/mastoids reference and an average reference was r = 0.81 for the midfrontal asymmetry measures (Hagemann et al., 2001; Reid et al., 1998). In light of the high reliability of these measures, this latter nding suggests that both references yield measures with a common factor that contributes shared variance (e.g. due to the asymmetry construct) and specic factors that contribute unique variance (e.g. due to method factors) in addition to measurement errors. Thus, construct validity comes in gradations but is never perfect because each method of measurement contributes at least some unique variance. In what follows, a closer look at the most frequently used reference schemes might illuminate their relative utilities and downsides for the measurement of the construct asymmetry of cortical activity and thus, may shed some light on the construct validity of the measures (for a tutorial on the reference issue and a more comprehensive comparison, see Hagemann et al., 2001). Since the target site activity is the signal of interest, the reference site activity may be denoted as noise; thus, the greater the amplitude of the target site activity becomes
166
in relation to the amplitude of the reference site, the higher the signal-to-noise ratio will become, and the topographic validity of the reference scheme will increase. For an application of this principle to different reference schemes one must bear in mind that the topography of alpha activity in a resting state is commonly presumed to have a rather monotonic increase from anterior to posterior (Creutzfeldt, 1995; Zschocke, 1995; see Hagemann et al., 2001, for a discussion of this assumption). The common vertex (Cz) reference is one of the most frequently used reference schemes in asymmetry research, but it may have a most unfavorable signal-to-noise ratio. The vertex is certainly an electrically active site, and it may be presumed that the frontal sites have a lower alpha activity than vertex; thus, the derivations of F3 and F4 against vertex hardly reect the activity of prefrontal structures but mostly vertex activity. Moreover, it can be shown that, depending on the amplitude and phase relations between the two target sites and the reference site, the true amplitude asymmetry of the target sites may be enhanced, mitigated, or even reverted (Hagemann et al., 2001). Therefore, if the (asymmetry of) local alpha activity at the left and right frontal lobes needs to be measured, then the common vertex reference appears to be less appropriate. One prospective solution to this problem might be the (mathematically) linked earlobes/mastoids (A1 + A2) reference, which is another commonly used montage. The topographic validity of this reference scheme rests on the assumption that the actual reference site (the average of A1 and A2) is substantially less active than the cephalic target sites (e.g. Hagemann et al., 1998). Thus, the signal-to-noise ratio would be quite favorable for the derivations of all scalp electrodes to this reference. Unfortunately, empirical data do not support this assumption. Earlier studies could demonstrate substantial event-related activity at A1 and A2 in a variety of paradigms (Curran et al., 1993; Dien, 1998; Lehtonen and Koivikko, 1971). A more recent study of Hagemann et al. (2001) demonstrated that the bipolar derivation of the ears (A1A2) in a resting state shows highly signicant alpha activity, which was greater than the alpha activity of bipolar derivation between the frontopolar (Fp1Fp2) and midfrontal (F3F4) sites, but smaller than activity of a central derivation (C3C4). This nding appears to be difcult to reconcile with the assumed inactivity of the A1 and A2 reference sites, but the data also suggests that the earlobes/mastoids may have lower activity than vertex. Thus, the signal-to-noise ratio of the A1 + A2 reference for the measurement of anterior resting activity might be less favorable than one may wish, but it appears to be greater than for the Cz reference. Therefore, the A1 + A2 reference may be more appropriate for the measurement of frontal resting asymmetry than the Cz reference. For practical applications of the A1 + A2 reference, it must be noted that physically linking the ears/mastoids may result in an electrical shunt across the head if the electrode impedances at A1 and A2 are very low (Katznelson, 1981; see also Miller et al., 1991), and may also distort the EEG if the impedances at A1 and A2 are high but asymmetrical (see Miller et al., 1991; Nunez, 1991).4 To avoid this problem effectively, all scalp electrodes may
4 In general, the pre-existing biological resistance is quite low compared to the linked earlobes or mastoids if the electrode impedance is within a typical range (e.g. between 1 and 10 K ). Thus, the effects of linking the ears or mastoids are not likely to be of practical signicance. However, if the electrode impedances are unusually low (i.e. less than 1 K ), then linking the ears or mastoids may create an effective shunt across the head (for detailed model calculations, see Miller et al., 1991).
167
be recorded to a common and arbitrary reference site, and rereferenced to a mathematically A1+A2 reference (for a comprehensive discussion of this issue, see Lehmann, 1987; Miller et al., 1991; Nunez, 1991; Nunez et al., 1991). Such a common reference scheme for EEG recording has the additional advantage that any other montage may be computed subsequent to the data acquisition. One different and quite promising solution to the EEG reference problem is the average reference, although this montage is less frequently applied in the context of frontal asymmetry research. In theory, Bertrand et al. (1985) could demonstrate that under certain conditions, the integral of the potential on the surface of a sphere is zero, and therefore, this integral would be a perfect (inactive) EEG reference. In practice, the surface integral of the head may be approximated by the average of the potentials across all scalp electrodes. However, this average reference will be electrically null only if the head is sampled with a high spatial electrode density and if the whole head is sampled (Bertrand et al., 1985). If this condition is not met, then a distortion of the alpha topography may occur, which has been described as the frontal mirroring of the alpha rhythm (Zschocke, 1995). Because alpha tends to be synchronous across broad areas of the scalp with small amplitudes at anterior and large amplitudes at posterior sites (e.g. Creutzfeldt, 1995), the average across the sites that are located within the limited area of the 1020 system will not result in a zero potential, but in an alpha activity of medium magnitude. Referencing the low-amplitude activity of anterior sites to the medium-amplitude activity of the AR scheme will then result in a medium-amplitude activity (which is largely due to the activity of the virtual AR site). Effectively, the posterior alpha activity appears to be mirrored at the central coronal line (for empirical demonstrations, see Hagemann et al., 2001; Hjorth, 1975; Katznelson, 1981; Maurer and Dierks, 1991; for a computer simulation, see Mac Gillivray and Sawyers, 1988). Although there is no scarcity of discussion about the electrode density and head coverage that are necessary to obtain unambiguous EEG results with the AR approach (Chung et al., 1996; Curran et al., 1993; Desmed et al., 1990; Junghfer et al., 1999; Srinivasan et al., 1998; Tomberg et al., 1990), no consensus has emerged so far. Nonetheless, the occurrence of artifactual frontal alpha activity due to the mirror phenomenon suggests that the use of an AR, which is based on 32 electrodes of the 1010 system, must be discouraged for the measurement of frontal resting asymmetry. A totally different approach that seeks to overcome the dilemma of the EEG reference problem is the use of current source density (CSD) measures. This approach aims to estimate the radial current density of the sources and sinks on the scalp that are generated by the electrical activity of the brain. Because the current source density is proportional to the curvature of the surface potential eld (i.e. the rate of change of the rate of change of the scalp potential when moving along the scalp; Hjorth, 1975; Nunez, 1981), the CSD measures can be computed by suitable difference operations of EEG potential data (Hjorth, 1975; for further computational procedures, see Hjorth, 1980; Perrin et al., 1989, 1990). It might be noted that this measure can be conceptualized as the second spatial derivative of the potential eld at a specic electrode with respect to each surrounding electrode. The resulting measures are strictly reference-free, i.e. they are invariant with respect to the reference site that was used for EEG recording. Moreover, the CSD transformation has also the property to act as a spatial high-pass lter, i.e. broadly distributed sources and sinks
168
will be attenuated. This latter property may be benecial for research on alpha asymmetry, because activity of one hemisphere or region that is introduced via volume conduction to the other hemisphere or region would be considerably reduced with a CSD transform. However, this same property might also have an unfavorable effect for the measurement of alpha activity. Because alpha tends to be broadly distributed, the transformation of EEG potential data into CSD measures might eliminateat least in theorythe very activity of interest. Fortunately, recent empirical ndings suggest that this nuisance might be of little practical concern. First, CSD alpha in a resting state shows the expected scalp topography with a monotonic increase from anterior to posterior and with greater right-sided activity (Hagemann et al., 2001). Second, a structural equation modeling analysis of CSD based resting asymmetry measures revealed an excellent reliability of these measures, with coefcients in the upper 0.90; moreover, these measures also showed a mediocre trait specicity with coefcients of about 0.60, and an excellent temporal stability of the latent trait component with retest correlations close to 1 (Hagemann et al., 2002). It may be interesting to note that the same analysis, when performed with A1 + A2 based asymmetry measures, revealed a somewhat lower trait specicity of resting asymmetry. Third, the CSD alpha asymmetry measures showed excellent convergent validity with Cz, A1 + A2, and AR derived asymmetry measures for the posterior region, with most correlations in the 0.90 s (Hagemann et al., 2001). It should be noted that posterior asymmetry measures appear to be rather invariant with respect to the EEG reference, which may be due to the greater signal-to-noise ratio for posterior sites (where alpha tends to be maximal). Fourth, CSD measures of resting alpha activity show convergent validity with another neuroimaging modality. Cook et al. (1998) obtained EEG and H2 15 O positron emission tomography data simultaneously from normal subjects in a resting state and during a motor task. Across conditions, scalp/cortex positions, and reference montages (ears or CSD), alpha band activity was inversely related to cortical perfusion. Taken together, none of these observations can easily be reconciled with the assumption that (the asymmetry of) resting alpha activity is eliminated by the CSD approach. Thus, the CSD approach might be a prospective solution to the EEG reference problem for research on resting asymmetry, in particular if the integration of EEG data with other neuroimaging modalities is desired. To sum up, some evidence suggests that there is only low convergent validity for anterior asymmetry measures that are based on different EEG reference schemes for at least some references, and the choice of the EEG reference may affect the outcome of the study. From the perspective of psychological/behavioral construct validity, it might be prudent to continue to use different reference schemes in each study, and to compare the relationships between the asymmetry measures and the psychological/behavioral variables of interest (i.e. emotion, motivation). Such a comparison may be facilitated by a data analytic approach that uses general linear model (GLM) analysis, which allows inclusion of a repeated measures factor for reference scheme (see Coan and Allen, 2003a, for an example). A signicant interaction involving this reference factor would indicate that the asymmetrybehavior relationship is moderated by the reference, and thus, allows for a selection of the reference with the strongest link between the variables. If conducted over several years, this type of research would eventually establish good psychological/behavioral construct validity of EEG asymmetry measures.
169
From the perspective of neurophysiological construct validity, however, a pre-selection of a particular reference scheme might be desired, and such a selection may consider the relative advantages and disadvantages of each method for the measurement of the asymmetry construct. The common Cz reference appears to be the least recommended montage because the alpha activity at Cz tends to be greater than at anterior sites, which may severely distort the measurement of anterior alpha activity. The ARif based on a limited head coverage as with the 1020 systemmay result in an increased anterior alpha activity, which might be an artifact of the reference scheme; thus, this method may be very problematic for the measurement of frontal alpha activity as well. The A1 + A2 reference sites show alpha activity, the magnitude of which may approach the activity at anterior sites, which is a disadvantage of this method. However, some data suggest that the A1 + A2 reference may yield greater signal-to-noise ratios for anterior sites than the Cz and AR montages. Thus, the A1 + A2 reference appears to be more appropriate for the measurement of anterior alpha activity than the other two schemes. Finally, an entirely different approach is the use of CSD measures, which are reference-free and enhance the spatial resolution by attenuating broadly distributed activity. Although this method has some shortcomings such as the need for large electrode arrays and the problem that the estimation of the eld curvature at the edge of the electrode system may be inaccurate (e.g. at lateral frontal or frontopolar sites), this approach may be recommended as a prospective solution to the reference problem. Because CSD and potential data measure different properties of the electrical eld of the head, a research approach that utilizes the CSD and the A1 + A2 reference as alternative methods within one study might be a viable route for frontal asymmetry research. This type of research allowsat least conceptuallya direct operationalization of the theoretical construct asymmetry of cortical activity, and thus, facilitates an empirical test of theories that operate with this construct. The comparison between the CSD and A1 + A2 reference may again be facilitated by a GLM analysis that includes a repeated measures factor for method. If the asymmetrybehavior relationship is moderated by method, then these relationships can be interpreted only with uttermost cautionthe discussion of the A1 + A2 reference and the CSD should have made clear that no denite conclusion can be drawn concerning the neurophysiological construct validity of both methods. If the asymmetrybehavior relationship is not moderated by method, however, then this association might be interpreted with some condence in terms of the intended construct.
5. State- and trait-composition of resting asymmetry After artifact treatment and rereferencing of the EEG, the next analysis step is the extraction of alpha activity via a spectral analysis of the data (see Dummermuth and Molinari, 1987; Pivik et al., 1993, for guidelines). The resulting alpha power values may be used as a proxy for cortical activity, with high values of alpha indicating a low level of cortical activity (e.g. Shagass, 1972; see also Davidson, 1998b), and asymmetry measures of cortical activity may be computed as the difference of alpha power of contralateral homologous sites. Unfortunately, the resulting measures are not trait variables, which is mainly due to the distinct state-dependence of the spontaneous EEG. This state-dependence is evidenced by rather low temporal stability of resting asymmetry.
170
It is well established that resting asymmetry measures of healthy subjects show retest correlations in the 0.50 or 0.60 s for time intervals between 2 and 6 weeks for a variety of reference schemes such as Cz, A1 + A2, nose, the average reference, and CSD measures (Allen et al., in press; Debener et al., 2000; Hagemann et al., 2002; Papoussek and Schulter, 1998; Sutton and Davidson, 1997; Tomarken et al., 1992). In contrast, most personality measures, corrected for unreliability, have annual stability coefcients of 0.980.99 (see Costa and McCrae, 1992). To explain this difference in stability, Tomarken et al. (1992) suggested that resting EEG asymmetry might reect the joint contribution of a trait that is superimposed on state-like factors (see also Davidson, 1992, 1993; Wheeler et al., 1993). Only recently Hagemann et al. (2002) provided direct evidence in support of the proposed dual nature of resting asymmetry. They recorded EEG for 12 min in four occasions of measurement each 4 weeks apart, transformed the data into CSD measures, and used a structural equation modeling approach to decompose asymmetry measures into latent state and trait components within the framework of latent statetrait (LST) theory (Steyer et al., 1992). As noted above, the authors found that about 60% of the variance of the resting asymmetry measure was due to a latent trait and that about 40% was due to state-like uctuations (effected by the situation or the interaction between person and situation). The variance due to measurement errors was rather negligible. The authors obtained similar results when they repeated their analysis with EEG data referenced to A1 + A2, although the trait specicity was now somewhat lower and the occasion specicity was somewhat higher. The trait component of the resting asymmetry measure is the variable of interest but the additional occasion-specic uctuation is unwanted and needs to be explained. Fortunately, several studies could shed light on the impact of situational factors that affect resting EEG asymmetry. For example some evidence suggests that states of positive and negative affect or approach and withdrawal motivation might be associated with shifts of anterior asymmetry (based on EEG references such as Cz or A1 + A2; e.g. Davidson et al., 1990; Sobotka et al., 1992; Tucker et al., 1981). Thus, transient mood state during the recording of the resting EEG may contribute to its state variance. Earlier studies investigated this issue and provided null ndings. Tomarken et al. (1990) recorded a resting EEG with a Cz reference and obtained mood ratings for the recording period after it was terminated. There was no association between mood and resting asymmetry. In a similar fashion, three more studies assessed transient mood before and after EEG recording, and found no correlation between the average of the mood ratings and resting asymmetry for Cz and A1 + A2 references (Hagemann et al., 1998, 1999; Wheeler et al., 1993). In addition, Henriques and Davidson (1991) reported that the severity of depressive states (as measured with the Beck Depression Inventory, Beck et al., 1961) was unrelated to frontal resting asymmetry for a Cz and an average reference, both in a group of depressed and control subjects (although this nding may be due to restricted range). In contrast to these ndings, Blackhart et al. (2002) suggested that the mood before the EEG recording might be more predictive, because the EEG preparation procedure (tting of an EEG cap, abrasion of the scalp, application of gel) is somewhat aversive, and thus, may induce a negative affective state. To test this hypothesis, they assessed mood before and after the EEG preparation, followed by the recording of the EEG with an A1 + A2 reference. There was a signicant mood shift during EEG preparation, with less pleasant
171
mood after the procedure was over. In addition, there were signicant associations between mood after EEG preparation and resting asymmetry if the mood before the EEG preparation was partialed out. Women who reported greater unpleasant mood showed greater relative right fontal activity, but men who reported greater unpleasant mood showed greater relative left fontal activity. This nding suggests that differences in transient mood may contribute to the state-like variance of resting asymmetry. Another factor that may contribute to the state variance of resting asymmetry is the transient motivational state during the preparation and recording of the EEG. Kline et al. (2002) investigated this issue recently. They pointed out that defensiveness is conceptually and empirically related to self-presentation motives (e.g. Baumeister and Cairns, 1992), it may moderate physiological responses to social stimuli (e.g. Newton and Contrada, 1992), and some evidence suggests that defensiveness may be associated with anterior resting asymmetry (Kline et al., 1998, 2001; Tomarken and Davidson, 1994). Because opposite-sex encounters can efciently affect self-presentation motives (see Leary et al., 1994), Kline et al. (2002) suggested that the defensiveness of a subject might interact with the presence of an opposite-sex experimenter, and thus, induce a shift of anterior asymmetry. To test this hypothesis, Kline et al. (2002) recorded a resting EEG while a same-sex or opposite-sex experimenter was present. During the presence of an opposite-sex experimenter, high-defensive subjects showed greater left frontal activity but low-defensive subjects showed greater right frontal activity, whereas no group effect occurred in the presence of a same-sex experimenter. This nding suggests that complex situational factors (such as the sex of the experimenter in relation to the sex of the subject), if not held constant, may contribute to the occasion-specic variance. Perhaps, a further factor that might contribute to the state variance of resting asymmetry is a transient state of approach and withdrawal motivation. Several studies have manipulated approach and withdrawal states, and provided evidence for a systematic association between this experimental manipulation and frontal EEG asymmetry as measured with the A1 + A2 reference (Harmon-Jones et al., 2002, 2003; Harmon-Jones and Sigelman, 2001; Sobotka et al., 1992). However, it is not known at present if such an association also exists for unprovoked states of approach and withdrawal motivation, thus, the contribution of this factor to the state variance of resting asymmetry remains uncertain. Because the supposed dual nature of resting EEG asymmetry may contribute to the inconsistency of ndings, Davidson (1998b) suggested that it might be imperative to employ a method for the control/reduction of state-like variability in the measures. Several procedures have been used for this purpose. For example Wheeler et al. (1993) endeavored a selection of stable subjects. The authors selected those participants from a larger sample whose standardized resting asymmetry measures on two occasions of measurement showed no difference greater than one third of a standard deviation. These participants were supposed to have a stable asymmetry, and the asymmetry measures were averaged across both occasions to further reduce its state variance. One grave limitation of this approach is that the selection of participants with similar asymmetry measures on two occasions does not necessarily imply that these participants have a stable asymmetry. Alternatively, the asymmetry of these persons may show a uctuation across time similar to the asymmetry of unstable persons, but just happened to be similar on two occasions by chance. This objection found empirical support in the study of Hagemann et al. (2002) who analyzed data
172
from four occasions of measurement. Subjects who showed similar asymmetry measures in the rst two sessions showed dissimilar asymmetry measures in the last two sessions; the retest correlation of a stability index across the rst two sessions was uncorrelated with a stability index across the last two sessions (r = 0.03 for the stability index of midfrontal asymmetry). In total, these ndings suggest that the similarity of asymmetry measures in two occasions is no stable characteristic of the person, and thus, appears not to be a suitable criterion for subject selection. A related approach is the selection of extreme and stable subjects. For example Tomarken and Davidson (1994) selected those persons from a larger sample whose asymmetry measure scored in the top or bottom 25 percentile in two occasions of measurement (for a similar approach, see Tomarken et al., 1992). A major limitation of this method is that each extreme group selection suffers from the problem of a regression to the mean, which is a hazard for any specic substantive interpretation of the data (see Cohen and Cohen, 1983, for a full discussion of this problem). Moreover, the objections against the selection of stable subjects also apply to the selection of subjects that are extreme on two separate occasions (which implies relative similarity in both occasions). Therefore, this selection procedure may be less appropriate. A very different approach is the aggregation of data across occasions. For example Sutton and Davidson (1997) averaged the EEG data of two occasions in an attempt to reduce the state-like variability of the asymmetry measures (for other examples of an aggregation across two occasions, see Debener et al., 2000; Tomarken and Davidson, 1994). The rationale of this approach leads back to work of Epstein (1983). If any measure reects the joint contribution of a trait and occasion-specic state-like factors, then aggregation across several occasions may reduce the state and increase the trait variance (see also Schmitt and Steyer, 1990). A limitation of this method is that it may reduce the state-ux contamination of the asymmetry measures, but it does not eliminate it. In particular, it is not self-evident that an aggregation across only two occasions would yield asymmetry measures with sufciently reduced state-variance. To estimate the effects of aggregation Hagemann et al. (2002) performed latent statetrait model calculations for the EEG asymmetry measures (these calculations are similar to the SpearmanBrown formula for prolonging scales; see Steyer and Schmitt, 1990, for theoretical foundations). For the case that the data were aggregated across two, three, and four occasions the median of the trait specicity across the whole head was estimated to be 0.76, 0.83, and 0.86, respectively. This nding suggests that an aggregation across two or three occasions may yield an asymmetry measure with mediocre or good trait specicity, respectively, but an aggregation across four occasions would provide only little further improvement. Finally, it may be mentioned that an elimination of unsystematic effects of the situation may be achieved with simultaneous latent statetrait modeling. In this approach, two or more latent traits may be represented with separate LST models within one simultaneous analysis, and the correlation of the latent traits or the latent state residuals may be assessed directly (see Eid et al., 1994; Steyer et al., 1990, for applications). Unfortunately, this type of analysis involves complex structural equation models, which require larger sample sizes for a reliable identication of the latent variables than those that are usually available. Taken together, the resting EEG asymmetry is not a trait variable, but about half of its true score variance may be attributed to a latent trait and the other half may be due to
173
occasion specic factors. Some evidence suggests that transient mood states and the social context may contribute to the state-like variations of the measure, but there is no doubt that a plethora of specic factors are at work that are not known at present. Nonetheless, these factors need to be controlled to ensure that the trait component of the measure is enhanced. Several procedures aimed to reduce the occasion-specic variance, but not all may be suited equally well. Some researchers selected subjects with similar (or extreme) asymmetry measures on two occasions and trusted that these subjects have a rather stable asymmetry, i.e. an asymmetry that is not likely to respond to situational factors. However, recent empirical evidence suggests that this selection procedure does not result in a group of stable subjects, because the similarity of the asymmetry in two subsequent occasions is no stable characteristic of the person. Other researchers aggregated the EEG data across two occasions of measurement. This approach appears to be far more appropriate, because 7080% of the variance of these measures may be due to the latent trait. A more favorable aggregation would include even three occasions, but four occasions would hardly improve the measures. Alternatively, the latent trait variable of the asymmetry measure might be correlated directly with the behavioral indicators in a structural equation modeling approach, but the necessary models are rather complex and require probably large sample sizes.
6. Statistical issues No matter how the alpha power measures of the resting EEG were acquired and processed, the nal step of the empirical endeavor is the statistical analysis of the data. Here, a rst problem is the positively skewed distribution of alpha power, which makes a normalization of the data necessary. For this purpose, Gasser et al. (1982) compared several transformations, and found that the natural logarithm yields the closest approximation to the normal distribution. Thus, this transformation should be applied before further statistical analyses take place. The associations between the topography of alpha activity and behavioral traits may be statistically evaluated in a multivariate fashion, either with a correlation analysis or with a general linear model approach. Both approaches may have several advantages and disadvantages that should be considered. For the correlation analysis, the asymmetry measures for anterior and posterior regions are correlated with the behavioral variables (the posterior asymmetry needs to be introduced into this analysis to secure that the associations under question are really specic to the anterior sites). The directional hypotheses allow one-tailed t-tests of signicance for each correlation involving the anterior sites, which will yield greater statistical test power than the usual two-tailed tests (see Hays, 1994). However, the null hypotheses for posterior sites are non-directional and allow only the use of two-tailed tests. The major drawback of the correlation approach is the hazard of an ination of type-I error rate due to multiple testing. In the typical study, several correlations need to be tested, and it is common practice to test each correlation separately. However, this must inevitably result in an increase of type-I error rate for the whole set of tests, i.e. the probability of making at least one type-I error (the familywise error rate) is greater than the nominal alpha level of each test. For example, if a total of only eight correlations is tested with a nominal
174
level of alpha = 0.05 for each test, the familywise error rate will amount to 0.34 (Hays, 1994, Eq. (11.13.1))which appears to be rather unacceptable. Fortunately, there are two ways to counteract this ination and the danger of erroneous decisions. First, the Bonferroni approach allows an adjustment of the testwise error rate such that the familywise error rate will not exceed the desired nominal alpha level. For the classical Bonferroni approach, the nominal error rate is divided by the number of tests, which yields the adjusted testwise error rate. For the example just given above, a familywise error rate of not greater than alpha = 0.05 will be achieved if the testwise error rate is set to 0.006 (=0.05/8). The major disadvantage of this method is its conservative nature, i.e. this method will yield signicant results only if the population correlations are large. However, a less problematic version of this approach is the sequentially rejective Bonferroni approach of Holm (1979). For this procedure, the correlations must be organized in a rank order, starting with the highest correlation. The rst correlation is tested with the classical Bonferroni method. If it is signicant, then the second correlation is tested with the Bonferroni method, but the number of tests is now reduced by 1. If this correlation is also signicant, then the next correlation is tested, always reducing stepwise the number of tests for the Bonferroni correction. The procedure nally stops when the rst non-signicant test result occurs. This approach has an intuitively appealing logic, it is easy to apply, and it is considerably less conservative than the classical approach. Second, the correlation pattern analysis developed by Steiger (1980) allows the omnibus test of the hypothesis that all correlations in a set are zero. For the application to the present problem, the correlations between asymmetry and affect indices might be divided into two subsets, one containing the associations for anterior and one for posterior sites. Then both subsets may be tested for signicance using Steigers (1980) Eq. (22) (which essentially yields a chi-square model test). If the proposed association between alpha topography and affect holds, then the statistical null hypothesis should be rejected for the anterior but not the posterior set. This approach has the advantage to reduce the total number of tests to two, and thus, minimizes type-I error ination. However, in case of a rejected null hypothesis, the question is open which of the correlations contribute to the overall association and which do not (the sequentially rejective Bonferroni approach might serve as a follow-up procedure). Another problem is that the employed test statistic approaches the chi-square distribution only in medium to large samples (Steiger, 1980). Finally, the underlying chi-square test statistic is usually rather low in power, and may be associated with substantial type-II error in small to medium sized samples (Cohen, 1988). Thus, in small samples, this test may not have adequate statistical power to detect a relationship between anterior asymmetry and affective dispositions that is present in the population. A totally different avenue than the correlation analysis is provided by the general linear model approach, which offers a hierarchical set of repeated measurement omnibus tests. Similar to the correlation pattern analysis, this method avoids multiple testing and thus avoids the ination of type-I error. In contrast to the correlation pattern analysis, however, the GLM employs the F-statistic, which usually has a higher test power than the chi-square statistic (Cohen, 1988). Two different versions of the GLM may be considered. The most commonly applied method is the ANOVA approach. The continuous affect dimensions are reduced to categorical between-subjects variables (e.g. by a median split in high and low scorers), andtogether with factors for hemisphere and regionintroduced
175
into the analysis as independent variables (see Kirk, 1995, for a full treatment). For example a typical ANOVA setup may be a hemisphere region affect group analysis of alpha power at single sites. In case of signicant effects, standard follow-up tests facilitate the decomposition of the omnibus effects (Kirk, 1995). Unfortunately, this approach has two serious disadvantages (see Aiken and West, 1991, for a discussion). First, the loss of information that will occur when a continuous dimension is reduced to a categorical variable with only few classes may impair statistical precision and test power (Cohen and Cohen, 1983). Second, split operations may yield statistical artifacts, in particular if applied to small samples (Bissonnette et al., 1990). However, this disadvantage may be overcome by the multiple regression approach. Like in an ANOVA, the topography may be introduced into the analysis with categorical factors (by effect coding), but the affect dimensions are now introduced as continuous independent variables. For the example given above, a typical setup may be a hemisphereregionaffect dimension analysis of alpha power density at single sites. Signicant main and interaction effects might be followed up with simple regression analyses (see Aiken and West, 1991, for procedures). For the example given above, a signicant three-way interaction would indicate that the slopes of the simple regressions of alpha activity on depression are signicantly different for different electrode sites along the saggital and lateral direction, or stated differently, that associations between asymmetry and affect differ between regions. This omnibus result may be decomposed with separate hemisphere affect analyses for each region. Now a signicant interaction effect would indicate that the slopes of the simple regressions of alpha activity on depression are different between both hemispheres, i.e. that there is an association between asymmetry and affect for that region. To aid the interpretation of this interaction one may plot the two simple regressions. This nested analysis procedure has several advantages. First, it minimizes the number of tests, and thus, counteracts a type-I error ination (Hays, 1994). Second, it makes use of the repeated measurements and employs the F-statistic, which increases the statistical power, and thus, minimizes the type-II error rate (Cohen and Cohen, 1983; Cohen, 1988). Third, this approach avoids the hazard of statistical artifacts (Bissonnette et al., 1990). To sum up, the associations between asymmetry and affect may be analyzed with a correlation or GLM approach. If the former were used, uncorrected tests for each correlation would result in an inated type-I error rate and a classical Bonferroni approach would result in increased type-II error rate, thus, these methods cannot be recommended. However, more sophisticated procedures like the sequentially rejective Bonferroni procedure or the correlation pattern analysis may effectively overcome these drawbacks. Within the GLM approach, the number of comparisons are kept low with hierarchical testing, which keeps the familywise type-I error low. However, the use of percentile splits in order to reduce continuous affect variables to categorical between-subjects factor for classical ANOVAs cannot be recommended because of statistical shortcomings such as increased type-II error rate and statistical artifacts. In contrast, a multiple regression approach allows the inclusion of continuous independent variables, and thus, circumvents this problem effectively. At the present time, no systematic comparisons of the three endorsed methods exist (rejective Bonferroni procedure, correlation pattern analysis, multiple regression), thus, the researcher will be wise to consider all of them and select the approach that best ts his or her needs. Nonetheless, it appears that the GLM/regression approach has the advantage of a greater
176
statistical test power, and additionally allows to treat different reference schemes as an independent factor (see above). Thus, this method may be recommended.
7. How to deal with methodological issues in practice? The measurement of individual differences in anterior EEG asymmetry is associated with a multitude of methodological problems, some of which are rather easy to resolve but others are more challenging. A rather undemanding issue is the assurance of good reliability, which can be easily archived with a sufcient length of EEG recording. Because a recording period of 8 min would ensure excellent reliability of asymmetry measures and since handling and storage of large data les is easily handled by modern computers, this may be considered as an advantageous procedure. Another manageable issue is to secure statistical validity, which might be threatened by an inated type-I error due to multiple testing. Several methods to counteract this problem are available such as the rejective Bonferroni procedure, the correlation pattern analysis, and multiple regression approaches. Each of these methods will work effectively and are easy to use, thus, the researcher may pick the method that best matches his or her question. In a similar manner, the treatment of ocular artifacts poses no serious difculty. Although the artifact rejection approach has the theoretical limitation that the specication of rejection criteria is associated with difculties, empirical evidence suggests that even great variations of these criteria do not substantially affect individual differences in measures of anterior EEG asymmetry. Thus, this method appears to serve the purpose. It should be noted, however, that the frontopolar sites show an exceptional susceptibility for ocular artifacts. If these sites are included into the data processing and analysis, a rather conservative rejection criterion might be considered. The regression approach has the disadvantage that it removes anterior brain activity from the data, but this method appears to have no distorting effects on frontal EEG asymmetry. Thus, as long as used for the study of individual differences, this approach might be recommended as an alternative to the rejection procedure. Finally, the MSEC approach may have its specic merits, but it has also limitations for the measurement of anterior EEG asymmetry and its robustness for this purpose is empirically not established yet. A more demanding issue is the reduction of unwanted state-like uctuations of anterior EEG asymmetry, which requires the collection of data in multiple sessions, and thus, is expensive. The easiest way to handle this problem is maybe the collection of data in two occasions and the aggregation of the asymmetry measures across both sessions; some evidence suggests that the resulting measures may have at least acceptable trait specicity. An aggregation across three occasions would be more favorable, because it may produce measures with even good trait specicity but it also increases the cost of a study. The inclusion of four or more sessions, however, has no obvious advantage. Other approaches that aim to reduce the state-like uctuations by selecting subjects with stable asymmetries appear to have grave limitations and may not be recommended. Another thorny issue is the EEG reference problem, which has presently no denite solution. In practice, the methodological strategy will depend on the research goal of the study. If the aim of a study is the search for a biological marker of emotional and motivational traits, then the psychological/behavioral construct validity of the measure is of
177
importance. For this research, different reference schemes may be used and those schemes will be selected that yield the most robust associations between asymmetry and the behavioral traits. If the aim of the research is a test of a theory that operates with the construct asymmetry of cortical activity, however, then good neurophysiological construct validity is paramount. Since each reference scheme has apparent limitations, the task is to choose the lesser of evils. At the present time, some empirical and theoretical arguments point to the (mathematically) linked ears/mastoids as a working solution as long as potential data are the basis for the asymmetry measures. Nonetheless, it must be observed that the ears/mastoids are not electrically inactive reference sites, thus, some distortion of anterior alpha asymmetry must be expected. A viable alternative to this approach may be the use of CSD data as a basis for asymmetry measures. These quantities are reference-free, but this advantage comes for the price of greater expenditure during data recording (more electrodes are needed for CSD estimation) and greater complexity of data processing. Ideally, a study might combine the linked ears/mastoids reference and the CSD approach. Other reference montages such as the common vertex or the average reference may have less favorable signal-to-noise ratios for anterior alpha activity, and may be less appropriate. No matter whether the psychological/behavioral or neurophysiological construct validity is of prior importance, the systematic comparison of reference schemes may be facilitated by a GLM analysis that includes a repeated measures factor for the EEG method. If the association between asymmetry and behavior is not dependent upon a particular method, this might indicate thatwithin the limits of the implemented measurement proceduresthe ndings can be interpreted with some condence. There is no doubt that research on individual differences in brain asymmetry and emotion is a fascinating topic. On a rst glance, the empirical investigation of this association seems to be an easy ventureall one has to do is measure anterior resting asymmetry and affect dispositions and correlate them. In a close up view, however, this task becomes challenging, and to pursue a better understand of how the brain organizes emotion may lead to another insight: there is no pleasure without pain. Acknowledgements I thank John Allen, Jim Coan, and Eddie Harmon-Jones for helpful comments on this article.
References
Aiken, L.S., West, S.G., 1991. Multiple Regression: Testing and Interpreting Interactions. Sage, Newbury Park. Barlow, J.S., 1986. Artifact processing (rejection and minimization) in EEG data processing. In: Lopes da Silva, F.H., Storm van Leeuwen, W., Rmond, A. (Eds.), Handbook of Electroencephalography and Clinical Neurophysiology. Revised Series. Vol. 2. Clinical Applications of Computer Analysis of EEG and Other Neurophysiological Signals. Elsevier, Amsterdam, pp. 1562. Barz, K., 2003. Okulare Artefakte im ElektroenzephalogrammEin empirischer Vergleich der Rejection Procedure und der Eye Movement Correction Procedure von Gratton und Coles (1989) [Ocular artifacts in the electroencephalograman empirical comparison of the rejection procedure and the Eye Movement Correction Procedure of Gratton and Coles (1989)]. Unpublished diploma thesis, Universitt Trier, Trier.
178
Basmajian, J.V., De Luca, C.J., 1985. Muscles Alive: Their Functions Revealed by Electromyography, fth ed. Williams & Wilkins, Baltimore. Baumeister, R.F., Cairns, K.J., 1992. Repression and self-presentation: when audiences interfere with self-deceptive strategies. Journal of Personality and Social Psychology 62, 851862. Beck, A.T., Ward, C.H., Mendelson, M., Erbaugh, J., 1961. An inventory for measuring depression. Archives of General Psychiatry 4, 561571. Berg, P., 1989. Comments on EOG correction methods. Journal of Psychophysiology 3, 4144. Berg, P., Scherg, M., 1994. A multiple source approach to the correction of eye artifacts. Electroencephalography and Clinical Neurophysiology 90, 229241. Bertrand, O., Perrin, F., Pernier, J., 1985. A theoretical justication of the average reference in topographic evoked potential studies. Electroencephalography and Clinical Neurophysiology 62, 462464. Bissonnette, V., Ickes, W., Bernstein, I.H., Knowles, E., 1990. Personality moderating variables: a warning about statistical artifact and a comparison of analytic techniques. Journal of Personality 58, 567587. Blackhart, G.C., Kline, J.P., Donohue, K.F., LaRowe, S.D., Joiner, T.E., 2002. Affective responses to EEG preparation and their link to resting anterior EEG asymmetry. Personality and Individual Differences 32, 167174. Borod, J.C., Haywood, C.S., Koff, E., 1997. Neuropsychological aspects of facial asymmetry during emotional expression: a review of the normal adult literature. Neuropsychological Reviews 7, 4160. Borod, J.-C., Koff, E., Yecker, S., Santschi, C., Schmidt, J.M., 1998. Facial asymmetry during emotional expression: gender, valence, and measurement technique. Neuropsychologia 36, 12091215. Brodal, P., 1992. The Central Nervous System. Oxford University Press, New York. Brunia, C.H.M., Mcks, J., van den Berg-Lenssen, M.M.C., 1989. Correcting ocular artifacts in the EEG: a comparison of several methods. Journal of Psychophysiology 3, 150. Campbell, D.T., Fiske, D.W., 1959. Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological Bulletin 56, 81105. Chelazzi, L., Corbetta, M., 2000. Cortical mechanisms of visuospatial attention in the promate brain. In: Gazzaniga, M.S. (Ed.), The New Cognitive Neurosciences, second ed. MIT Press, Cambridge, pp. 667686. Chung, G., Tucker, D.M., West, P., Potts, G.F., Liotti, M., Luu, P., Hartry, A.L., 1996. Emotional expectancy: brain electrical activity associated with an emotional bias in interpreting life events. Psychophysiology 33, 218233. Coan, J.A., Allen, J.J.B., 2003a. Frontal EEG asymmetry and the behavioral activation and inhibition systems. Psychophysiology 40, 106114. Coan, J.A., Allen, J.J.B., 2003b. The state and trait nature of frontal EEG asymmetry in emotion. In: Hughdahl, K., Davidson, R.J. (Eds.), Brain Asymmetry, second ed. MIT Press, Cambridge, pp. 565-615. Cohen, J., 1988. Statistical Power Analysis for the Behavioral Sciences. Erlbaum, Hillsdale, NY. Cohen, J., Cohen, P., 1983. Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences, second ed. Lawrence Erlbaum, Hillsdale. Cook, I.A., OHara, R., Uijtdehaage, S.H.J., Mandelkern, M., Leuchter, A.F., 1998. Assessing the accuracy of topographic EEG mapping for determining local brain function. Electroencephalography and Clinical Neurophysiology 107, 408414. Cook, T.D., Campbell, D.T., 1979. Quasi-Experimentation. Design and Analysis Issues for Field Settings. Rand McNally, Chicago. Costa, P.T., Jr., McCrae, R.R., 1992. Trait psychology comes of age. In: Sonderegger, T.B. (Ed.), Nebraska Symposium on Motivation: Psychology and Aging. University of Nebraska Press, Lincoln, NE, pp. 169204. Creutzfeldt, O.D., 1995. Cortex Ceribri. Performance, Structural and Functional Organization of the Cortex. Oxford University Press, Oxford. Cronbach, L.J., Meehl, P.E., 1955. Construct validity in psychological tests. Psychological Bulletin 52, 281302. Curran, T., Tucker, D.M., Kutas, M., Posner, M.I., 1993. Topography of the N400: brain electrical activity reecting semantic expectancy. Electroencephalography and Clinical Neurophysiology 88, 188209. Davidson, R.J., 1988. EEG measures of cerebral asymmetry: conceptual and methodological issues. International Journal of Neuroscience 39, 7189. Davidson, R.J., 1992. Anterior cerebral asymmetry and the nature of emotion. Brain and Cognition 20, 125151. Davidson, R.J., 1993. Cerebral asymmetry and emotion: conceptual and methodological conundrums. Cognition and Emotion 7, 115138. Davidson, R.J., 1998a. Affective style and affective disorders: perspectives from affective neuroscience. Cognition and Emotion 12, 307330.
179
Davidson, R.J., 1998b. Anterior electrophysiological asymmetries, emotion, and depression: conceptual and methodological conundrums. Psychophysiology 35, 607614. Davidson, R.J., Ekman, P., Saron, C.D., Senulis, J.A., Friesen, W.V., 1990. Approach-withdrawal and cerebral asymmetry: emotional expression and brain physiology I. Journal of Personality and Social Psychology 58, 330341. Davidson, R.J., Fox, N.A., 1982. Asymmetrical brain activity discriminates between positive and negative affective stimuli in human infants. Science 218, 12351237. Davidson, R.J., Fox, N.A., 1989. Frontal brain asymmetry predicts infants response to maternal separation. Journal of Abnormal Psychology 98, 127131. Davidson, R.J., Schwartz, G.E., 1976. Patterns of cerebral lateralization during cardiac biofeedback versus the self-regulation of emotion: sex differences. Psychophysiology 13, 6268. Debener, S., Beauducel, A., Nessler, D., Brocke, B., Heilemann, H., Kayser, J., 2000. Is resting anterior EEG alpha asymmetry a trait marker for depression. Neuropsychobiology 41, 3137. Desmed, J., Chaklin, V., Tomberg, C., 1990. Emulation of somatosensory evoked potential (SEP) components with the 3-shell head model and the problem of ghost potential elds when using an average reference in brain mapping. Electroencephalography and Clinical Neurophysiology 77, 243258. Dien, J., 1998. Issues in the application of the average reference: review, critiques, and recommendations. Behavior Research Methods, Instruments and Computers 30, 3443. Dummermuth, G., Molinari, L., 1987. Spectral analysis of EEG background activity. In: Gevins, A.S., Rmond, A. (Eds.), Handbook of Electroencephalography and Clinical Neurophysiology, revised series ed., vol. 1. Elsevier, Amsterdam, pp. 85130. Eid, M., Notz, P., Steyer, R., Schwenkmezger, P., 1994. Validating scales for the assessment of mood level and variability by latent statetrait analysis. Personality and Individual Differences 16, 6376. Elbert, T., Lutzenberger, W., Rockstroh, B., Birbaumer, N., 1985. Removal of ocular artifacts from the EEGa biophysical approach to the EOG. Electroencephalography and Clinical Neurophysiology 60, 455463. Epstein, S., 1983. Aggregation and beyond: some basic issues on the prediction of behavior. Journal of Personality 51, 360392. Friedman, B.H., Thayer, J.F., 1991. Facial muscle activity and EEG recordings: redundancy analysis. Electroencephalography and Clinical Neurophysiology 79, 358360. Gasser, T., Bcher, P., Mcks, J., 1982. Transformations towards the normal distribution of broad band spectral parameters of the EEG. Electroencephalography and Clinical Neurophysiology 53, 119124. Gasser, T., Bacher, P., Steinberg, H., 1985. Testretest reliability of spectral parameters of the EEG. Electroencephalography and Clinical Neurophysiology 60, 312319. Gasser, T., Sroka, L., Mcks, J., 1985. The transfer of EOG activity into the EEG for eyes open and closed. Electroencephalography and Clinical Neurophysiology 61, 181193. Gasser, T., Ziegler, P., Gattaz, W.F., 1992. The deleterious effect of ocular artefacts on the quantitative EEG, and a remedy. European Archives of Psychiatry and Clinical Neuroscience 241, 352356. Gevins, A.S., 1987. Overview of computer analysis. In: Gevins, A.S., Rmond, A. (Eds.), Handbook of Electroencephalogryphy and Clinical Neurophysiology. Revised Series. Vol. 1. Methods of Analysis of Brain Electrical and Magnetic Signals. Elsevier, Amsterdam, pp. 3183. Gotman, J., Ives, J.R., Gloor, P., 1981. Frequency content of EEG and EMG at seizure onset: possibility of removal of EMG artefacts by digital ltering. Electroencephalography and Clinical Neurophysiology 52, 626639. Gratton, G., 1998. Dealing with artifacts: the EOG contamination of event-related brain potential. Behavior Research Methods, Instruments, and Computers 30, 4453. Gratton, G., Coles, M.G.H., 1989. Generalization and evaluation of eye-movement correction procedures. Journal of Psychophysiology 3, 1416. Hagemann, D., Naumann, E., 2001. The effects of ocular artifacts on (lateralized) broadband power in the EEG. Clinical Neurophysiology 112, 215231. Hagemann, D., Naumann, E., Becker, G., Maier, S., Bartussek, D., 1998. Frontal brain asymmetry and affective style: a conceptual replication. Psychophysiology 35, 372388. Hagemann, D., Naumann, E., Lrken, A., Becker, G., Maier, S., Bartussek, D., 1999. EEG asymmetry, dispositional mood and personality. Personality and Individual Differences 27, 541568. Hagemann, D., Naumann, E., Thayer, J.F., 2001. The quest for the EEG reference revisited: a glance from brain asymmetry research. Psychophysiology 38, 847857.
180
Hagemann, D., Naumann, E., Thayer, J.F., Bartussek, D., 2002. Does resting EEG asymmetry reect a trait? An application of latent statetrait theory. Personality and Social Psychology 82, 619641. Harman, D.W., Ray, W.J., 1977. Hemispheric activity during affective verbal stimuli: an EEG study. Neuropsychologia 15, 457460. Harmon-Jones, E., Abramson, L.Y., Sigelman, J., Bohlig, A., Hogan, M.E., Harmon-Jones, C., 2002. Proness to hypomania/mania symptoms or depression symptoms and asymmetrical frontal cortical responses to an anger-evoking event. Journal of Personality and Social Psychology 82, 610618. Harmon-Jones, E., Allen, J.J.B., 1998. Anger and frontal brain asymmetry: EEG asymmetry consistent with approach motivation despite negative affective valence. Journal of Personality and Social Psychology 74, 13101316. Harmon-Jones, E., Sigelman, J., 2001. State anger and prefrontal brain activity: evidence that insult-related relative left-prefrontal activation is associated with experienced anger and aggression. Journal of Personality and Social Psychology 80, 797803. Harmon-Jones, E., Sigelman, J.D., Bohlig, A., Harmon-Jones, C., 2003. Anger, coping, and frontal cortical activity: the effect of coping potential on anger-induced left activity. Cognition and Emotion 17, 124. Hayes, K.J., 1960. Wave analyses of tissue noise and muscle action potential. Journal of Applied Physiology 15, 749752. Hays, W.L., 1994. Statistics, fth ed. Harcourt Brace, Fort Worth. Heller, W., Nitschke, J.B., 1998. The puzzle of regional brain activity in depression and anxiety: the importance of subtypes and comorbidity. Cognition and Emotion 12, 421447. Henriques, J.B., Davidson, R.J., 1990. Regional brain electrical asymmetries discriminate between previously depressed and healthy control subjects. Journal of Abnormal Psychology 99, 2231. Henriques, J.B., Davidson, R.J., 1991. Left frontal hypoactivation in depression. Journal of Abnormal Psychology 100, 535545. Hjorth, B., 1975. An on-line transformation of EEG scalp potentials into orthogonal source derivations. Electroencephalography and Clinical Neurophysiology 39, 526530. Hjorth, B., 1980. Source derivation simplies topographical EEG interpretation. American Journal of EEG Technology 20, 121132. Holm, S., 1979. A simple sequentially rejective multiple test procedure. Scandinavian Journal of Statistics 6, 6570. Iacono, W.G., Lykken, D.T., 1981. Two-year retest stability of eye tracking performance and the comparison of electro-oculographic and infrared recording technique: evidence of EEG in the electro-oculogram. Psychophysiology 18, 4955. Junghfer, M., Elbert, T., Tucker, D.M., Braun, C., 1999. The polar average reference effect: a bias in estimating the head surface integral in EEG recording. Clinical Neurophysiology 110, 11491155. Katznelson, R.D., 1981. EEG recording electrode placement, and aspects of generator localization. In: Nunez, P.L. (Ed.), Electric Fields of the Brain. Oxford University Press, New York, pp. 176213. Kirk, R.E., 1995. Experimental Design: Procedures for Behavioral Sciences, third ed. Brooks/Cole, Pacic Grove. Kline, J.P., Allen, J.J.B., Schwartz, G.E.R., 1998. Is left frontal brain activation in defensiveness gender specic. Journal of Abnormal Psychology 107, 149153. Kline, J.P., Blackhart, G.C., Joiner, T.E., 2002. Sex, lie scales, and electrode caps: an interpersonal context for defensiveness and anterior electroencephalographic asymmetry. Personality and Individual Differences 33, 459478. Kline, J.P., Knapp-Kline, K., Schwartz, G.E.R., Russek, L.G.S., 2001. Anterior asymmetry, defensiveness, and perceptions of parental caring. Personality and Individual Differences 31, 11351145. Klingmann, P.O., 2003. Korrektur okularer Artefakte im EEGEin Methodenvergleich [Correction of ocular artifacts in the EEGa comparison of methods]. Unpublished diploma thesis, Universitt Trier, Trier. Kosslyn, S.M., Thompson, W.L., 2000. Shared mechanisms in visual imagery and visual perception: insights from cognitive neurosciences. In: Gazzaniga, M.S. (Ed.), The New Cognitive Neurosciences, second ed. MIT Press, Cambridge, pp. 975985. Leary, M.R., Nezlek, J.B., Downs, D., Radford-Davenport, J., Martin, J., McMullen, A., 1994. Self-presentation in everyday interactions: effects of target familiarity and gender composition. Journal of Personality and Social Psychology 67, 664673.
181
Lehmann, D., 1984. EEG assessment of brain activity: spatial aspects, segmentation and imaging. International Journal of Psychophysiology 1, 267276. Lehmann, D., 1987. Principles of spatial analysis. In: Gevins, A.S., Rmond, A. (Eds.), Handbook of Electroencephalography and Clinical Neurophysiology. Methods of Analysis of Brain Electrical Signals and Magnetic Signals, revised series ed., vol. 1. Elsevier, Amsterdam, pp. 309354. Lehmann, D., Skrandies, W., 1984. Spatial analysis of evoked potentials in mana review. Progress in Neurobiology 23, 227250. Lehtonen, J.B., Koivikko, M.J., 1971. The use of a non-cephalic reference electrode in recording cerebral evoked potentials in man. Electroencephalography and Clinical Neurophysiology 31, 154156. Mac Gillivray, B.B., Sawyers, F.J.P., 1988. A comparison of common reference, average and source derivations in mapping. In: Samson-Dollfus, D. (Ed.), Statistics and Topography in Quantitative EEG. Elsevier, Paris, pp. 7287. Maurer, K., Dierks, T., 1991. Atlas of Brain Mapping. Topographic Mapping of EEG and Evoked Potentials. Springer, Berlin. Miller, G.A., Lutzenberger, W., Elbert, T., 1991. The linked-reference issue in EEG and ERP recording. Journal of Psychophysiology 5, 273276. Mcks, J., Gasser, T., Sroka, L., 1989. Approaches to correcting EOG artifacts. Journal of Psychophysiology 3, 2126. Newton, T.L., Contrada, R.J., 1992. Repressive coping and verbal/autonomic response dissociation: the inuence of social context. Journal of Personality and Social Psychology 62, 159167. Niedermeyer, E., 1998. The normal EEG of the waking adult. In: Niedermeyer, E., Lopes da Silva, F. (Eds.), Electroencephalography. Basic Principles, Clinical Applications, and Related Fields, fourth ed. Lippincott Williams & Wilkins, Philadelphia, pp. 149173. Nitschke, J.B., Heller, W., Palmieri, P.A., Miller, G.A., 1999. Contrasting patterns of brain activity in anxious apprehension and anxious arousal. Psychophysiology 36, 628637. Nunez, P.L., 1981. Electric Fields of the Brain. The Neurophysics of EEG. Oxford University Press, New York. Nunez, P.L., 1991. Comments on the paper by Miller, Lutzenberger and Elbert. Journal of Psychophysiology 5, 279280. Nunez, P.L., Pilgreen, K.L., Westdorp, A.F., Law, S.K., Nelson, A.V., 1991. A visual study of surface potentials and laplacians due to distributed neocortical sources: computer simulations and evoked potentials. Brain Topography 4, 151168. ODonnell, R.D., Berkhout, J., Adey, W.R., 1974. Contamination of scalp EEG spectrum during contraction of cranio-facial muscles. Electroencephalography and Clinical Neurophysiology 37, 145151. Papoussek, I., Schulter, G., 1998. Different temporal stability and partial independence of EEG asymmetries from different locations: implications for laterality research. International Journal of Neuroscience 93, 87100. Perrin, F., Pernier, J., Bertand, O., Echallier, J.F., 1989. Spherical splines for scalp potential and current density mapping. Electroencephalography and Clinical Neurophysiology 72, 184187. Perrin, F., Pernier, J., Bertrand, O., Echallier, J.F., 1990. Corrigenda. Electroencephalography and Clinical Neurophysiology 76, 565. Pham, D.T., 1989. A comment from viewpoint of time series analysis. Journal of Psychophysiology 3, 4648. Pivik, R.T., Broughton, R.T., Coppola, R., Davidson, R.J., Fox, N., Nuwer, M.R., 1993. Guidelines for the recording and quantitative analysis of electroencephalographic activity in research contexts. Psychophysiology 30, 547 558. Reid, S.A., Duke, L.M., Allen, J.J.B., 1998. Resting frontal electroencephalographic asymmetry in depression: inconsistencies suggest the need to identify mediating factors. Psychophysiology 35, 389404. Robertson, L., Rafael, R., 2000. Disorders of visual attention. In: Gazzaniga, M.S. (Ed.), The New Cognitive Neurosciences, second ed. MIT Press, Cambridge, pp. 633649. Schaffer, C.E., Davidson, R.J., Saron, C., 1983. Frontal and parietal electroencephalogram asymmetry in depressed and nondepressed subjects. Biological Psychiatry 18, 753762. Scherg, M., von Cramon, D., 1985. Two bilateral sources of the late AEP as identied by a spatio-temporal dipole model. Electroencephalography and Clinical Neurophysiology 62, 3244. Schmitt, M.J., Steyer, R., 1990. Beyond intuition and classical test theory: a reply to Epstein. Methodika 4, 101107.
182
Shagass, C., 1972. Electrical activity of the brain. In: Greeneld, N.S., Sternbach, R.A. (Eds.), Handbook of Psychophysiology, Holt, Rinehart and Winston, New York, pp. 263328. Sobotka, S.S., Davidson, R.J., Senulis, J.A., 1992. Anterior brain electrical asymmetries in response to reward and punishment. Electroencephalography and Clinical Neurophysiology 83, 236247. Srinivasan, R., Tucker, D., Murias, M., 1998. Estimating the spatial Nyquist of the human EEG. Behavioral Research Methods, Instruments, and Computers 30, 819. Steiger, J.H., 1980. Tests for comparing elements of a correlation matrix. Psychological Bulletin 87, 245251. Steyer, R., Ferring, D., Schmitt, M., 1992. States and traits in psychological assessment. European Journal of Psychological Assessment 8, 7998. Steyer, R., Schmitt, M., 1990. The effects of aggregation across and within occasions on consistency, specicity and reliability. Methodika 4, 5894. Steyer, R., Schwenkmezger, P., Auer, A., 1990. The emotional and cognitive components of trait anxiety: a latent statetrait model. Personality and Individual Differences 11, 125134. Sutton, S.K., Davidson, R.J., 1997. Prefrontal brain asymmetry: a biological substrate of the behavioral approach and inhibition systems. Psychological Science 8, 204210. Tomarken, A.J., Davidson, R.J., 1994. Frontal brain activation in repressors and nonrepressors. Journal of Abnormal Psychology 103, 339349. Tomarken, A.J., Davidson, R.J., Henriques, J.B., 1990. Resting frontal brain asymmetry predicts affective responses to lms. Journal of Personality and Social Psychology 59, 791801. Tomarken, A.J., Davidson, R.J., Wheeler, R.E., Doss, R.C., 1992. Individual differences in anterior brain asymmetry and fundamental dimensions of emotion. Journal of Personality and Social Psychology 62, 676687. Tomarken, A.J., Davidson, R.J., Wheeler, R.E., Kinney, L., 1992. Psychometric properties of resting anterior EEG asymmetry: temporal stability and internal consistency. Psychophysiology 29, 576592. Tomberg, C., Noel, P., Ozaki, I., Desmed, J., 1990. Inadequacy of the average reference for the topographic mapping of focal enhancements of brain potentials. Electroencephalography and Clinical Neurophysiology 77, 259265. Tucker, D.M., Stenslie, C.E., Roth, R.S., Shearer, S.L., 1981. Right frontal lobe activation and right hemisphere performance: decrement during a depressed mood. Archives of General Psychiatry 38, 169174. van Boxtel, A., 2001. Optimal signal bandwidth for the recording of surface EMG activity of facial, jaw, oral, and neck muscles. Psychophysiology 38, 2234. van Boxtel, A., Goudswaard, P., Shomaker, L.R.B., 1984. Amplitude and bandwidth of the frontalis surface EMG: effects of electrode parameters. Psychophysiology 21, 699707. van Driel, G., Woestenburg, J.C., van Blokland-Vogelesang, A.W., 1989. Frequency domain methods: a solution for the problems of EOGEEG contamination in ERPs. Journal of Psychophysiology 3, 2934. Verleger, R., 1991. The instruction to refrainfrom blinking affects auditory P3 and N1 amplitudes. Electroencephalography and Clinical Neurophysiology 78, 240251. Weerts, C.W., Lang, P.J., 1973. The effects of eye xation and stimulus and response location on the contingent negative variation (CNNV). Biological Psychology 1, 119. Wheeler, R.E., Davidson, R.J., Tomarken, A.J., 1993. Frontal brain asymmetry and emotional reactivity: a biological substrate of affective style. Psychophysiology 30, 8289. Zschocke, S., 1995. Klinische Elektroenzephalographie. Springer, Berlin.
Methodology
Issues and assumptions on the road from raw signals to metrics of frontal EEG asymmetry in emotion
John J.B. Allen , James A. Coan, Maria Nazarian
Department of Psychology, University of Arizona, P.O. Box 210068, Tucson, AZ 85721-0068, USA
Abstract There exists a substantial literature examining frontal electroencephalographic asymmetries in emotion, motivation, and psychopathology. Research in this area uses a specialized set of approaches for reducing raw EEG signals to metrics that provide the basis for making inferences about the role of frontal brain activity in emotion. The present review details some of the common data processing routines used in this eld of research, with a focus on statistical and methodological issues that have captured, and should capture, the attention of researchers in this eld. 2004 Published by Elsevier B.V.
Keywords: Frontal electroencephalographic asymmetry; EEG; Emotion; Methods; State and trait
The eld of research examining frontal electroencephalographic (EEG) asymmetries in emotion and psychopathology is now over two decades old, with over 80 published studies documenting relationships between asymmetries in frontal EEG power and emotion-related traits and states (see Coan and Allen, 2004, this issue, for review). Although data reduction and analytic techniques have varied across studies, there are many common approaches that have become quite popular for transforming raw EEG signals to metrics that provide the basis for making inferences about the role of frontal brain activity in emotion. These approaches involve many transformations of the data, and in that process involve assumptions that can impact the interpretations scientists can levy from a given pattern of results. The aim of this paper, therefore, is to provide a general overview of some of the common steps involved in data processing in this eld, highlighting the assumptions and the impact of violations of these assumptions for interpreting ndings. It is important to note that none of the issues raised in this paper call to question the now well-replicated relationships between
Corresponding author. Tel.: +1-520-621-4992. E-mail address: jallen@u.arizona.edu (J.J.B. Allen).
0301-0511/$ see front matter 2004 Published by Elsevier B.V. doi:10.1016/j.biopsycho.2004.03.007
184
J.J.B. Allen et al. / Biological Psychology 67 (2004) 183218
the metrics of EEG asymmetry and emotional constructs. The issues will, however, have implications for how the ndings best be interpreted.
1. From raw signals to handy metrics Investigators who examine frontal EEG asymmetry use a set of relatively specialized signal processing routines, which will be reviewed anon. This review is not intended so serve as a primer for basic signal processing, but rather is designed to highlight the data reduction trail typically followed in this specic research domain. For a basic primer, many sources are available, including easily accessible chapters by Gratton (2000) and by Reilly (1987), and a more in depth treatment by Glaser and Ruchkin (1976). Fig. 1 depicts the many steps typically involved in transforming electroencephalographic signals into metrics that putatively are related to how active various brain regions may be. This process involves taking a signal collected in the time-domain (Panel A, left side), and converting it to a frequency-domain representation, usually in the form of a power spectrum (Panel A, right side). This spectrum, which collapses data across time, summarizes which frequencies are present to greater or lesser degrees in the time-domain signal. Whether data are collected from an extended resting period involving several minutes, or from discrete and relatively short emotion-related segments, this spectral analysis approach always involves examining the frequency composition of short epochs (Panel B), on the order of 1 or 2 s each, and averaging power spectra across many such epochs. In the case of resting data, this involves epoching a large data segment into many smaller epochs. In the case of EEG acquired in the context of eeting emotional expression or experience, the data segment might still require being epoched into a few smaller epochs, and data from several such expressions or experiences would then be aggregated. By using epochs that are only 1 or 2 s-long, one more closely approximates an assumption underlying the Fourier transform, the method used to derive power spectra from raw signals. Fourier analyses assume a periodic signal (the stationarity assumption), and moreover that any periodic signal can be decomposed into a series of sine and cosine functions of various frequencies, with the function for each frequency beginning at its own particular phase. A periodic signal is one that repeats, and does so at uniformly spaced intervals of time. Although strictly speaking EEG signals are not periodic, as the repetition of features is not precisely spaced at uniform intervals, by selecting short epochs one can analyze small segments of data that will have features that repeat in a highly similar fashion at other points in the waveform. Epoching typically involves the construction of overlapping epochs (Panel B), as weighting functions applied in the process of windowing (described below) prior to frequency analysis result in the central portion of the epoch receiving the most weight, and distal portions receiving negligible weight (Panel C). By overlapping the epochs, all data points receive maximum weighting in some epoch. Windowing is used to avoid creating artifactual frequencies in the resultant power spectra. Because Fourier transforms assume a periodic signal, it is assumed that the signal in the epoch repeats innitely both forwards and backwards in time, and without the windowing function to reduce the ends of the epoch to near-zero values, discontinuities would
J.J.B. Allen et al. / Biological Psychology 67 (2004) 183218 185
Fig. 1. Depiction of the various data reduction steps typically used in frontal EEG asymmetry research. Panel A depicts a 10-s segment of raw data from a single channel on the left, and the spectral representation of this epoch on the right. Panel B illustrates the process of epoching the longer segment into shorter overlapping 2-s epochs. Panel C depicts the impact of the Hamming window (dotted bell curve) on a single epoch, with the gray line representing the raw signal and the black line representing the signal after the application of the window. Note that a discontinuity would result if a copy of the raw (gray) signal were concatenated following this signal, but no such discontinuity would result for a similarly concatenated windowed (black) signal. Panel D displays the net weighting (black line, scaled to t graph) of overlapping hamming windows (gray lines) for 2-s epochs. Panel E illustrates the impact of averaging power spectra. The top nine gray lines are the spectral representation of nine 2-s epochs, and the lower black line is the average spectrum. Note that alpha power (813 Hz) is somewhat variable from epoch to epoch, but that the average spectrum reveals a distinct alpha peak. Vertical axis in Panel E is power in V2 .
186
result if one were to place a copy of the epoch immediately before or after itself. Fourier methods would introduce a variety of spurious frequencies to reconstruct a signal with such discontinuity. By windowing, the discontinuity is avoided (Panel C), but at the expense of preventing the data near the end of the epoch from being fully represented in the resultant power spectrum. The overlapping of epochs (Panel D) provides a solution to this latter problem, as data minimally weighted at the end of epoch x will be weighted more heavily in epoch x + 1. Most computer signal processing packages use a fast Fourier transform (FFT), which as the name implies is considerably faster and computationally less complex than the more general case discrete Fourier transform (DFT). The FFT requires that the epochs to be analyzed have 2n data points. Data are often sampled at a rate that is a power of two, thus allowing epochs of 1 or 2 s, but in other cases of sample rates that deviate from a power of two (e.g., 250 Hz), epoch length will need to be tailored accordingly (e.g., 2.048 s). For a data segment of 1024 data points, the DFT will take about 10 times longer to arrive at the same result as the FFT.1 The result of the FFT is two spectra, a power spectrum and a phase spectrum. The power spectrum reects the power in the signal at each frequency from dc to the Nyquist frequency,2 with a spectral value every 1/T points, where T is the length of the epoch analyzed. The phase spectrum presents the phase of the waveform at each interval 1/T. These two spectra can jointly be used to reconstruct the original time-domain waveform. Psychophysiologists, however, typically discard the phase spectrum and focus their analyses only on the power spectrum. As an FFT is applied to each epoch, many power spectra result, and the average of these power spectra is ultimately taken as the basis for analysis (Panel E). The data in this resultant spectrum might entail between 20 and 200 data points (the precise number being T (f/2), dependent on the epoch length T and the sample rate f), a substantial reduction from the raw data signal that will likely have hundreds of data points per second for several minutes. The spectra represent, therefore, a relatively economical representation of the original signal, with higher sampling frequencies and longer epochs resulting in more spectral points. Further reduction is accomplished by summarizing data within conventionally-dened frequency bands. Alpha power, either total ( V2 ) or density ( V2 /Hz), is most often examined, and is typically operationalized as power between 8 and 13 Hz in adults, although lower frequencies have been examined in children (for review see Coan and Allen, 2003b), as these lower frequencies in the developing brain are assumed to be equivalent to adult alpha
1 The DFT transform is a general case instantiation of the Fourier transform for discretely sampled signals, but it is computationally intensive, with the time taken to compute the spectral representation being proportional to the square of the number on points in the series. The comparable computation time using the FFT, by contrast, is proportional to N(log2 (N)). For an epoch of 1024 points (N = 1024), the DFT will take 102.4 times longer than the FFT to compute the spectral representation of the signal. 2 The Nyquist frequency, named after Henry Nyquist, is the fastest frequency that can be represented for a given sampling rate, and is equal to 1/2 the sampling rate. Nyquist, whose entire career was at AT&T Bell Laboratories, published a 1928 paper (Nyquist, 1928) in which he proposed a theorem that a sample rate twice as fast as the highest signal frequency will capture that signal perfectly. Stated differently, the highest frequency which can be accurately represented is one-half of the sampling rate, and this frequency has come to be known as the Nyquist frequency.
187
3 2.5 2
Raw
Skewness
1.5 1 0.5 0 -1.5 -1 -0.5 0
Ln-Transformed
9 8 7 6 5 4 3 2 1 0 -1 0
Kurtosis
Raw
2 Ln-Transformed
Fig. 2. Skewness statistic (top panel) and Kurtosis statistic (lower panel) for natural log transformed (X-axis) and raw (Y-axis) power values. Statistics were calculated on 34 subjects with complete resting EEG data reported in Coan and Allen (2003a) for each of 18 scalp sites (FP1, FP2, F3, F4, F7, F8, FTC1, FTC2, C3, C4, T3, T4, T5, T6, TCP1, TCP2, P3, P4) using the average reference. The solid line in each plot represents the demarcation between improvement towards normality (above the line) from greater deviation from normality (below the line) as a result of the natural-log transformation.
(e.g., Fox and Davidson, 1987). Alpha power is then taken as an index of the inverse of cortical activity (Davidson, 1988), an assumption that will be explored further below. Alpha power at any given site then is typically natural log transformed, as untransformed power values tend to be positively skewed, as depicted in Fig. 2. The top panel of Fig. 2 depicts the Skewness Statistic for the raw power values (Y-axis) and the natural-log transformed values (X-axis) at each of 18 scalp sites. The lower panel similarly depicts the Kurtosis statistic for the same data set. The solid line in each plot represents the demarcation between improvement towards normality (above the line) and greater deviation from normality (below the line) as a result of the natural-log transformation. As can be seen from the gure, the transformation improves the skewness for 89% of the scalp sites, and improves kurtosis for 83% of the scalp sites. In absolute terms, using the 95% condence intervals, prior to natural-log transformation, 94% of sites deviated signicantly from normality in terms of skewness, and 83% deviated signicantly in terms of kurtosis. Following transformation, only 33 and 39% of sites still deviated signicantly from normality in terms
188
of skewness and kurtosis, respectively. Thus although some sites still exhibit distributions of natural-log transformed scores that deviate from normality, the natural-log transformation substantially improves the distributional characteristics of the data. 2. Comparing left and right activity Because asymmetrical activity is of interest, investigators often use a difference score (ln(Right) ln(Left) alpha power) to conveniently summarize the relative activity at homologous right and left leads.3 The difference score thus provides a simple unidimensional scale representing the relative activity of the right and left hemispheres, with higher scores putatively indicative of relatively greater left frontal activity (assuming that alpha is inversely related to activity). An additional benet of this difference score metric is that it provides some degree of correction for overall alpha power, as large individual differences in overall alpha power could be confounded with the magnitude of the asymmetry. The correction for overall power stems from the fact that the natural log difference score metric is not a simple difference score, but a difference of natural log transformed scores. Rules of logarithmic subtraction state that the difference of two natural-log transformed scores is equivalent to the natural log transform of the ratio of these scores: ln(R) ln(L) = ln R L (1)
Thus this difference metric is actually the natural log transform of the ratio of right to left alpha power, which provides some degree of correction for overall power expressing each subjects asymmetry in terms of a ratio. The extent of the correction is conrmed by comparing the values of the natural log difference score metric to another sometimes-utilized metric, a normalized difference score computed as (R L)/(R + L). This normalized difference score metric correlates over 0.99 with the natural log asymmetry metric (ln(Right) ln(Left); Allen et al., 2004). There is in fact a nonlinear function relating these two metrics over a broad range of scores, because when either R or L gets very small, the normalized metric is bounded by the values 1 and 1 and the natural-log asymmetry metric will not have such bounds. Over the range of values encountered in asymmetry research, however, the function is almost perfectly linear, as illustrated in Fig. 3. The difference metric is rather handy in several respects, notably that it mitigates the impact of individual differences in skull thickness that would have sizeable inuences on recorded signal amplitude (Eshel et al., 1995; Leissner et al., 1970; Pfefferbaum, 1990), and the difference scores can simplify analyses, such as those involving correlations between frontal asymmetry (as a difference score) and an individual difference measure (e.g., Be3 Similar distributional improvements as a result of natural log transformation are seen for the asymmetry scores based on these log transformed values. Comparing asymmetry scores based on the difference of natural-log transformed and untransformed values, the transformation improves the skewness of the asymmetry score for 67% of the scalp sites, and improves kurtosis for 89% of the scalp sites. In absolute terms, using the 95% condence intervals, prior to natural-log transformation, 67% of the differences scores deviated signicantly from normality in terms of skewness, and also kurtosis, but following transformation, only 22 and 33% of asymmetry scores still deviated signicantly from normality in terms of skewness and kurtosis, respectively.
189
Fig. 3. The relationship of the asymmetry metric (ln(Right) ln(Left)) and a metric normalized for overall power ((RL)/(R+L)), over a large range of possible alpha power values. In asymmetry research, the ln(Right)ln(Left) metric produces scores that typically are in the range of 0.5, the range demarcated by the two lines, where the relationship is linear. From Allen et al. (2004), reprinted with permission from Blackwell Publishing. 2004, Society for Psychophysiological Research.
havioral Activation Scale; Coan and Allen, 2003a; Harmon-Jones and Allen, 1997; Sutton and Davidson, 1997). Difference scores have been criticized for their potential unreliability, as errors of measurement with each of the constituent scores are compounded when the difference score is calculated. The reliability of change scores is of greatest concern, however, when the constituent scores have modest reliability. Alpha power at a given lead, however, demonstrates extremely high reliability, with coefcient alpha values typically over 0.90 based on 8 min of data. Moreover, the reliability of the difference score for frontal regions has been calculated in several studies and routinely is high, excepting frontal pole sites (e.g., coefcient alphas for frontal asymmetry (difference) scores ranging from 0.85 to 0.90 at baseline in Allen, Urry, Hitt, and Coan (2004); from 0.76 to 0.91 in Coan and Allen (2003a); a median of 0.83 in Coan et al. (2001); from 0.80 to 0.93 in Reid et al. (1998); and from 0.81 to 0.92 in Tomarken et al. (1992)).4
4 A separate issue concerns the power of statistical tests that employ difference scores. The power of signicance tests using difference scores is only indirectly inuenced by the reliability of these scores. Signicance tests of differences can be powerful even if the reliability of the difference scores is near zero (Overall and Woodward, 1975; Zimmerman et al., 1993). The paradox pointed out by Overall and Woodward (1975) is that difference scores with zero reliability can in fact give rise to high power to detect a signicant difference. The paradox is resolved when one considers that reliability of the difference scores depends on the existence of variance in the difference score that can reliably rank-order individuals in terms of the magnitude of their difference scores, but that the power to detect a difference involves assessing a mean difference between the two scores relative to the variance in
190
Despite the simplicity of the difference score, the contribution of activity in each hemisphere is ultimately of interest, which will require analyses involving the examination of data from each hemisphere as a difference metric is uninformative with respect to the contribution of each constituent hemisphere (Davidson et al., 2000a). The most straightforward approach involves analyzing (ln-transformed) power at left and right sites in an analysis of variance (ANOVA) or the more general linear model (GLM), with not only region (anterior to posterior) as a factor, but hemisphere (left versus right) as well. In these models, individual differences in overall power are removed, and region specic variations in power (e.g., occipital alpha is greater than frontal alpha) are partitioned as region main effects. In such a model, with EEG power as the dependent variable, the interaction of an independent variable with hemisphere will yield the same information as a main effect of this independent variable when using asymmetry scores as the dependent variable. The follow-up tests to decompose the interaction can then examine the contribution of each hemisphere individually. The standard ANOVA approach provides a straightforward method of examining the power from each hemispheres lead or leads when the independent variable of interest is amenable to the ANOVA approach, such as when comparing depressed and nondepressed subjects, or when comparing two or more emotion elicitation conditions. This approach is limited, however, as the standard ANOVA with a between subjects or within subjects factor fails to allow for an examination of the power at a given lead or leads with a continuously varying independent variable such as ratings of emotional valence or intensity, or an individual difference variable such as behavioral activation. There exist a few published approaches that have included a continuous predictor in the model, the whole-head and homologous-lead residualized power approach rst reported by Wheeler et al. (1993), and the hierarchical general linear model strategy (e.g., Coan and Allen, 2003a) or mixed model strategy (e.g., Kline et al., 2002). 2.1. Residualized power approach Wheeler et al. (1993) adopted a two-stage analytic approach, examining rst the correlation between the asymmetry difference score and continuous measures of self-reported
this difference score. Thus if one constituent score (e.g. Left activity) were for every subject a constant k less than the other constituent score (e.g. Right activity), then there would be no variability in the difference scores, and no reliability. On the other hand, the mean difference score would be k, with no variance around that mean, allowing for a powerful statistical test that the mean difference is signicantly different than zero, and that a statistically signicant difference has been found. The pragmatic implications are that the reliability of difference scores if are of little consequence if one wishes to test the signicance of such a difference (e.g. to test that Right activity is greater than Left activity for the group as a whole), but the reliability of the difference score will be highly relevant when one is using the difference score to examine how individual differences in that difference score relate to other variables of interest (e.g. how individual differences in the asymmetry score relate to individual differences in BAS scores). In the latter case, the reliability of the difference score will impose constraints on the magnitude of the correlation that can be observed, as the maximum correlation that can be observed between two variables will be the square root of the product of the reliability of the two variables. Thus, because a sizable portion of the research examining frontal EEG asymmetry is concerned with the relationship of individual differences in frontal EEG asymmetry to other individual difference measures, the reliability of the asymmetry metric assumes great importance.
191
Table 1 Correlations between natural-log transformed alpha power at homologous leads collected for 8 min under resting conditions Sites Reference AR FP1FP2 F7F8 F3F4 FTC1FTC2 C3C4 T3T4 TCP1TCP2 P3P4 T5T6 0.997 0.983 0.990 0.975 0.977 0.918 0.944 0.965 0.907 LM 0.998 0.971 0.992 0.943 0.981 0.891 0.948 0.982 0.932
Note. AR: average reference, LM: computer linked mastoid reference; data from 34 subjects reported in Coan et al. (2001).
affect. Upon nding signicant correlations, the second stage was to investigate the contribution of each hemisphere, but unconfounded by the large individual differences in power due to irrelevant factors such as scalp thickness. Power at a given electrode (e.g., F3) was residualized, using a hierarchical regression, rst entering the average power across available scalp sites, and as the second step entering power from the homologous lead (e.g., F4). The resultant residualized values were then correlated with the variable of interest (e.g., self-reported affect). The rst step of this procedure preserves individual patterns of activity across scalp sites, adjusted for overall power. The second step of this procedure was introduced by Wheeler et al. (1993) ostensibly to statistically account for volume-conducted activity from the homologous electrode. It is unclear why one would be more concerned with volume conduction from a lead over the opposite hemisphere, which in many instances is considerably further away from the site of interest than ipsilateral leads adjacent to the site. On the other hand, the activity between homologous leads is often highly correlated, with alpha power values being correlated on the order of 0.95 or even higher (see Table 1),5 and could in part reect the dense contralateral cortico-cortical connections between some homologous regions as well as volume conduction effects. As seen in Table 1, correlations are uniformly high, but higher yet between closely spaced homologous leads (e.g., FP1 and FP2) as compared to more widely spaced homologous leads (e.g., T5 and T6). Whether volume conducted, or the result of interconnectivity, the second step of the regression approach of Wheeler et al. then statistically controlled for shared variance between left and right homologous leads, which is likely to be substantial.
5 The fact that the difference between these highly correlated sites is nonetheless predictive of state affect and individual differences merits a brief comment. The asymmetry score reects the difference between the contribution of the activity of the left and right leads within subjects, whereas the correlations between sites reect the similarities of activity at each lead across subjects. It is thus the case that between-person differences in alpha power at a given site are substantially larger than the within person differences between sites, but that the latter nonetheless have some degree of predictive validity.
192
Fig. 4. Correlations between Behavioral Activation Scale (BAS) scores and EEG asymmetry (left panels) and residualized power at constituent sites (right panels), for data under an averaged reference (AR) and computer-averaged mastoids (CAM) reference. Data from subjects presented in Coan and Allen (2003a).
The results of this procedure produce what has become a fairly typical pattern: for each signicant correlation between the R L difference score and a criterion, two signicant correlations emerge, approximately equal in magnitude to the original correlation, but opposite in sign to one another, at the constituent leads. Correlations at the right lead maintain the sign of the R L difference score, and correlations at the left lead reverse direction. For example, Wheeler et al. (1993) found that the F4F3 (ln-transformed) asymmetry score correlated with positive affect 0.45, and that residualized ln-transformed power at F4 correlated 0.44 and at F3 correlated 0.49 with positive affect. Similarly, Harmon-Jones and Allen (1998) found that the F4F3 (ln-transformed) asymmetry score correlated 0.48 with trait anger, and that residualized ln-transformed power at F4 and F3 correlated 0.45 and 0.46, respectively, with trait anger. To illustrate more generally this pattern, Fig. 4 presents correlations between BAS scores and asymmetry scores (left panels) and residualized power at constituent sites (right panels). Treating these correlations themselves data points, the obtained values from residualized power at right leads correlated 0.94 with the values obtained using the difference score, and the values from residualized power at left leads correlated 0.88 with those obtained using the difference score. Additionally, the values obtained
193
using residualized left values correlated 0.87 with those obtained using residualized right values. Thus, the procedure that was originally devised to examine the independent contributions of each hemisphere would appear to distribute the variance relatively equally and in opposite directions across the two hemispheres, which would be expected if activity at homologous leads is extremely highly correlated, as is the case with homologous left and right lead power (Table 1). To demonstrate why such a pattern would be expected with such highly correlated data, consider the impact on residualizing left hemisphere power on right hemisphere power. The residualized score (Lresid ) for a left hemisphere lead (L) is given by Lresid = L L (2)
where L is the predicted power at the left hemisphere lead given power at the right hemisphere lead, determined by the raw score regression (prediction) formula: L = a + b(R) (3)
where a is the intercept and b is the unstandardized regression coefcient. In the case where L and R are nearly perfectly positively correlated (see Table 1), with the distribution of each having virtually identical means and standard deviations, the intercept a will approach 0, and the regression coefcient b will approach one,6 reducing Eq. (3) to: L 0 + 1(R) = R (4)
Substituting the results of Eq. (4) for L in Eq. (2), it is revealed that, when L and R are nearly perfectly correlated: Lresid = L L L R (5)
Thus this residualization procedure produces residual values for left hemisphere leads that will approach the value L R as the correlation between left and right leads approaches 1.0, provided that the unstandardized regression coefcient approaches 1 and the intercept approaches 0. Similarly, by implementing Eqs. (2)(5) for right hemisphere residualized (Rresid ) and predicted (R) scores, it will be the case that residual values for right hemisphere leads will approach the value RL as the correlation between left and right leads approaches 1.0. Therefore, this procedure will make it appear that right hemisphere leads correlate with a criterion variable in the same direction and approximate magnitude as the R L difference score, and that left hemisphere leads correlate with a criterion variable in the opposite direction but same approximate magnitude as the R L difference score.
6 Empirically, it appears to be the case that the unstandardized regression weight is very close to one and the intercept is very close to zero. Resting data for 34 subjects (from Coan et al., 2001) were used to predict left hemisphere frontal activity from the homologous right hemisphere frontal activity. For the prediction of four frontal sites (FP1, F7, F3, and FTC1), each from its homologous right hemisphere site (FP2, F8, F4, and FTC2), across both LM and AR reference schemes (for a total of eight separate regressions), the median unstandardized regression coefcient was 1.028 (range: 1.0121.071) and the median intercept was 0.039 (range: 0.098 to 0.031).
194
2.2. Revised residualized power approach More recently, Davidson et al. (2000a) have proposed an improved variant on the method of Wheeler et al. (1993), one that does not include homologous lead power in the residualization calculations. This approach rst residualizes the criterion variable on whole head power, and then calculates correlations between the residualized criterion variable and power at each individual site (Davidson et al., 2000a, p. 41; Davidson, 2002, personal communication). This method obviates the problem detailed above using the homologous lead to residualize power at each site, but will produce a large set of correlations (one for each scalp site) that are not tested formally in a model that can control for experiment-wise alpha slippage. Such correlations are quite informative, but ultimately must be regarded as descriptive. To adequately test the relationship between power at each site and the criterion variable, an omnibus model is required. Although the precise model will depend on the nature of the investigation, and the theory being tested, an alternative approach might be for investigators to specify a hierarchical general linear model in testing the relationship of left and right sites to criterion variables, as highlighted below. Such an approach might limit the undesirable probabilistic artifacts involved in multiple statistical tests, optimizing risk for both type 1 and type 2 errors in estimating both the impact of whole head power and effects of interest. Further, such single model approaches may economize data analytic effort and reporting. 2.3. Hierarchical general linear models Hierarchical general linear models can simultaneously account for the multiple sources of variance contributing to the relationship between cortical asymmetry and criterion variables. Such models can include both categorical and continuous predictors, and can be constructed to test a variety of specic hypotheses of interest, including those related to overall power, hemisphere, and even reference scheme, all in a single model. In fact, interactions with reference scheme can be entered into such a model in order to determine whether relationships between asymmetry and the criterion variable are dependent upon reference scheme. In constructing the model, some general principles may guide the investigator. First, the model should be explicitly specied, and whenever possible should be an omnibus model that can test all effects of interest at once. Second, the investigator should use theory to guide the ordering of the main effects followed by the interactions of these main effects. In most cases, main effects per se will not be of interest (e.g., they may reect the contribution of overall power to the prediction of the criterion variable, or differences in overall power between anterior and posterior regions), but the interactions will be of interest. Interactions of hemisphere and region in predicting the criterion variable, for example, would be found if there are frontally-specic hemispheric differences in the contribution of left and right leads to the prediction of the criterion variable. A higher-order interaction with reference scheme would further indicate that the pattern of ndings is reference-scheme dependent. Of course, the use of theory and careful sensitivity to the possibility of spurious interaction effects is particularly important as the complexity of interactions increases. As interaction effects reach beyond third order, the probability of overtting the observed dataessentially
195
modeling meaningless residual varianceincreases. While testing for most or all effects of interest in an omnibus model is highly desirable, it is not advisable if doing so requires the modeling of very complex (e.g., greater than fourth order) interactions. Thus a third principle might be to reduce the potential for spurious ndings and complex interactions in the hierarchical linear model designed to test for specic contribution of hemisphere by rst running a simpler but conceptually related model using the asymmetry scores. Then, following this simpler model, natural-log transformed power at constituent sites can be entered for the relevant regions where the asymmetry score identied a relationship between asymmetry and the criterion variable. As an illustration, data from Coan and Allen (2003a) are presented, the same data that were used in Fig. 4 to illustrate the residualization approach. First, an omnibus hierarchical linear model using asymmetry scores from eight regions across the scalp under both averaged reference and computer averaged mastoids reference schemes were used to construct a model predicting BAS scores. To code reference scheme, data from each reference scheme were concatenated, and a contrast-coded variable was used to code for reference scheme (cf. Aiken and West, 1991). The model rst entered the main effect of reference scheme, followed by the main effects of regions ordered according to theoretical interest and results of previous studies. Sites entered rst were frontal and anterior temporal sites, followed by sites from central to parietal: F4F3, F8F7, FTC2FTC1, T4T3, C4C3, T6T5, TCP2TCP1, and P4P3. Finally, interactions of each region with reference scheme were entered to test for the reference-specic effects. In this model, only the main effect of F4F3 was signicant in predicting BAS scores (F(1, 46) = 8.5, P < 0.01), with trends for contributions from the main effects of F8F7 (F(1, 46) = 3.6, P < 0.10) and C4C3 (F(1, 46) = 3.6, P < 0.10). Reference scheme did not interact with any effects in this model. Thus the focus of the subsequent analysis was to examine the contribution of left and right hemisphere in the signicant midfrontal region. In this hierarchical general linear model, BAS scores were the dependent variable to be predicted by (1) whole head power; (2) reference scheme and (3) natural log-transformed alpha power in the left (F3) and right (F4) hemispheres. This model, with whole head power entered rst, is akin to the procedure described by Davidson et al. (2000a) to statistically partial out the effect of overall power in predicting BAS score. In this model, main effects of each site were of interest. Interactions with reference scheme were entered into the model in order to determine whether any relationships between site and BAS scores were dependent upon reference scheme.7 The overall model was approached statistical signicance (F(9, 54) = 1.94, P = 0.07, adjusted-R2 = 0.12). Results indicated a main effect of the right hemisphere at F4 (F(1, 54) = 9.61, P < 0.01, 2 = 0.15) but not of the left hemisphere at F3 (F(1, 54) = 2.49, P = 0.12, 2 = 0.04). No interactions with reference scheme were signicant, indicating that individual differences in right frontal activity were related to differences in BAS scores, and that this main effect of the right hemisphere was not dependent upon reference scheme. Fig. 5 depicts specic left/right relationships with BAS scores for F4 and F3. To estimate regression lines for both F3 and F4 separately, two hierarchical gen7 Had the rst model included additional regions of signicance, such a model could also include the interaction between hemisphere (left, right) and region (e.g. mid-frontal and lateral-frontal).
196

BAS
55 50 45 40 35 30 25 20
-2 -1 0 1 ln(Alpha Power) 2
right
left
Fig. 5. Regression lines for left and right midfrontal (F3 and F4) ln-transformed alpha power predicting BAS scores. The regression equations depicted are: BAS = (5.12) (Left) + 37.41; BAS = (1.98) (Right) + 36.658.
eral linear models were run, one each for F3 and F4. For these models, reference scheme, whole head alpha power, and each site were entered such that (1) the effects of reference scheme and whole head alpha power were each removed before the b-coefcient for each site was estimated, and (2) dependence upon both reference scheme and whole head power could be estimated. Notably, neither model is, by itself, statistically signicant, but it is nevertheless useful to estimate such curves in the service of understanding the signicant effects reported above. Apparent using this method, but not the residualization approach, is that the relationships between BAS and each hemisphere are not a mirror opposites, but in fact BAS is robustly related to right hemisphere activity, and largely not related to left hemisphere activity, a surprising result given the theoretical notions concerning the left hemisphere and approach-related motivation (Coan and Allen, 2003a,b; Davidson, 1992, 1998; Harmon-Jones and Allen, 1997).
3. Data acquisition 3.1. How much raw data should be acquired? Sufcient data are required to ensure that reliable estimates of EEG activity are derived. Although the power spectrum derived from any single epoch via the FFT will reect both frequencies that are common across epochs as well as those idiosyncratic to any given epoch, averaging together multiple spectra can allow those frequencies to emerge that are present in a reasonably large proportion of epochs (see Fig. 1, Panel E), while mitigating the inuence of infrequent or irregular signals (Nunez, 1981), which might often be considered noise. Thus an investigator, by averaging across epochs, makes the implicit assumption that the frequencies that appear commonly across epochs are of interest, and epoch-specic variations are of little interest. In the case of estimating trait asymmetry with the goal of predicting psychological traits or psychopathology, this is clearly a reasonable assumption. On the other hand, a recent investigation found that variability from epoch to epoch was itself an important correlate of neuroticism (Minnix and Kline, 2004).
197
To reliably estimate EEG asymmetry at any given assessment session, investigators and reviewers often suggest that 8 min of resting EEG asymmetry are necessary to obtain adequate internal consistency reliability, as this was the number reported in the rst psychometric investigation of resting EEG alpha asymmetry (Tomarken et al., 1992). Substantially fewer 1-min samples, however, also can produce acceptable estimates of internal consistency (Tomarken et al., 1992), and estimates based on even shorter time frames of 2 min have proven similarly reliable (Coan et al., 2001). Tomarken et al. (1992) assessed the reliability of fewer than 8 min of data in a way that confounded the length of recording with the number of discrete items included in the calculation of coefcient alpha; i.e., they used the SpearmanBrown prophecy formula to estimate the reliability for shorter recording periods, estimating alpha based on six asymmetry values for 6 min of data, seven values for 7 min, and eight values for 8 min. To adequately test whether fewer minutes of recording would produce estimates of internal consistency comparable to those obtained with more minutes of recording, it would be required to keep the number of values constant despite changes in the length of recorded data, as Cronbachs alpha will be higher given more minutes (items) for analysis (Lord and Novick, 1968). In a recent study (Allen et al., 2004), reliability estimates from 2, 4, 6, and 8 min of data were compared. Specically, the rst 2, 4, and 6 min as well as all 8 min of recorded data were divided into eight blocks each. Each block contained 2-s overlapping epochs that were subjected to Fourier analysis as reviewed above. In each case, eight asymmetry values were obtained, reecting the asymmetry score averaged across 1/8 of the total time of recording (15 s for the 2-min data, 30 s for 4-min data, 45 s for the 6-min data, and 60 s for the 8-min data). These eight values were then treated as items on an eight-item scale to assess internal consistency reliability. Fig. 6 shows the results for frontal regions as a function of reference scheme. As can be seen in the gure, the number of minutes of recording exerts relatively little inuence on the estimate of internal consistency compared to the number of blocks included in creating the estimate. Whether 2, 4, 6, or 8 min of data are utilized, very small differences are apparent when all eight data segments are used as items for the purpose of estimating internal consistency reliability. Reliability estimates begin to diverge, however, when fewer segments are utilized to estimate reliability. Thus highly internally consistent measures of asymmetry can be obtained with considerably fewer than the conventionally accepted 8 min of recorded data, provided that internal consistency is estimated with a sufcient number of constituent blocks. To highlight this point, consider a comparison of two comparable data points from Fig. 6: four 60-s blocks or eight 30-s blocks, which correspond to identical timepoints from the EEG record. In all nine cases (3 regions 3 reference schemes), the internal consistency of the latter is higher than the former, by an average of 0.06 reliability units. It also appears to be the case that when fewer than four blocks are used to estimate the reliability, the expected rank ordering of reliabilities becomes less orderly, in some cases with longer recording blocks demonstrating lower reliability than shorter blocks. Thus, regardless of the total length of data collected, attempting to estimate reliability with insufcient blocks will lead to misleading estimates of internal-consistency reliability. If investigators have fewer than 8 min of data available, reliable estimates of asymmetry can likely be derived, but it is recommended that investigators report the internal consistency reliability of asymmetry scores based on the data
198 J.J.B. Allen et al. / Biological Psychology 67 (2004) 183218 Fig. 6. Cronbachs alpha internal consistency estimates for resting alpha asymmetry as a function of region, reference scheme, length of data recording, and number of blocks (items) used to calculate alpha. The number of subjects ranges from 19 to 28, reecting that some subjects did not have enough artifact-free 2-s epochs to compute power spectra for the for shorter recording intervals, or that a recording site was bad for a given subject. Midfrontal: F4F3, lateral frontal: F8F7, and Fro-Tem-Cen: fronto-tempo-central. From Allen et al. (2004), reprinted with permission from Blackwell Publishing. 2004, Society for Psychophysiological Research.
199
available, detailing how many epochs were treated as items in the calculation of Cronbachs alpha. 3.2. What reference montage is preferred? The choice of reference has been referred to as perhaps the most divisive issue among current EEG researchers (Davidson et al., 2000a, p. 33). Although rational arguments have been levied in favor of one or another reference scheme (e.g., Hagemann et al., 2001; Reid et al., 1998), it remains an empirical question which reference scheme has the greatest predictive validity with respect to motivation, emotion, and psychopathology. Investigators would ideally like measures of spectral power at a given site to reect the activity at that site, and not at the reference lead. For this purpose, investigators often search for a relatively inactive reference, and have used linked ears or mastoids, averaged ears or mastoids, or an average reference comprised of the average of activity at all recorded EEG sites. The average reference, given a sufciently large array of electrodes in a spherical arrangement around the head, will nicely approximate an inactive reference, as activity generated from dipoles will be revealed as positivity at one site and negativity at a site 180 opposite this site, with the sum across sites thus approaching zero with a sufciently representative sample of the sphere. Smaller montages, and those that do not provide coverage approximating the sphere, however, will have more residual activity in the average reference. Especially troubling is the Cz reference, which has been utilized more often in the EEG asymmetry literature than other reference montages (see Coan and Allen, 2003b for review). The Cz reference has been criticized as potentially under- or over-estimating activity at the target site (Hagemann et al., 2001). Moreover, empirical comparisons of data from different reference schemes have found Cz to be the least related to other reference schemes (e.g., Hagemann et al., 2001; Reid et al., 1998). The fact that many studies have successfully identied predicted relationships using the Cz reference suggests at least two non-mutually exclusive possibilities: (1) signicant results using the Cz reference reect, in part, not only the relationship of constructs with frontal asymmetry, but also with sources of variance unique to the Cz reference (e.g., overall alpha power); and/or, (2) asymmetry scores using the Cz reference may have more irrelevant variance (error or systematic) with respect to asymmetry, and may therefore resultacross studiesin inconsistencies in the pattern of empirical relationships with motivation, emotion, and psychopathology. Distinguishing between these possibilities will be facilitated if investigators report results from multiple reference montages. Moreover, various reference schemes can be conceptualized as contributing unique sources of error variance to any given analysis, providing the researcher with semi-independent measures of EEG activity, with ndings that are statistically independent of reference scheme being considered the most generalizable, being less likely to reect only the reference-specic method variance (cf. Campbell and Fiske, 1959). 3.3. Impedances in asymmetry research It has been customary in EEG asymmetry research to strive to obtain low and symmetrical impedances during subject preparation. Intuitively, this seems desirable, as one would wish to have a strong noise-free signal by lowering impedance, and would like to guarantee that
200
any observed alpha power asymmetries reect an underlying asymmetry in activity, rather than an asymmetry in impendence to recording the underlying signal. As pointed out by Ferree et al. (2001), however, contemporary high impedance ampliers mitigate the impact of scalp impedances on the recorded signal, as the loss in the observed signal due to scalp impedances is directly related to the average impedance of the measurement and the reference electrode, and inversely related to the amplier input impedance. Because amplier input impedances are typically on the order of tens (or even hundreds) of megaohms, and scalp impedances on the order of a few kilohms, small changes in scalp impedance do not appreciably impact the observed signal, as the magnitude of the amplier input impedance is at least 1000 times greater than the scalp impedance. The observed voltage for a given electrode E with a given reference electrode R is the measured voltage difference between these electrodes, or VE VR . This difference, VE VR , is inuenced by electrode impedance ZE and reference impedance ZR and input impedance Zin as follows (Ferree et al., 2001, p. 538): VE VR = VD 2 Z E + ZR Zin + VC ZE Z R Zin +O 1 Zin
2
(6)
where VD is the actual differential-mode signal (VE true VR true )/2, and VC is the commonmode signal (VE true VR true )/2. The latter term VC results primarily 60 cycle (US) or 50 cycle (Europe) ambient noise, and the extent to which it emerges is a function of impedance mismatch. The former term VD is primarily the signal of interest, resulting from voltage potential differences between the two sites, but attenuated by the ratio of the scalp impedances to the amplier input impedance. Since the scalp impedances are a tiny fraction of the size of the input impedance, even appreciable differences in scalp site impedance will not measurably attenuate the voltage potential difference observed between the two sites. The nal term in the equation is a residual term to account for other sources in the differential amplier circuit that inuence the observed voltage potential difference, the sum of which are negligible (Ferree, 2002, personal communication). The mathematical notation O is standard for order in Taylor series, and simplies the expression without appreciably altering the result obtained with the simplied equation involving only the rst two terms. In the full equation there are a series of higher order terms involving powers of 1/Zin , which the term O(1/Zin )2 denotes. With a high input impedance Zin , the impact of (1/Zin )2 will be negligible, and the impact of higher powers approaches zero. Fig. 7 depicts the impact of mismatched impedances under conditions likely to be encountered in a psychophysiological laboratory. The data depicted in Fig. 7 show the observed asymmetry score (ln(Right) ln(Left)) as a function of amplier input impedance, and impedance at left (zleft ) and right (zright ) leads. Data in the left panel depict the impact of mismatched left and right lead impedances with an input impedance of 10 m (that of the Neuroscan Synamps system, Neuroscan a Compumedics Company, El Paso, TX), and data in the right panel depict the same with an input impedance of 20 m (that of the Grass Model 12 Neurodata system, Grass Telefactor an Astro-Med Inc. Product Group, West Warwick, RI). Data were obtained by solving Eq. (5) independently for left and right leads, for impedances ranging from 0.2 to 10 k , assuming a reference electrode impedance of
201
Fig. 7. Asymmetry score (difference of natural log scores) as a function of amplier input impedance, and impedance at left (zleft ) and right (zright ) leads. For both plots, reference electrode impedance is set to 1 k . The computed voltage at both left and right leads was squared to produce power units, and the difference of the natural log transformed power values (ln(Right) ln(Left)) was plotted on the vertical axis. Top two panels depict the observed asymmetry score when the right leads true signal was 5 V larger than the left, and the lower two panels depict the asymmetry score when the right leads true signal was 0.5 V larger than the left. Note that the top two panels are on the same scale, and the bottom two panels share a different scale.
1 k .8 The resultant values for left and right leads were then used to compute the asymmetry score (ln(Right) ln(Left)) for all combinations of left and right impedances. Two aspects of Fig. 7 warrant comment. First, the overall impact of impedance mismatch ranging from 0 to 10 k between left and right leads is negligible, and apparent only in the sixth decimal place of the asymmetry score. Differences between hemispheres and between groups of subjects, on the other hand, are readily apparent in the rst decimal place (cf. Henriques and Davidson, 1990, 1991; Reid et al., 1998). Differences in left and right lead
8 Results remain essentially unchanged if reference impedance is higher than 1 k , as the reference impedance appears similarly in the equation for calculating observed left lead and observed right lead voltage.
202
impedances are thus unlikely to spuriously create or mask veritable differences in left and right alpha power. Second, the impact of the mismatch is further attenuated as a function of the higher input impedance amplier. Asymmetry scores vary by 7.7 106 with the 10 M input impedance, and by 3.9 106 with the 20 M input impedance, when true voltage differences are 0.5 V. Variation would be even less with higher input impedances (such as the 200 M input impedance of the Net Amps, Electrical Geodesics Inc., Eugene, OR). 3.4. Dealing with ocular and muscular artifacts EEG recordings may contain not only brain electrical activity, but non-cerebral contributions to the observed signal, including artifactual contributions of the scalp muscles and potentials generated by eye movements and blinks (Gratton, 1998; Picton et al., 2000). Although careful screening and rejection of data segments contaminated by the artifacts is likely to remove many of the artifacts, it may be desirable to obtain an estimate of the extent to which such artifacts may be inuencing the results of an investigation. 3.4.1. Electrooculographic (EOG) inuences The eyes, being ion-lled imperfect spheres, carry a positive charge at the relatively leptokurtotic cornea, and a negative charge at the relatively platykurtotic retina. Being mobile, these charged spheres create electrical elds that are observed as signal in the case of EOG recordings, or artifact in the case of EEG recordings. Moreover, the conductive eyelid acts as a variable resistor as it slides across the cornea, momentarily distributing the ocular potential across the scalp. Thus ocular movements and blinks can be observed in scalp-recorded EEG, with the magnitude of the EOG signal decreasing as a function of the distance from the eyes. Although a majority of the signal of ocular origin is in the delta and theta range (Gasser et al., 1985; Hagemann and Naumann, 2001), slower than the 813 Hz alpha range of interest in EEG asymmetry research, some activity in the alpha band will inevitably be present, some potentially of neural origin (cf. Iacono and Lykken, 1981). The concern that activity of ocular origin may contaminate scalp-recorded EEG has prompted investigators utilizing EEG asymmetry to often reject epochs containing blinks or other ocular artifact. Moreover, the concern that the EOG signal may contain alpha-band activity of neural origin has discouraged investigators from employing a simple correction procedure that subtracts a portion of the time-domain EOG signal from the time-domain EEG signal, for doing so might also subtract alpha activity of neural origin. Hagemann and Naumann (2001) carefully examined the contribution of ocular signals to scalp-recorded EEG asymmetry scores. Reviewing the literature, Hagemann and Naumann (2001) suggested that ocular artifacts are not likely to artifactually create or mitigate alpha-band asymmetries from homologous scalp leads, because: (1) power in the alpha band that is observed in EOG recordings is predominantly neural in origin, thus making it unlikely that ocular movements and blinks will appreciably alter scalp-recorded alpha activity, and (2) eye-movements and blinks are propagated relatively symmetrically. The symmetric propagation of vertical eye movements and blinks is apparent in the raw signal, but even lateral eye movements will be reected similarly in the power spectra of left and right sites due to such movements creating similar magnitude but phase reversed deec-
203
tions; the FFT will produce similar power spectra but different phase spectra, with only the power spectra of interest in EEG asymmetry research. Assessing the contribution empirically, Hagemann and Naumann (2001) found that alpha asymmetry scores (ln(Right)ln(Left)) derived from 8 min of resting EEG were highly similar when computed with versus without epochs containing ocular artifacts. The correlations between asymmetry scores from a dataset that included all epochs free of non-ocular artifacts and the same dataset with ocular-contaminated epochs were greater than 0.82 for all regions except the frontal pole, which was substantially lower. Effects of ocular-contaminated epochs, however, were larger for single sites than for corresponding asymmetry scores, further supporting the notion that ocular artifacts propagate symmetrically across most of the scalp. Hagemann and Naumann (2001) concluded that the control of ocular artifacts may thus be unnecessary for correlational analyses involving alpha asymmetry scores, but that analyses involving mean levels may be inuenced by ocular artifacts. Although the data in support of their conclusion is relatively strongas the correlations are high between asymmetry scores from data with versus without artifactsit is worth noting two issues that were not considered or assessed fully, and that remain to be investigated empirically. First, no relationships between alpha asymmetry and a criterion variable (e.g., BAS scores, cf. Coan and Allen, 2003a; Harmon-Jones and Allen, 1997; Sutton and Davidson, 1997) were investigated. Although the high correlation between asymmetry scores obtained from data with and without ocular artifacts would suggest that each would demonstrate similar correlations to a criterion variable, it is possible that the variance that is not shared by the two sets of asymmetry scores is differentially related to the criterion variable. Because the correlation between the two sets of asymmetry scores is attenuated the most at frontal leads by the inclusion or exclusion of epochs with ocular artifacts, and because it is precisely these regions that are of greatest interest with respect to the criterion variable, the possibility is amplied that the two sets of scores may relate differentially to a criterion variable. The second issue to consider derives from this possibility. Despite the careful analysis of Hagemann and Naumann (2001), one cannot differentiate between two possibilities: (1) that the true brain activity is invariant across epochs with and without ocular artifacts, but the presence of the ocular activity inuences the observed EEG recording, or (2) the asymmetry scores differ because the true EEG activity differs as a function of whether blinks or eye movements are occurring. Given that eye-blinks show predictable relationships to cognitive processing and attention (Stern et al., 1984), this latter possibility must be considered in earnest. 3.4.2. Facial electromyographic (EMG) activity Scalp-recorded EEG alpha activity may artifactually reect the contribution of EMG activity (Cacioppo et al., 1990; Friedman and Thayer, 1991). Although the vast majority of the power in the EMG signal is faster than the alpha band, EMG activity has broad frequency characteristics with some small proportion of activity evident in the alpha band. This problem is potentially exaggerated by the fact that facial EMG asymmetriessometimes similar in direction to reported cortical EEG asymmetrieshave been observed (Borod et al., 1997), although the consistent and robust nding to emerge from this literature is an asymmetry characterized by greater left side activity in facial expressions in general, across all specic emotions and elicitation procedures.
204
Friedman and Thayer (1991) examined the potential magnitude of the EMG contribution to EEG recordings with the use of a redundancy analysis, which can be used to account for overlap between cortically derived alpha power and alpha power due to facial muscle activation. In their analysis, facial EMG accounted for 7% of the variance in cortical EEG activity, while cortical EEG activity accounted for only 3% of the variance in facial EMG, suggesting that facial EMG is likely to be responsible for a small but potentially important portion of the variance in scalp-recorded EEG. This study did not, however, specically address the extent to which asymmetries in facial EMG activity were contributing to asymmetries in scalp EEG. Coan et al. (2001) assessed the inuence of EMG on scalp recorded alpha during a directed facial action task using two strategies. The rst, a strategy used also by Davidson and colleagues (cf. Davidson, 1988; Davidson et al., 2000b), involves assessing EMG frequencies at scalp sites of interest. This approach extracts EMG frequencies (7080 Hz in Davidson et al., 2000b or 7090 Hz in Coan et al., 2001) from the power spectrum at each site involved in the EEG analysis. Because Coan et al. (2001) were analyzing EEG asymmetry scores, EMG asymmetry scores (ln(Right) ln(Left)) were computed on this EMG frequency bandfor all the same regions as were included in the EEG analysis. These EMG range asymmetries were used as changing covariates in a multivariate repeated measures analysis of covariance (MANCOVA), which assumes that the EMG covariate changes within groups with the dependent variable across levels of the independent variablein this case the particular facial expression. This changing covariate approach then correlates the change in the covariate with the change in the dependent variable and subsequently analyzes the residual variance in a standard MANOVA. Using this strategy, Coan et al. (2001) found that statistically adjusting for the EMG variance in this way did not change any of the signicant relationships between facial pose and EEG asymmetry. The second strategy used by Coan et al. (2001) involved an examination of alpha frequencies derived from bipolar EMG leads. This analysis was motivated by noting that the previous method assumes that all frequencies of the EMG are equally likely to show asymmetry effects that differ by the manipulation. But it is conceivable that EMG activity in the 7090 Hz band may relate differently to a criterion variable than EMG activity in the 813 Hz band, the band of particular interest. Thus whereas the rst analysis strategy examined EMG frequencies in EEG leads, the second examined alpha frequencies in EMG leads. This second approach derived power spectra from bipolar EMG activity in the frontalis and the temporalis muscle regions. Alpha power asymmetry scores derived from these EMG leads were thus included as changing covariates the analyses. Because unlike the rst approach, where each region had its own covariate, this approach produced solely a frontalis alpha asymmetry score and a temporalis asymmetry score, one way repeated measures MANCOVAs were conducted separately for each region (e.g., one for F4F3, one for F8F7, etc.) since each region could not have its own changing covariate. Analyzing each region separately, thus increasing the number of analyses conducted, actually provided a more stringent test of whether the relationships between the manipulation and EEG asymmetry were inuenced by myogenic contributions, because the chances increased of nding that the covariates rendered a previously signicant effect nonsignicant. Statistically controlling for the EMG variance in this way, however, did not change any of the signicant relationships between facial pose and EEG asymmetry.
205
4. Interpretive issues 4.1. Alpha and activity A guiding assumption underlying the interpretation of ndings involving frontal EEG alpha asymmetry is that greater alpha power is indicative of less cortical activity in broad underlying regions (cf. Davidson, 1988). Although there is good evidence to support this assumption, one might question whether this relationship is ubiquitous. It has been well documented that sensory input shows modality-specic blocking of alpha activity at cortical regions involved in processing such input. With visual stimuli, blocking of alpha over the occipital region occurs about 0.3 s after the presentation of the visual stimulus (Berger, 1932; Jasper and Cruickshank, 1937; Knott, 1938), but this latency has been found to vary with intensity and duration of the stimulus (Cruickshank, 1937; Durup and Fessard, 1936a,b) and to diminish somewhat with a motor response related to the stimulus (Knott, 1938, 1939; Travis et al., 1937). Recovery time from blocking is generally about 1 s, but it too varies with stimulus intensity and duration (Cruickshank, 1937; Jasper and Cruickshank, 1937; Motokawa and Tosiada, 1941). Similar but less dramatic effects are observed with other sensory modalities. Auditory stimuli, for example, block occipital alpha less effectively than visual stimuli and with a somewhat longer latency (Berger, 1930; Gibbs et al., 1935; Travis et al., 1937). Other sensory stimuli, such as tactile, cutaneous, pain (Berger, 1931, 1932; Jasper and Cruickshank, 1937; Livanov, 1940; Travis and Barber, 1938) and gustatory (Kitamura, 1939) have been found to block alpha, at least in their respective cortical areas. Thus, sensory stimulation that should require active cortical processing leads to modalityspecic alpha blocking, a principle that might lead to the inference that diminished alpha recorded over any cortical region signies greater cortical activity. A test of this hypothesis in regions other than primary sensory regions is hampered by the lack of clearly dened stimuli to specically engage those cortical regions in active processing, although several studies have provided data quite consistent with the notion that greater alpha power is indicative of less cortical activity in the underlying regions thought to subserve task performance (e.g., Davidson et al., 1990). A consideration of the genesis of the alpha rhythm might prove illuminative for the assumption that diminished alpha recorded over any cortical region signies greater cortical activity. A series of studies by Andersen and colleagues (Andersen et al., 1967a,b) suggest that thalamic rhythmicity drives cortical ensembles, the latter which comprise a large portion of scalp-recorded EEG activity. Andersen et al. (1967b) examined spindles in animals anesthetized with barbiturates, making the inference that such spontaneous rhythmic spindle activity is homologous with the human alpha rhythm. Several ndings highlight the basis of their conclusion that thalamic rhythmicity drives cortical rhythmicity, including: (1) destruction or cooling of cortical regions leaves thalamic spindle activity unchanged (Andersen et al., 1967b); (2) damage or removal of the thalamus abolished cortical spindle activity (Andersen et al., 1967b); (3) unilateral destruction of thalamic tissue resulted in the disappearance of ipsilateral cortical spindle activity (Andersen et al., 1967b); (4) synchronous cortical spindles were not observed in relatively closely spaced cortical regions (those separated by 2 mm or more), but were observed over a much larger distance
206
between a group of thalamic cells and the cortical area to which they projected (Andersen et al., 1967a). Thus Andersen and colleagues concluded that spontaneous cortical rhythmicity was generated exclusively by thalamic neurons (Andersen et al., 1967b, p. 258). Although cortical systems provide inputs to the thalamus that can disrupt the rhythmicity, the thalamus, and particularly the reticularis nucleus (Steriade et al., 1985), appears to be responsible for synchronizing cortical EEG activity. Recent conrmation of a relationship between thalamic activity and scalp-recorded alpha activity in humans derives from a positron emission tomography (PET) study (Larson et al., 1998). During approximately 30 min, EEG and [18 F]-2-uoro-2-deoxy-d-glucose (FDG) PET recordings were obtained. Global alpha (813 Hz) power was then correlated with glucose metabolism, and cortical alpha power was strongly and inversely related to glucose metabolism in the thalamus (Larson et al., 1998). This nding is consistent with the notion that thalamic activity in response to sensory or cortical input will disrupt alpha rhythmicity. Thus scalp recorded EEG alpha activityin a very coarse sense both spatially and temporallyis inversely related to thalamic activity. Global alpha power across electrodes and across 30 min relates to thalamic metabolism. Ultimately, however, investigators would wish to know whether EEG alpha at a given scalp lead is related to cortical activity in the tissue beneath that lead, a question addressed by Cook et al. (1998). Using H2 15 O PET imaging allowed them to examine activity in 2 min segments, with a total of eight such segments per subject. EEG power was calculated for 4 Hz wide bins, starting at 0 Hz and extending to 40 Hz, at 1 Hz intervals (e.g., 04, 15, 26, etc.). Cerebral perfusion under each electrode was calculated, and then correlated with each of the EEG spectral bins, resulting in a plot of correlations between EEG power and cerebral perfusion as a function of frequency. Frequency range played a major role in the relationship of EEG power with perfusion, such that lower frequencies (those bins with a center frequency below 8 Hz) had a positive relationship to perfusion, middle frequencies (bins with center frequencies from 8 to 12 Hz) had a negative relationship, and upper frequency ranges (center frequency varied depending on specic operationalization) had a positive relationship (Fig. 8). Apparent from the gure is that relative power shows a closer correspondence to underlying cortical activity than does absolute power, which may reect that the latter is confounded by variations in scalp thickness much more than the former. Additionally, although alpha power is inversely related to underlying cortical activity no matter which of the montages was used, the relationship of beta power (1330 Hz) to underlying activity varied substantially as a function of recording montage, exhibiting either a positive or negative relationship depending on the particular recording montage used. Thus there is reasonable support for the assumption that greater alpha at a scalp lead reects less cortical activity in a broad region(s) contributing to electrical activity recorded at that lead. Recent data of Cook et al. (1998) suggest, however, that a tighter correspondence between cortical activity and scalp-recorded EEG is possible with a reattribution technique these authors have called cordance, although the correspondence in the alpha band is not vastly improved using this cordance measure (see the lower panel of Fig. 8, taken from Cook et al., 1998). Whether asymmetry in reattributed power demonstrates relationships with emotion and individual differences, however, remains an empirical question, but one worth investigating given the tighter coupling of EEG to brain function that appears possible using this technique (see also Leuchter et al., 1994, 1999, 2002).
207
Fig. 8. Correlations between EEG power and PET perfusion values at tissue under the EEG scalp lead. Top panel depicts absolute power, and lower panel depicts relative power (i.e., power in the 4-Hz wide bin divided by total power across all spectral frequencies). Ear reference is a computer linked ears reference, source derivation is that described by Hjorth (1975) that weights immediate neighboring electrodes in the time domain prior to frequency-domain transformation, and reattributed power is a weighting of power derived from bipolar channels of nearest neighbors (Cook et al., 1998). Statistical signicance is indicated by horizontal lines representing the magnitude at which a correlation coefcient attains signicance: solid line for P = 0.05; large dashed line for P = 0.01; ne dashed line for P = 0.001. From Cook et al. (1998), reprinted with permission from Elsevier.
208
Worth noting from Cook et al. (1998), and as apparent in Fig. 8, is that the inverse relationship between cortical perfusion and EEG power is relatively constant throughout the entire alpha band in these adult subjects. Thus although the 813 Hz denition of alpha is somewhat arbitrary, the correspondence of activity in this range (and perhaps a bit of the 68 Hz portion of the upper Theta band) to underlying cortical activity at rest is relatively uniform in adult subjects. This observation does not suggest a strong functional distinction between smaller bandwidth divisions, such as upper or lower alpha, at rest in a psychiatrically and neurologically healthy population (although some investigators have made an argument for the utility of subdividing the alpha band for specic tasks and applications, e.g., Klimesch et al., 1997). A nal issue with respect to interpreting alpha and cortical activity concerns how to conceptualize resting EEG data. Such resting data necessarily summarizes activity across several minutes, which will collapse across many variations in brain and psychological state during the recording period. Although investigators refer to such periods as resting periods, one might alternatively think of EEG activity during these periods as task-related, with individual differences in how subjects approach this task of resting for several minutes underlying the observed individual differences in EEG activity. The resting state is relatively uncontrolled, allowing for individual differences in mentation (broadly construed) during the resting period to inuence the measure (cf. Schwartz et al., 1976 for a similar phenomenon when depressed and nondepressed subjects pondered a typical day while facial EMG was recorded). 4.2. Robustness or capitalization on chance: the impact of reference schemes, specic sites, and other variations Reviewing the literature (see Tables 14 in Coan and Allen, 2004, this issue), one is impressed by the fact that signicant relationships involving frontal EEG asymmetry: (1) derive from data analyzed under a variety of reference schemes, with different studies using different reference montages; (2) appear to involve different specic frontal regions in different studies; and (3) sometimes involve different frequency cutpoints for operationalizing alpha band activity. It is premature to know how best to interpret such a pattern of ndings, but these observations suggest at least three non-mutually exclusive possibilities. First, for those who like to see glasses as half-empty, this could reect that this research eld suffersas all do to some degreefrom signicant ination of the likelihood of Type I error, with alpha ination resulting from the poor control for multiple comparisons, compounded by the many permutations of variables possible when recording from multiple regions under multiple reference montages, with the possibility of different operationalizations of alpha-band activity. In the absence of strong theory to suggest that investigators should nd effects at one specic frontal region and not another, or under one reference scheme and not another, investigators wisely examine multiple sites and reference schemes, but the incumbent risk of this strategy is that the eld as a whole may be inadequately protected against reporting spurious ndings. Second, for those who like to see glasses as half-full, the fact that relationships with frontal EEG asymmetry appear with data from different reference schemes at different times, and at different regions at different times, suggests that these observed variables are
209
imperfect manifest representations of what should be considered a latent variable, i.e., functional frontal brain asymmetry. This argument would imply that only by recording multiple regions and under multiple reference schemes can one adequately assess an individuals true score on the latent variable. By aggregating over multiple measures, one gains power and reliability, thus enhancing the ability to nd relationships between frontal brain asymmetry and criterion variables. This argument further implies that the impact of these factors, such as reference scheme and specic frontal region, should be tested explicitly in a statistical model, such as those detailed previously in the section on hierarchical general linear models. In the absence of such a model to protect from isolated chance ndings appearing signicant, the eld may indeed suffer from ination of Type I statistical error. The latent trait argument would of course predict that the manifest variables should all show some modest correlations with the latent trait, and likely with one another. Although this is true of data recorded under the average reference and the computer averaged mastoids references, it is less true of data recorded using the Cz reference (Hagemann et al., 1998; Reid et al., 1998), which shows much lower correspondence with data using the other reference montages. It is worth noting, moreover, that by far the most common reference scheme used in frontal EEG asymmetry research is the Cz reference (see Coan and Allen, 2004, this issue), and that most studies do not employ multiple reference montages in their analysis of the data. The third possibility is that there exists some systematic relationship between measured or unmeasured variables and asymmetry at specic sites or under specic reference schemes. This line of reasoning suggests the differential engagement of various frontal systems as a function of particular task demands, as a function of factors in the experimental environment, and as a function of various individual difference variables under study. At present, it is difcult to assess the likelihood that such effects exist, as most studies do not assess a range of variables and attempt to relate them to asymmetry at specic regions. A notable exception comes from a recent study of Miller and Tomarken (2001), in which manipulations of expected reward or punishment produced changes in mid-frontal EEG asymmetry (that varied by sex), and manipulation of the required response produced changes in central asymmetry. These results suggest that there may indeed be task variables that will impact the specic region in which EEG alpha asymmetry effects are likely to appear. Nonetheless, there are many other non-task variables that may impact the regional specicity of EEG asymmetry effects, but such variables may not be known or assessed, thus making it impossible to discern whether there exists a systematic cause underlying the appearance of EEG asymmetry effects at some sites in some studies, and other sites in other studies. 4.3. Consistency and variability Estimates of EEG alpha asymmetry are averages that summarize patterns of brain activity across several minutes, either contiguous time segments in the case of resting EEG asymmetry, or collapsed across numerous discrete but separated time segments in the case of state-manipulated EEG asymmetry. Although it has been adequately demonstrated that such estimates possess excellent internal consistency reliability (e.g., Allen et al., 2001; Coan and Allen, 2003b; Coan et al., 2001; Reid et al., 1998; Tomarken et al., 1992), these estimates of internal consistency are derived from several segments of data, each of which
210
is an average reecting the pattern of brain activity across many seconds of recorded data. There is a sense in which frontal EEG alpha asymmetry derived from these segments of data ignores variability on a ner temporal scale. Whether such variability will prove meaningful is ultimately an empirical question, but the utility of explicitly examining such variability appears promising. For example, Minnix and Kline (2004) examined the variance estimate associated with the average FFT from a resting assessment. Subjects who show more secondto-second variability in frontal EEG alpha asymmetry will have higher variance estimate across the entire recording epoch. Minnix and Kline (2004) found that increased variability of this sort was related to higher neuroticism. Thus a trait characterized by greater emotional lability was found to be associated with more lability in frontal EEG asymmetry as well. Another way of assessing the stability of frontal EEG asymmetry was rst reported by Baehr et al. (1998), who computed the percentage of time that right alpha is greater than left alpha at homologous leads. Baehr et al. (1998) found that the percenttime measure better discriminated psychometrically-dened depressed subjects from nondepressed subjects than the traditional asymmetry measure that averaged across the recording period. Allen et al. (2001) used the percenttime measure as well, nding that it produced comparable ndings to the traditional asymmetry score. Thus although it may be premature to suggest that this metric has distinct advantages, the extant data suggest its promise and moreover suggest that it would not result in the elimination of signicant ndings with the traditional asymmetry score. 4.4. Keeping straight the states and the traits Substantial data support the contention that frontal EEG asymmetry can serve as a relatively stable individual difference variable, yet also show predictable state-related uctuations (see Coan and Allen, 2004, this issue). Evidence in support of the trait-like quality of frontal EEG asymmetry derives from studies specically examining stability over time. Tomarken et al. (1992) assessed the psychometric properties of trait-like frontal EEG asymmetries, nding that frontal EEG asymmetry demonstrated acceptable testretest stability (intra-class correlations ranging from 0.69 to 0.84 across 3 weeks). Similarly, Jones et al. (1997) found that frontal EEG asymmetry recorded at 3 months of age was highly correlated with asymmetry at 3 years (r = 0.66, P < 0.01). Similar gures come from Hagemann et al. (2002), who found that across four different measurement occasions, 60% of the variance in EEG asymmetry measures was due to individual differences in a temporally stable latent trait. To enhance the ability to identify trait-related variance, some studies have specically examined subjects who show the greatest cross-session consistency (e.g., Wheeler et al., 1993), reasoning that the strongest relationships to other traits should be shown by those who are consistent on the measure of trait EEG asymmetry (cf. Bem and Allen, 1974). Others have averaged data across multiple sessions to mitigate occasion-specic uctuations and presumably derive a better estimate of the trait-related variance in EEG asymmetry (e.g., Sutton and Davidson, 1997). When attempting to account for the nonstable variance in frontal EEG asymmetry, three sources must be considered: reliable changes from one session to the next, reliable and systematic changes within session, and unreliability of measurement. Because frontal EEG
211
asymmetry demonstrates high internal consistency reliability at any given assessment session (Cronbachs alphas typically above 0.95; Reid et al., 1998), it is unlikely that attenuated testretest reliability of frontal EEG asymmetry is due to random measurement error. Rather, a majority of the variance in EEG asymmetry can be accounted for reliable and systematic sources of variation (Coan and Allen, 2003b) due to: (1) Stable trait consistency across multiple assessments, which is presumably indicative of temperamental style and a tendency to respond in a characteristic way when confronted with emotionally evocative situations. An individuals trait level of frontal EEG asymmetry represents a quality of that individuala quality that the individual brings to a variety of situations and contexts. This trait level is necessarily estimated, by averaging across multiple occasions of measurement (e.g., Sutton and Davidson, 1997; Wheeler et al., 1993), by modeling it as a latent trait (e.g., Hagemann et al., 2002), or by accounting for it within the context of a generalizability analysis (as described below). (2) Occasion-specic variance refers to reliable variations in frontal asymmetry that characterize the variation in resting EEG assessments across multiple sessions of measurement. Such variation may reect systematic but unmeasured sources such as current mood, recent life events and/or factors in the testing situation. (3) State-specic variance refers to changes within a single assessment that characterize the difference between two experimental conditions or between baseline resting levels and an experimental condition. State-specic changes as conceptualized here are proximal effects in response to specic experimental manipulations. Such manipulations should be reversible and of relatively short duration. These state-related uctuations stand in contrast to the occasion-specic uctuations, which are assumed to characterize the individual throughout the measurement occasion, reecting the high internal consistency reliability estimates such measurement occasions typically show. Occasion variance is hypothesized to reect the effects of time- or context-limited individual difference variables (e.g., mood on the day of assessment, recent or imminent life events, daily hassles) or alternatively the interaction of the individual with the experimental milieu in a manner that varies from assessment to assessment (e.g., effects of experimental milieu or procedures, Blackhart et al. (2002), or experimenter effects, Kline et al. (2002)). Such effects would not be the result of purposeful state-related experimental manipulations, but would rather represent an interaction of the subject with other experimentally uncontrolled stimuli. Most studies of trait frontal asymmetry are not designed to allow for the separation of trait variance and occasion variance, as most studies entail only a single occasion of measurement of resting frontal asymmetry. If occasion-specic uctuations were not sizable, then a single assessment of trait levels would prove sufcient. Recent evidence (Hagemann et al., 2002), however, suggests that reliable occasion-specic uctuations account for approximately 40% of reliable variance in resting frontal asymmetry, while the consistency across multiple sessions, presumably reecting a stable trait, accounts for approximately 60%. Further, there may exist individual differences in the magnitude of occasion-specic uctuations. For example, Wheeler et al. (1993) selected a subset of 26 from among 81 women (i.e., 32% of the sample) who were classied as possessing stable asymmetry, meaning that 68% of the sample was classied as having unstable asymmetry.
212
4.4.1. Methods for assessing state, occasion, and trait variance To reliably delineate sources of variance in frontal EEG asymmetry, an analytic strategy for decomposing variance components is needed. One such promising general strategy is generalizability or G theory. Generalizability theory (Cronbach et al., 1972), or G-theory, was developed for the purpose of identifying the generalizability and dependability of different independent variables thought to contribute to a measures score (e.g., Di Nocera et al., 2001). The generalizability of a measure is analogous to more conventional estimates of reliability, such as the intraclass correlation, while dependability of a measure refers to a measures reliability across contexts. Generalizability and dependability estimates may be obtained for each independent variable thought to contribute variance to a measure. In practice, an independent variable with high dependability is one that contributes variance that is relatively independent of other independent variables affecting the measure of interest. For example, an estimate of the dependability of trait variance in frontal EEG asymmetry would allow one to assess how independent and stable trait variance is from state manipulations and measurement occasions. In addition to estimates of generalizability and dependability, actual variance components may be estimated for each independent variable hypothesized to contribute to an individuals score at any one time, including variance components attributable to the interaction of independent variables. G-theory is based fundamentally on an ANOVA model in the estimation of variance components. A critical difference between G-theory analyses and classical ANOVA models is that G-theory requires the computation of expected, as opposed to observed variance components. Expected variance components are estimated by using specic algorithms employed in very few statistical packages (e.g., SAS PROC VARCOMP). As applied to questions of state, occasion and trait variance in frontal EEG asymmetry, such a model might be dened as follows (cf. Di Nocera et al., 2001):
2 2 2 2 2 2 2 y = t2 + o + s + to + ts + os + tos 2 where y is the total variance for a given variable, in this case frontal EEG asymmetry, across all occasions and manipulations, t2 is the variance in frontal EEG asymmetry at2 tributable to individuals (here considered trait variance), o is the variance in frontal EEG 2 asymmetry attributable to measurement occasion, s is the variance in frontal EEG asym2 metry attributable to experimentally manipulated states, to is the variance in frontal EEG 2 asymmetry attributable to the interaction of trait and occasion variance, ts is the variance 2 in frontal EEG asymmetry attributable to the interaction of trait and state variance, os is the variance in frontal EEG asymmetry attributable to the interaction of occasion and state 2 variance, and tos is the variance in frontal EEG asymmetry attributable to the interaction of trait, occasion and state variance (confounded with error of measurement). G-theory thus provides variance component estimates (percent of variance accounted for by each component), coefcients of generalizability (2 ), as well as coefcients for dependability ( or phi), for each independent variable of interest (in this case, trait, occasion and state components). If, for example, trait variance in frontal EEG asymmetry shows high dependability in addition to high generalizability, such a nding would bolster the likelihood that it would prove useful as a liability indicator for risk for psychopathology, or index a
213
trait-like affective style, as this nding would indicate that trait frontal EEG asymmetry can be assessed reliably, and that trait frontal EEG asymmetry is independent of state manipulations and occasion-related uctuations. Further, to date no researcher has examined the stability of state manipulations in frontal EEG asymmetry, over time or otherwise. G-theory provides useful estimates of reliability that mirror and extend approaches designed to understand intraindividual dynamics, such as advocated by Mischel and colleagues (e.g., Shoda and Mischel, 1996; Shoda et al., 1994). Using Shoda and Michels approach, such patterns of behavior are represented as if . . . then . . . probabilities that vary from individual to individual and that presumably reect an individuals underlying personality type (Mendoza-Denton et al., 2001). Effects of this type are also easily accommodated by a generalizability analysis, as they would be reected 2 in ts , the trait by state interaction term. Thus, in addition to assessing the dependability of trait variance across states and occasions, the dependability of state manipulations across individuals is estimable, as is the dependability of occasion variance across individuals and states. Although no investigators have applied the model specied above to data collected across multiple occasions of measurement, Coan and Allen (2004, this issue) did assess the extent to which state changes in frontal EEG asymmetry were reliably elicited across subjects, and the extent to which trait levels of frontal EEG asymmetry were preserved across state manipulations. The results indicated that trait-specic variance, state-specic variance, as well as variance attributable to their interaction, each accounted for approximately 10% of the total explained variance in frontal EEG asymmetry. Trait stability as measured by the g coefcient (intraclass correlation) was estimated to be moderately high (0.47), whereas state stability was extremely high (g coefcient = 0.92). Although these results identify stability and sizable contributions of both trait and state frontal EEG asymmetry, trait variance as estimated from this single measurement occasion will necessarily include both stable trait and occasion specic inuences. Indeed, while the state variance in response to the manipulation was highly stable, the trait variance was only moderately so. This may be due to the inuence of unmeasured but relevant occasion specic factors, which future efforts might protably explore. A nal note with respect to the G-theory approach concerns its exibility to assess the impact of a variety of other factors, such as the effect of reference scheme and specic frontal region. By including terms to account for variance due to particular reference scheme, or to the particular frontal region (e.g., F4F3 versus F8F7 versus FTC2FTC1), the magnitude of these sources of variance and their interactions with trait, occasion, and state variance can be assessed. Similarly, estimates of the stability of the effects across these factors can be quantitatively assessed.
5. Synopsis Research on frontal EEG asymmetry and emotion now represents a substantial body of literature. There are numerous methodological issues to which the eld may have paid insufcient attention, while at the same time paying potentially too much attention to other factors. The eld may have been too concerned with recording at least 8 min of data to ob-
214
tain reliable estimates of asymmetry, overly concerned about the impact of blink artifacts, and overly concerned with closely matching impedances at homologous leads. By contrast, too little concern has generally been given to assessing the impact of reference scheme, disentangling left from right hemisphere effects using appropriate statistical models, and discerning whether specic regions are differentially involved in various tasks or as a function of individual differences. It remains to be determined whether the impact of myogenic activity substantially inuences ndings involving EEG alpha asymmetry. As one reviews the frontal EEG asymmetry and emotion literature (Coan and Allen, 2003a; 2004, this issue), it is apparent that many different data analytic approaches have been used, resulting in a collection of ndings that converge despite rather dramatic differences in: (1) the conditions under which data were recorded; (2) the manner in which data were reduced; and (3) the manner in which data were subsequently analyzed. The optimist will see this as a testament to the robustness of the underlying systems reected in frontal EEG asymmetry, and the curmudgeon will see this as representing considerable literature-wide alpha slippage due to the many permutations of data reduction and analysis. A conservative intermediate interpretation is that the larger enterprise of interpreting the data and theory building will benet from a more solid empirical foundation, one that will require that careful attention be given to EEG data recording and analysis. The issues highlighted here may best be regarded as fundamentals that may inform future efforts, to assist in the creation of a more methodologically consistent and precise data base. Only with such a foundation can researchers then explore the underlying functional, anatomical and neurochemical systems that may be tapped by frontal EEG asymmetry.
Acknowledgements This research was supported, in part, by a grant from the National Alliance for Research on Schizophrenia and Depression (NARSAD) and an Exploratory/Development Grant from the National Institutes of Health (1 R21 RR09492) to J.J.B.A., and by a Graduate Research Fellowship from the National Science Foundation to J.A.C.
References
Aiken, S.L., West, S.G., 1991. Multiple Regression: Testing and Interpreting Interactions. Sage Publications, Thousand Oaks, CA. Allen, J.J.B., Harmon-Jones, E., Cavender, J.H., 2001. Manipulation of frontal EEG asymmetry through biofeedback alters self-reported emotional responses and facial EMG. Psychophysiology 38, 685693. Allen, J.J.B., Urry, H.L., Hitt, S.K., Coan, J.A., 2004. The stability of resting frontal electroencephalographic asymmetry in depression. Psychophysiology 41, 269280. Andersen, P., Andersson, S.A., Lomo, T., 1967a. Nature of thalamo-cortical relations during spontaneous barbiturate spindle activity. Journal of Physiology 192, 283307. Andersen, P., Andersson, S.A., Lomo, T., 1967b. Some factors involved in the thalamic control of spontaneous barbiturate spindles. Journal of Physiology 192, 257281. Baehr, E., Rosenfeld, J.P., Baehr, R., Earnest, C., 1998. Comparison of two EEG asymmetry indices in depressed patients vs. normal controls. International Journal of Psychophysiology 31 (1), 8992.
215
Bem, D.J., Allen, A., 1974. On predicting some of the people some of the time: the search for cross-situational consistencies in behavior. Psychological Review 81 (6), 506520. Berger, H., 1930. Uber das elektrenkephalogramm des Menschen II. Journal fur Psychologie und Neurologi 40, 160179. Berger, H., 1931. Uber das elektrenkephalogramm des Menschen III. Archiv fur Psychiatrie und Nervenkrankheiten 94, 1660. Berger, H., 1932. Uber das elektrenkephalogramm des Menschen IV. Archiv fur Psychiatrie und Nervenkrankheiten 97, 626. Blackhart, G.C., Kline, J.P., Donohue, K.F., LaRowe, S.D., Joiner, T.E., 2002. Affective responses to EEG preparation and their link to resting anterior EEG symmetry. Personality & Individual Differences 32 (1), 167174. Borod, J.C., Haywood, C.S., Koff, E., 1997. Neuropsychological aspects of facial asymmetry during emotional expression: a review of the normal adult literature. Neuropsychology Review 7, 4160. Cacioppo, J.T., Tassinary, L.G., Fridlund, A.J., 1990. The skeletomotor system. In: Cacioppo, J.T., Tassinary, L.G. (Eds.), Principles of Psychophysiology. Cambridge University Press, New York, NY. Campbell, D.T., Fiske, D.W., 1959. Convergent and discriminant validation by the multitraitmultimethod matrix. Psychological Bulletin 56, 81105. Coan, J.A., Allen, J.J.B., 2003a. Frontal EEG asymmetry and the behavioral activation and inhibition systems. Psychophysiology 40, 106114. Coan, J.A., Allen, J.J.B., 2003b. The state and trait nature of frontal EEG asymmetry in emotion. In: Hugdahl, K., Davidson, R.J. (Eds.), The Asymmetrical Brain, second ed. MIT Press, Cambridge, MA, pp. 565615. Coan, J.A., Allen, J.J.B., 2004. Frontal EEG asymmetry as a moderator and mediator of emotion. Biological Psychology 67, 749. Coan, J.A., Allen, J.J.B., Harmon-Jones, E., 2001. Voluntary facial expression and hemispheric asymmetry over the frontal cortex. Psychophysiology 38, 912925. Cook, I.A., OHara, R., Uijtdehaage, S.H., Mandelkern, M., Leuchter, A.F., 1998. Assessing the accuracy of topographic EEG mapping for determining local brain function. Electroencephalography & Clinical Neurophysiology 107 (6), 408414. Cronbach, L.J., Gleser, G.C., Nanda, H., Rajaratnam, N., 1972. The Dependability of Behavioral Measurements: Theory of Generalizability of Scores and Proles. Wiley, New York. Cruickshank, R.M., 1937. Human occipital brain potentials as affected by intensity-duration variables of visual stimulation. Journal of Experimental Psychology 21, 625641. Davidson, R.J., 1988. EEG measures of cerebral asymmetry: conceptual and methodological issues. International Journal of Neuroscience 39 (12), 7189. Davidson, R.J., 1992. Anterior cerebral asymmetry and the nature of emotion. Brain & Cognition 20 (1), 125151. Davidson, R.J., 1998. Affective style and affective disorders: perspectives from affective neuroscience. Cognition & Emotion 12 (3), 307330. Davidson, R.J., 22 and 24 September 2002. Personal communication. Davidson, R.J., Chapman, J.P., Chapman, L.J., Henriques, J.B., 1990. Asymmetrical brain electrical activity discriminates between psychometrically-matched verbal and spatial cognitive tasks. Psychophysiology 27 (5), 528543. Davidson, R.J., Jackson, D.C., Larson, C.L., 2000a. Human electroencephalography. In: Cacioppo, J.T., Tassinary, L.G., Berntson, G.G. (Eds.), Handbook of Psychophysiology, second ed. Cambridge University Press, Cambridge, UK, pp. 2752. Davidson, R.J., Marshall, J.R., Tomarken, A.J., Henriques, J.B., 2000b. While a phobic waits: regional brain electrical and autonomic activity in social phobics during anticipation of public speaking. Biological Psychiatry 47 (2), 8595. Di Nocera, F., Ferlazzo, F., Borghi, V., 2001. G theory and the reliability of psychophysiological measures: a tutorial. Psychophysiology 38, 796806. Durup, G., Fessard, A., 1936a. Lelectrencephalogramme de lhomme. Donnees quantitatives sure larret provoque par des stimuli visuels ou auditifs. Comptes Rendus des Seances de la Societe de Biologie et de ses Filiales 122, 756758. Durup, G., Fessard, A., 1936b. Lelectrencephalogramme de lhomme observations psychophysiologiques relative a laction des stimuli visuels et auditifs. Annee Psychologie 36, 132.
216
Eshel, Y., Witman, S., Rosenfeld, M., Abboud, S., 1995. Correlation between skull thickness asymmetry and scalp potential estimated by a numerical model of the head. IEEE Transactions on Biomedical Engineering 42 (3), 242249. Ferree, T.C., 27 November 2002. Personal communication. Ferree, T.C., Luu, P., Russell, G.S., Tucker, D.M., 2001. Scalp electrode impedance, infection risk and EEG data quality. Clinical Neurophysiology 112 (3), 536544. Fox, N.A., Davidson, R.J., 1987. Electroencephalogram asymmetry in response to the approach of a stranger and maternal separation in 10 month old infants. Developmental Psychology 23, 233240. Friedman, B.H., Thayer, J.F., 1991. Facial muscle activity and EEG recordingsredundancy analysis. Electroencephalography and Clinical Neurophysiology 79, 358360. Gasser, T., Sroka, L., Mocks, J., 1985. The transfer of EOG activity into the EEG for eyes open and closed. Electroencephalography & Clinical Neurophysiology 61 (2), 181193. Gibbs, F.A., Davis, H., Lennox, W.G., 1935. The electro-encephalogram in epilepsy and in conditions of impaired consciousness. Archives of Neurological Psychiatry 34, 11331148. Glaser, E.M., Ruchkin, D.S., 1976. Principles of Neurobiological Signal Analysis. Academic Press, New York. Gratton, G., 1998. Dealing with artifacts: the EOG contamination of the event-related brain potential. Behavior Research Methods, Instruments, & Computers 30 (1), 4453. Gratton, G., 2000. Biosignal processing. In: Cacioppo, J.T., Tassinary, L.G., Berntson, G.G. (Eds.), Handbook of Psychophysiology, second ed. Cambridge University Press, New York, pp. 900923. Hagemann, D., Naumann, E., 2001. The effects of ocular artifacts on (lateralized) broadband power in the EEG. Clinical Neurophysiology 112 (2), 215231. Hagemann, D., Naumann, E., Becker, G., Maier, S., Bartussek, D., 1998. Frontal brain asymmetry and affective style: a conceptual replication. Psychophysiology 35 (4), 372388. Hagemann, D., Naumann, E., Thayer, J.F., 2001. The quest for the EEG reference revisited: a glance from brain asymmetry research. Psychophysiology 38, 847857. Hagemann, D., Naumann, E., Thayer, J.F., Bartussek, D., 2002. Does resting electroencephalograph asymmetry reect a trait? An application of latent statetrait theory. Journal of Personality & Social Psychology 82 (4), 619641. Harmon-Jones, E., Allen, J.J.B., 1997. Behavioral activation sensitivity and resting frontal EEG asymmetry: covariation of putative indicators related to risk for mood disorders. Journal of Abnormal Psychology 106 (1), 159163. Harmon-Jones, E., Allen, J.J.B., 1998. Anger and frontal brain activity: EEG asymmetry consistent with approach motivation despite negative affective valence. Journal of Personality & Social Psychology 74 (5), 13101316. Henriques, J.B., Davidson, R.J., 1990. Regional brain electrical asymmetries discriminate between previously depressed and healthy control subjects. Journal of Abnormal Psychology 99 (1), 2231. Henriques, J.B., Davidson, R.J., 1991. Left frontal hypoactivation in depression. Journal of Abnormal Psychology 100 (4), 535545. Hjorth, B., 1975. An on-line transformation of EEG scalp potentials into orthogonal source derivations. Electroencephalography & Clinical Neurophysiology 39 (5), 526530. Iacono, W.G., Lykken, D.T., 1981. Two-year retest stability of eye tracking performance and a comparison of electro-oculographic and infrared recording techniques: evidence of EEG in the electro-oculogram. Psychophysiology 18 (1), 4955. Jasper, H.H., Cruickshank, R.M., 1937. Electro-encephalography: II. Visual stimulation and the after-image as affecting the occipital alpha rhythm. Journal of General Psychology 17, 2948. Jones, N.A., Field, T., Davalos, M., Pickens, J., 1997. EEG stability in infants/children of depressed mothers, Child Psychiatry. Child Psychiatry & Human Development 28 (2), 5970. Kitamura, K., 1939. Die elektrencephalographische untersuchung der gerschmacksempndlichkeit. Tohoku Psychologica Folia 7, 1332. Klimesch, W., Doppelmayr, M., Pachinger, T., Ripper, B., 1997. Brain oscillations and human memory: EEG correlates in the upper alpha and theta band. Neuroscience Letters 238 (12), 912. Kline, J.P., Blackhart, G.C., Joiner, T.E., 2002. Sex, lie scales, and electrode caps: an interpersonal context for defensiveness and anterior electroencephalographic asymmetry. Personality & Individual Differences 33 (3), 459478.
217
Knott, J.R., 1938. Reduced latent time of blocking of the Berger rhythm to light stimuli. Proceedings of the Society for Experimental Biology and Medicine 38, 216217. Knott, J.R., 1939. Some effects of mental set upon the electrophysiological processes of the human cerebral cortex. Journal of Experimental Psychology 24, 384405. Larson, C.L., Davidson, R.J., Abercrombie, H.C., Ward, R.T., Schaefer, S.M., Jackson, D.C., 1998. Relations between PET-derived measures of thalamic glucose metabolism and EEG alpha power. Psychophysiology 35 (2), 162169. Leissner, P., Lindholm, L.E., Petersen, I., 1970. Alpha amplitude dependence on skull thickness as measured by ultrasound technique. Electroencephalography & Clinical Neurophysiology 29 (4), 392399. Leuchter, A.F., Cook, I.A., Lufkin, R.B., Dunkin, J., Newton, T.F., Cummings, J.L., 1994. Cordance: a new method for assessment of cerebral perfusion and metabolism using quantitative electroencephalography. Neuroimage 1 (3), 208219. Leuchter, A.F., Cook, I.A., Witte, E.A., Morgan, M., Abrams, M., 2002. Changes in brain function of depressed subjects during treatment with placebo. American Journal of Psychiatry 159 (1), 122129. Leuchter, A.F., Uijtdehaage, S.H., Cook, I.A., OHara, R., Mandelkern, M., 1999. Relationship between brain electrical activity and cortical perfusion in normal subjects. Psychiatry Research 90 (2), 125140. Livanov, M.N., 1940. Rhythmical stimuli and the interrelation between the areas of the cerebral cortex. Journal of Physiology 28, 172194. Lord, F.M., Novick, M.R., 1968. Statistical Theories of Mental Test Scores. Addison-Wesley, Reading, MA. Mendoza-Denton, R., Ayduk, O., Mischel, W., Shoda, Y., Testa, A., 2001. Person situation interactionism in self-encoding (I am . . . when . . . ): implications for affect regulation and social information processing. Journal of Personality & Social Psychology 80 (4), 533544. Miller, A., Tomarken, A.J., 2001. Task-dependent changes in frontal brain asymmetry: effects of incentive cues outcome expectancies and motor responses. Psychophysiology 38, 500511. Minnix, J.A., Kline, J.P., 2004. Neuroticism predicts resting frontal EEG asymmetry variability. Personality and Individual Differences 36, 823832. Motokawa, K., Tosiada, M., 1941. Die elektrenkephalographische untersuchung uber den adaptationsmechanismus des zentralnervensystems. Japanese Journal of Medical, Science and Biophysics 7, 213233. Nunez, P., 1981. Electrical Fields of the Brain: The Neurophysics of EEG. Oxford University Press, New York. Nyquist, H., 1928. Certain topics in telegraph transmission theory. Transactions of the American Institute of Electrical Engineers 47, 617644. Overall, J.E., Woodward, J.A., 1975. Unreliability of difference scores: a paradox for measurement of change. Psychological Bulletin 82, 8586. Pfefferbaum, A., 1990. Model estimates of CSF and skull inuences on scalp-recorded ERPs. Alcohol 7 (5), 479482. Picton, T.W., Bentin, S., Berg, P., Donchin, E., Hillyard, S.A., Johnson Jr., R., 2000. Guidelines for using human event-related potentials to study cognition: recording standards and publication criteria. Psychophysiology 37 (2), 127152. Reid, S.A., Duke, L.M., Allen, J.J.B., 1998. Resting frontal electroencephalographic asymmetry in depression: inconsistencies suggest the need to identify mediating factors. Psychophysiology 35 (4), 389404. Reilly, E.L., 1987. EEG recording and operation of the apparatus. In: Niedermeyer, E., Silva, F.H.L.D. (Eds.), Electroencephalogaphy: Basic Principles, Clinical Applications, and Related Fields, second ed. Urban & Schwarzenberg, Inc., Baltimore, pp. 5777. Schwartz, G.E., Fair, P.L., Salt, P., Mandel, M.R., Klerman, G.L., 1976. Facial muscle patterning to affective imagery in depressed and nondepressed subjects. Science 192 (4238), 489491. Shoda, Y., Mischel, W., 1996. Toward a unied, intra-individual dynamic conception of personality. Journal of Research in Personality Special Issue: The Future of Personality 30 (3), 414428. Shoda, Y., Mischel, W., Wright, J.C., 1994. Intraindividual stability in the organization and patterning of behavior: incorporating psychological situations into the idiographic analysis of personality. Journal of Personality & Social Psychology 67 (4), 674687. Steriade, M., Deschenes, M., Domich, L., Mulle, C., 1985. Abolition of spindle oscillations in thalamic neurons disconnected from nucleus reticularis thalami. Journal of Neurophysiology 54 (6), 14731497. Stern, J.A., Walrath, L.C., Goldstein, R., 1984. The endogenous eyeblink. Psychophysiology 21 (1), 2233.
218
Sutton, S.K., Davidson, R.J., 1997. Prefrontal brain asymmetry: a biological substrate of the behavioral approach and inhibition systems. Psychological Science 8 (3), 204210. Tomarken, A.J., Davidson, R.J., Wheeler, R.E., Kinney, L., 1992. Psychometric properties of resting anterior EEG asymmetry: temporal stability and internal consistency. Psychophysiology 29 (5), 576592. Travis, L.E., Barber, V., 1938. The effect of tactile stimulation upon the Berger rhythm. Journal of Experimental Psychology 22, 269272. Travis, L.E., Knott, J.R., Grifth, P.E., 1937. Effect of response on the latency and frequency of the Berger rhythm. Journal of General Psychology 16, 391401. Wheeler, R.E., Davidson, R.J., Tomarken, A.J., 1993. Frontal brain asymmetry and emotional reactivity: a biological substrate of affective style. Psychophysiology 30 (1), 8289. Zimmerman, D.W., Williams, R.H., Zumbo, B.D., 1993. Reliability of measurement and power of signicance tests based on differences. Applied Psychological Measurement 17, 19.
Commentary
What does the prefrontal cortex do in affect: perspectives on frontal EEG asymmetry research
Richard J. Davidson
Laboratory for Affective Neuroscience, W.M. Keck Laboratory for Functional Neuroimaging and Behavior, University of Wisconsin-Madison, 1202 West Johnson Street, Madison, WI 53706, USA
Abstract This commentary provides reections on the current state of affairs in research on EEG frontal asymmetries associated with affect. Although considerable progress has occurred since the rst report on this topic 25 years ago, research on frontal EEG asymmetries associated with affect has largely evolved in the absence of any serious connection with neuroscience research on the structure and function of the primate prefrontal cortex (PFC). Such integration is important as this work progresses since the neuroscience literature can help to understand what the prefrontal cortex is doing in affective processing. Data from the neuroscience literature on the heterogeneity of different sectors of the PFC are introduced and more specic hypotheses are offered about what different sectors of the PFC might be doing in affect. A number of methodological issues associated with EEG measures of functional prefrontal asymmetries are also considered. 2004 Elsevier B.V. All rights reserved.
Keywords: Frontal EEG asymmetry; Emotion; Prefrontal cortex; Neuroimaging
1. Introduction It is very gratifying for me personally to be writing this commentary since the publication of this Special Issue marks the fact that there now exists a substantial corpus of scientic literature investigating methodological and conceptual issues surrounding the use of measures of asymmetric prefrontal electrical signals recorded from the scalp surface to make inferences about emotional processes. With my colleagues and students at the time, I rst reported 25 years ago on the use of asymmetries in scalp-recorded frontal brain
Tel.: +1-608-262-8972; fax: +1-608-265-2875. E-mail address: rjdavids@wisc.edu (R.J. Davidson).
220
R.J. Davidson / Biological Psychology 67 (2004) 219233
electrical activity to make inferences about emotional processes (Davidson et al., 1979). At the time, it was a lonely eld though there were investigators from other domains of neuroscience who underscored the important role played by the prefrontal cortex in different aspects of emotional processes (e.g., Nauta, 1971). This Special Issue has brought together the best of current research using these non-invasive measures to investigate different aspects of the methodology and construct validity of these measures. When I began to work in this area, research on the central nervous system substrates of emotional processes was largely restricted to studies in non-human species that focused on the role of subcortical structures in emotional and motivational processes, leading to a view, still championed by some (e.g., Panksepp, 2003; but see Davidson, 2003a for a rebuttal), that the fundamental circuitry for emotion and motivation lies in subcortical zones and that cortical tissue has little if anything directly to do with emotion. The only exceptions to this view at the time came from observations of patients with localized cortical brain damage (e.g., Gainotti, 1972) and from early studies on the role of prefrontal lesions on the socioemotional behavior of monkeys (e.g., Myers, 1972). The research featured in this Special Issue is part of a larger body of work (see e.g., Rolls, 1999 for a modern example) that forcefully underscores the importance of prefrontal cortex for emotional and motivational processes. However, the work represented in this Special Issue has, for the most part, evolved with little connection to the core neuroscience research on the structure and function of the prefrontal cortex. I will argue that this state of affairs must change if this work is to become an accepted part of the inuential body of neuroscientic research on the prefrontal cortex. Just what specic role the primate prefrontal cortex (PFC) might be playing in emotion still remains elusive and will be one of the several issues I will address in some detail below. The substantive portions of this commentary will be divided into two parts. The rst part will address some of the fundamental conceptual issues that lie at the core of this work, most of which were not addressed in the articles in this Special Issue. I will focus on what the PFC is doing in emotion and will draw on other neuroscience literature on PFC from which insights about its role in emotion might be gleaned. In this section, I will also emphasize the fact that the PFC is heterogeneous, both anatomically and functionally and any serious discussion of PFC function must make distinctions among the sub-territories of the PFC. Also to be emphasized is the fact that the PFC is part of a larger overall circuit and that other components of the circuitry are crucial for understanding how the brain implements emotional and motivation processes. It cannot be solely through the prefrontal cortex that the brain implements emotional and motivational processes. One unwitting consequence of the work my laboratory began on prefrontal asymmetries and affect is the view that the prefrontal cortex is the center or at least the primary region for all aspects of the emotional and/or motivational processes in question. It is essential that we remind ourselves that the prefrontal cortex is part of a larger and more complex circuit and that other components of this circuitry will undoubtedly be important for many of psychological phenomena of interest to the readers of this Special Issue. A related issue concerns the nature of the measure with which we are dealing. It has been easy for those who study EEG frontal asymmetry to ignore other components of the circuitry because the measure in question here does not reect many of the other features of the circuit. In particular, scalp-recorded brain electrical signals reect activity primarily from cortex and without more complex source localization
221
methods which have not been used in any of the articles in this Special Issue, one cannot make inferences about subcortical signals.
2. Conceptual issues that underlie research on EEG asymmetries associated with emotion 2.1. Heterogeneity of prefrontal cortex Many recent reviews have documented the important functional and anatomical divisions within the prefrontal cortex (e.g., Rolls, 1999). For the purposes of the present discussion, it is important to simply call attention to this fact and to indicate that the majority of the neuroscience literature indicates that the prefrontal sector most directly associated with emotion is the sector that is least likely to be reected in scalp-recorded brain electrical signalsnamely, orbital frontal cortex (e.g., Hornak et al., 2003). Both the lateral orbital sectors and the ventromedial sectors have been directly implicated in various aspects of emotion, with the orbital sectors in particular thought to play an important role in the assignment of affective value to stimuli (e.g., Rolls et al., 2003). It is true that there are major interconnections among the different sectors of the PFC and that the dorsolateral sector receives input from the orbital sector (Rolls, 1999). However, according to most investigators, the dorsolateral sector of the PFC is primarily associated with various cognitive processes, particularly with cognitive control (see Miller and Cohen, 2001 for review). It is the dorsolateral sector which is likely most directly reected in the scalp signals from which metrics of functional prefrontal asymmetry are constructed. Recent neurophysiological data from awake behaving rhesus macaques in whom single cell recordings were simultaneously acquired from multiple neurons in both dorsolateral and orbitofrontal sectors of PFC (Wallis and Miller, 2003) provide a new understanding of what these different regions of PFC may be doing and how they interact. These investigators recorded single cell activity as well as behavior during a reward preference task. The monkeys had to choose between pictures associated with different amounts of juice reward. Neuronal activity in both regions of PFC reected the reward amount. However, neurons in the dorsolateral sector encoded both the reward amount and the monkeys forthcoming response while neurons in the orbital sector primarily encoded the reward amount alone. The authors propose that their data are consistent with a model where reward information enters the PFC via the orbital sector where it is then passed on to the dorsolateral sector and used to guide behavior. Unfortunately since neurophysiological studies of this kind are so difcult to perform and since there were only two animals tested, it was not possible for the authors to compare left- versus right-side recordings. However, one important conclusion to draw from this work is that tasks that include a response component will be more likely to show affect-related PFC activation asymmetry in the dorsolateral regions and it is activity in these regions that are most likely to be reected in scalp-recorded brain electrical signals. Recent studies have also identied a ventrolateral region of prefrontal cortex that shows robust asymmetries during response inhibition tasks. In response to cues that signal the requirement to inhibit a dominant response, activation of right inferior and middle frontal gyrus has been found (e.g., Garavan et al., 1999; Konishi et al., 1999). Moreover, patients
222
with focal damage to the right inferior frontal gyrus show impairments on such tasks (Aron et al., 2003). Right ventromedial PFC also appears to play a specialized role in the inhibition of impulsive affective urges. Patients with damage specically to this PFC sector on the right side, but not on the left side, have profound abnormalities in emotion-related decision making (such as on the Iowa Gambling task; Tranel et al., 2002; see also Clark et al., 2003), making more disadvantageous choices compared with their left-damaged counterparts. It appears that this sector of right prefrontal cortex may be particularly sensitive to punishment so that when it is damaged, patients no longer have the usual cues that signal threat and danger and so tend to act impulsively. It will be important in future research using EEG measures to utilize tasks that have been well-studied in imaging and lesion studies to determine if predictable changes in PFC electrical asymmetries occur. 2.2. PFC is part of a complex circuit involving other cortical and subcortical components Emotion and motivation are instantiated in complex circuitry involving both cortical and subcortical components. Most major neuroscientists working on these topics have reached this conclusion (see Rolls, 1999; LeDoux, 1996; Damasio, 1994; Davidson et al., 2000c; Davidson et al., 2003). For example, we (Davidson et al., 2003) have described a circuit that primarily involves different sectors of the PFC, the amygdala, hippocampus, insula and anterior cingulate. Each of these structures plays a different, complementary role in specic features of emotion. Just as cognitive neuroscience has powerfully demonstrated the utility of differentiating among specic subcomponents of cognitive processes, so it is essential that we do the same with emotion. Toward the end of this section, we will explicitly consider what different sectors of the PFC might be doing in emotion and suggest different functions for the different sectors. Unfortunately though, little work at the human level has begun to parse emotion in a fashion that derives from and honors the distinctions that are made at the neural level. This will be an important task for future research. With respect to the frontal EEG asymmetry literature, it is imperative that we be mindful of the fact that PFC, and particularly that sector of PFC likely contributing most to the brain electrical signals that are the subject of this Special Issue, represents only a small portion of the critical circuitry of emotion. Thus, for certain types of emotional processes, the presence of a particular pattern of functional prefrontal asymmetry may be necessary but not sufcient for the emotional state in question, or may simply be a contributory cause of the emotional state. For example, as Harmon-Jones (this issue) has demonstrated, certain types of anger under specic conditions may be associated with relative left prefrontal activation but this does not mean that the left prefrontal cortex is in any way a center for anger in the brain. We must ask what component of anger might be represented in the left prefrontal cortex and since left prefrontal activation occurs during other types of emotions, we must further ask what other brain regions must be recruited for anger to occur. Without additional measures that reect other components of the circuitry, it is not possible to effectively address the question of what the PFC is uniquely doing in emotion. When the patterns of anatomical connectivity are examined, it is apparent that one major target of prefrontal neurons from some PFC sectors, is the amygdala. Some researchers have suggested, based upon this anatomical arrangement, that one function of at least certain regions of PFC is to modulate or inhibit activity in the amygdala. Davidson (2000, 2002) has
223
specically suggested that regions of the left PFC in particular may play an important role in inhibiting the amygdala. Broadly consistent with this view are new data at the animal level. For example, Quirk et al. (2003) have shown that stimulation of medial PFC in rodents decreases the responsiveness of output neurons in the central nucleus of the amygdala. Caution must be exercised in generalizing from rodent to human PFC but nevertheless, these data raise the important suggestion that reciprocal relations between at least certain PFC regions and the amygdala may be important in understanding the functional signicance of this emotion-related circuitry. Using a paradigm to study the voluntary regulation of emotion modeled after one we developed (Jackson et al., 2000), Ochsner et al. (2002) reported strong inverse relations between activation in the left ventrolateral PFC and the amygdala when subjects were requested to voluntarily downregulate their negative affect. These ndings collectively imply that what the PFC is doing in emotion is clearly not mediating emotional responses but rather moderating patterns of activity in other parts of the circuit that control the primary emotional response. 2.3. Should measures of emotional experience be used to guide and test theory? An issue that surfaces in several of the articles in this Special Issue and one that was noted by Cacioppo (this issue) in his commentary concerns the role of emotional experience. Most, though certainly not all of the extant research on frontal EEG asymmetry and emotion is based upon associations with self-report measures of emotional experience or of emotional traits, our own work included (e.g., Tomarken et al., 1992; Sutton and Davidson, 1997). While there are some important, notable exceptions to this, including some of our own work examining relations with hormonal and immune measures (e.g., Davidson et al., 1999; Kalin et al., 1998; Rosenkrantz et al., 2003), much of the literature is based upon associations with self-report. If we are to uncover what the PFC is doing in emotion, it will be important that we develop laboratory probes that do not rely solely upon self-report measures to provide information on aspects of affective processing that may not be directly reected in conscious report. A recent example of this is a new study from our lab that investigated relations between baseline functional prefrontal asymmetry derived from brain electrical measures and startle responses during and following unpleasant pictures (Jackson et al., 2003). In this study we specically hypothesized that functional prefrontal asymmetry would predict startle measures following the offset of a negative stimulus, rather than startle magnitude during the stimulus itself. This hypothesis was formulated based upon our suggestion that the left lateral prefrontal cortex may play an important role in the inhibition of the amygdala and thus in the downregulation of negative affect. On this view, the recovery following an unpleasant picture as reected in the diminution of startle magnitude post-stimulus offset would reect the decrease in negative affect following the negative stimulus and may be modulated by PFC inhibition of the amygdala. Our results strongly conrmed our hypothesis. Frontal EEG asymmetry did not predict startle magnitude during the unpleasant picture; it only predicted startle magnitude following the offset of the negative picture. It may well be the case that frontal EEG asymmetry measures are associated with self-report measures of mood, dispositional affect and/or behavioral inhibition and activation. However, these associations are unlikely to be particularly revealing about mechanism since such associations can arise for a multitude of reasons. Recovering
224
slowly following the offset of an aversive stimulus may be associated with higher levels of dispositional negative affect and/or behavioral inhibition but such associations by themselves will provide little information to help mechanistically understand what role the PFC may be playing. Our study with startle measures during and following unpleasant stimuli is offered as an example of how we can begin to mechanistically dissect the stream of affective processing to help us better understand the role of the PFC in these processes. On the other hand, it is important that self-report measures be retained in our future studies because it is still of great interest to know how conscious mood is associated with prefrontal function and its correlates. Moreover, conscious mood may be a useful summary index that integrates and reects a multitude of processes that are unlikely to be fully captured in any single process measure, and conscious mood may play an important regulatory function in behavior. 2.4. Left prefrontal activity and angerwhat are the implications? Harmon-Jones (this issue) reviews a growing corpus of literature that suggests that specic forms of anger, or anger elicited in particular contexts, is associated with left-sided prefrontal activation. He argues that these data are consistent with the approachwithdrawal framework rather than with a valence-based model for frontal EEG asymmetry. More than 20 years ago, I argued the same point (Reuter-Lorenz and Davidson, 1981) though at that time had little data to buttress our suggestion. We did publish the rst data showing that anger elicited in infants when it was not accompanied by crying was associated with left-sided frontal EEG activation (Fox and Davidson, 1988). On the basis of the evidence from his lab, Harmon-Jones questions whether relatively greater left-sided frontal activity is a positive trait. There are several issues that are embedded here that should be disentangled. First is the distinction between baseline levels of asymmetry versus state-related changes. Most, though not all, of the evidence provided by Harmon-Jones and colleagues is based upon the latter type of measure, not the former. Secondly, it is important to ask just what the left PFC might be doing in anger-related affective processing. In light of the important data Harmon-Jones has gathered, it appears reasonable to suggest that one will nd activation of the left PFC in anger-related situations where there are response options that enable the individual to overcome whatever obstacle is thwarting the goal. As such, in these situations it may indeed be indeed be the case that relatively high left frontal brain activity is more psychologically and physically healthy than relatively less left frontal brain activity (Harmon-Jones, this issue), despite suggestions to the contrary, though the requisite evidence to carefully evaluate this claim has not been gathered. The negative behaviors that frequently accompany anger may well be subserved by other brain regions that require other methods to detect. And then there is a growing literature on the peripheral biological correlates of individual differences in baseline functional prefrontal asymmetry where we have shown that individuals with more left prefrontal activity have lower levels of the stress hormone cortisol (Kalin et al., 1998), lower levels of cerebrospinal uid measures of corticotropic releasing factor (CRF; Kalin et al., 2000), the central molecule that initiates the cascade of changes in the HPA system in response to stressful events, higher levels of natural killer activity both at baseline (Kang et al., 1991) and in response to challenge (Davidson et al., 1999), and higher antibody titers in response to inuenza vaccine (Rosenkrantz et al., 2003). Collectively, these ndings
225
powerfully reveal that individuals with higher levels of left-sided prefrontal activity do indeed have a more positive prole of peripheral biological indicators. When these ndings are considered in the context of other recent evidence from our lab showing that subjects with greater relative baseline left-sided prefrontal activity recover more quickly following a negative event (Jackson et al., 2003) and report higher levels of psychological well-being (Urry et al., in press), they indicate the left-sided prefrontal activity is associated with a nomological network that consists of effective coping with negative events and a resilient prole of peripheral neuroendocrine and immune function. If such individuals are also more prone to anger in certain situations, it may well be a form of anger that facilitates the rapid removal of obstacles that are thwarting goals. Of course, this perspective is still very much in need of additional conrmation and will be informed by studies that investigate both trait levels of prefrontal asymmetry as well as state variations in anger experience and expression and state variations in asymmetry. Moreover, the classication of a particular constellation of biological and/or psychological indicators as positive or benecial is fraught with complexity. It is not my intention here to unpack these complicated issues, but only to suggest that the compelling new data on anger offered by Harmon-Jones (this issue) is not at all incompatible with the view that higher levels of left prefrontal activity may be associated with aspects of resilience. 2.5. What do we know and what must we still learn about what different sectors of the prefrontal cortex are doing in emotion? Despite the current popularity of frontal EEG asymmetry measures, there is preciously little research that has systematically manipulated task parameters in a neurally-informed strategy to test specic hypotheses about the role of the left and right prefrontal cortex in different aspects of emotional processing. Harmon-Jones (this issue) and Nitschke et al. (this issue) are notable exceptions. As noted earlier, in light of the limitations of scalp-recorded brain electrical measures and what they reect, it is likely that the frontal EEG asymmetry measures mostly reect dorsolateral sectors of PFC. On the basis of a large body of both human and animal studies, we (Davidson et al., 2003) have proposed that greater left-sided dorsolateral activity may be associated with approach-related, goal-directed action planning. This would include instances of both pre-goal attainment positive affect as well as the approach component of anger when response options are available. This analysis implies that these effects will be particularly pronounced during periods of anticipation and planning. The temporal dynamics of shifting patterns of prefrontal asymmetry during pre versus post-goal periods is something that still must be systematically examined. We have been examining a number of tasks that explicitly manipulate reward and isolate periods in anticipation of reward (e.g., Sobotka et al., 1992; Skolnick and Davidson, 2002; Zinser et al., 1999; see also Miller and Tomarken, 2001). Though limited, the evidence accumulated to date indeed suggests that it is particularly during the period in anticipation of reward that the most pronounced left prefrontal activation is observed. There is less consensus on what the right lateral sector of PFC might be doing in withdrawal-related emotion. On the basis of neuroimaging studies of spatial working memory, we have suggested that activation of right lateral PFC during withdrawal-related emotion may be associated with threat-related vigilance. In studies with rhesus monkeys conducted
226
collaboratively with Kalin, we (Kalin et al., 2001) have reported that in response to a manipulation that elicits powerful increases in freezing behavior, there is also a parallel increase in behavioral signs of vigilance. Moreover, animals who display higher levels of these behaviors have greater right-sided baseline levels of lateral prefrontal activity (Kalin et al., 1998). This line of reasoning leads to a number of testable hypotheses concerning relations between right-sided PFC activity and the capacity to detect threat-related cues. Because of the strategic location of PFC, with extensive anatomical connections to posterior cortical regions as well as subcortical regions, this sector of cortex is ideally located to recruit other processing regions to facilitate adaptive responding. This view of PFC borrows heavily from that formulated in the cognitive neuroscience literature on the role of PFC in cognitive control (e.g., Miller and Cohen, 2001). What of other sectors of PFC that are not directly reected in frontal EEG? First, we know that there is extensive intrinsic connectivity among the various subregions of PFC (Rolls, 1999). This pattern of anatomical connectivity can therefore provide the basis for orbital, ventromedial and ventrolateral sectors of PFC to modulate processing in the dorsolateral sector. There are also direct roles for these other sectors of PFC in emotion and in fact, it is precisely these sectors that are not directly reected in scalp-recorded brain electrical activity that historically have been most closely associated with emotion. Based largely upon the work of Rolls in non-human primates and more recently in humans (Rolls, 1999; Rolls et al., 2003; see also work from our lab by Nitschke et al., 2004), the orbital sector of PFC can be thought of as cortex that computes affective value. It appears to track the hedonic response to a stimulus and not the stimulus properties itself. For example, OFC neurons will re in response to food in a food-deprived animal. However, once the animal is fed to satiety, a completely different response in OFC is found to the identical food stimulus (Scott et al., 1995). In humans, similar effects have been reported. Unfortunately there has been very little attention paid to possible asymmetries in ventral and orbital sectors of PFC though there have been some suggestions of asymmetries in these regions OFC consistent with the general pattern observed in dorsolateral sectors but this needs to carefully studied (e.g., Kawasaki et al., 2001). And there is increasingly evidence, reviewed earlier, both from neuroimaging and lesion studies, of asymmetries in the ventrolateral sector of the PFC. It appears that right-sided ventrolateral PFC is particularly sensitive to cues for punishment and when damaged, patients display a profound insensitivity to such cues (e.g., Tranel et al., 2002). In the cognitive domain, this sector of PFC appears to be sensitive to stop cues that require the inhibition of behavior (Aron et al., 2003; Garavan et al., 1999; Konishi et al., 1999). The ventromedial sector has been extensively studied by Damasio and colleagues (e.g., Damasio, 1994). They have suggested that ventromedial PFC provides a crucial substrate for affect-guided decision making. Patients with lesions in this area often have signicant abnormalities in decision making. Based upon recent evidence in rodents, this sector of PFC may also play a very important role in modulation of the amygdale and extinction learning. Lesions in this region result in failures to exhibit normal patterns of extinction following the learning of a classically conditioned aversive association (Morgan et al., 2003). Whether there are functional asymmetries in this PFC sector and whether they behave in ways that are similar to those in the dorsolateral sector is something that critically needs study in future research.
227
Finally, I wish to end this section by raising the issue of the distal causes of individual differences in EEG measures of functional prefrontal asymmetry. Tomarken et al. (this issue) report the striking observation of relations between socioeconomic status (SES) and prefrontal asymmetry in adolescents. The lower the SES, the greater the relative right-sided prefrontal activity. Although there are clearly genetic inuences on EEG measures of prefrontal asymmetry (Coan, 2003), we have suggested that environmental inuences, particularly early in development, are likely to be present and to shape aspects of prefrontal function, including baseline measures of functional prefrontal asymmetry (Davidson et al., 2000a). At this stage, although the evidence for experience-dependent changes in prefrontal asymmetry in humans is strictly correlational and does not imply cause, the extant animal data that explicitly manipulates early environment nds both structural and functional asymmetric changes in prefrontal cortex (e.g., Lyons et al., 2002), thus clearly implying that important environmental inuences on this circuitry are likely to exist in humans. This very brief discussion is meant to underscore the importance of functional heterogeneity within the PFC and to call attention to the limitations of focusing either explicitly or in many cases, unwittingly, on just one sector of this massive expanse of cortex.
3. Empirical and methodological issues in the use of frontal EEG asymmetry measures in studies of emotional processes 3.1. Evidence of progress The growing literature on methodological issues in the assessment of EEG asymmetry is leading to important advances that will improve the reliability and validity of these measures. The evidence presented in this Special Issue clearly establishes prefrontal EEG asymmetry measures as reecting, at least in part, trait-like variations in brain function that appear to predict interesting and important features of affective style. As such, these measures have much to add to our current armamentarium of methods used to probe links between the brain and affect. Functional MRI (fMRI) is being used increasingly to study the neural basis of affect (see Davidson and Irwin, 1999; Murphy et al., 2003; Wager et al., 2003 for recent reviews). One important limitation of fMRI methods is that they cannot be used to make inferences about baseline levels of activity though perfusion MR methods could potentially be used for this purpose (Wang et al., 2003). An important direction for future research is the use of EEG asymmetry measures to predict individual differences in fMRI studies of localized neuronal activation patterns in response to affective challenges (see Shackman et al., 2003, from our lab for recent example). Another critical opportunity for future research is the use of source localization methods to more precisely describe the intracerebral sources of the scalp-recorded signals that we measure. Several recent studies from our lab illustrate the potential utility of these measures and underscore the differences in information yield when such source localization methods are utilized (e.g., Pizzagalli et al., 2002, 2003), although these methods have yet to be compared directly with functional neuroimaging data as I have suggested above. Also not yet evaluated is the differential reliability and validity of asymmetry metrics derived from direct surface recordings versus from source-localized intracerebral signals. And nally perfusion MRI
228
can be used to obtain baseline data on localized cortical perfusion, and asymmetry can be computed from these measures and compared to EEG measures obtained from the same subjects. 3.2. The problem of the reference electrode As several authors in this Special Issue have illustrated, the different reference electrode locations used in research on frontal EEG asymmetry continues to be a vexing problem. While some authors have demonstrated robust and consistent effects across different reference electrodes, this has not always been the case and it underscores the fact that EEG recordings reect the potential difference between two locations and there is no electrically neutral location anywhere on the body. There still remains much methodological work to accomplish on this issue including the simultaneous assessment of brain electrical and hemodynamic or metabolic measures (see e.g., Goldman et al., 2002; Laufs et al., 2003). By using regional hemodynamic or metabolic measures as a gold standard against which to compare various EEG metrics, it may be possible to identify a set of procedures that consistently yields the most robust relations between brain electrical signals and more spatially precise measures of brain function. We have adopted this strategy and have begun to empirically dissect this complex issue and determine which EEG reference schemes and frequency bands most directly correlate with direct measures of regional glucose metabolism assessed with positron emission tomography (Oakes et al., 2004). One issue of considerable importance to research on EEG measures of functional brain asymmetry is the use of the linked ears reference, a common choice in studies of frontal EEG asymmetry (see e.g., Minnix et al., this issue). As Allen et al. (this issue) correctly show, variations in electrode impedance cannot account for any detectable difference in signals across electrodes using modern high impedance ampliers. This is true in all but one circumstance and that is when electrodes are linked prior to inputting the signal into the amplier (Senulis and Davidson, 1989). Thus when a linked ears or mastoid montage is used, even slight variations in electrode impedance on each side can asymmetrically bias the signal. It is for this reason that we have suggested that a computer-averaged ears (or mastoids) reference be used if an investigator wishes to compute a linked ears or linked mastoid reference (see Davidson et al., 2000b). 3.3. Frequency bands other than alpha It is striking that none of the articles in this Special Issue that focused on adult subjects included an analysis of frequency bands other than alpha. In several recent studies we have examined frequency bands other than alpha including theta, beta and gamma (Pizzagalli et al., 2001, 2002, 2003). In these studies, we also examined alpha power and in a number of cases, asymmetrical effects were found in bands other than alpha while effects in the alpha band were absent. Moreover, in our recent direct simultaneous comparison of EEG and PET measures of regional glucose metabolism (Oakes et al., 2004), we found that although alpha shows the predicted inverse relation with metabolism (see also Cook et al., 1998), the frequency band most consistently and strongly associated with glucose metabolism was gamma. Thus, although the frontal EEG asymmetry literature has traditionally focused on
229
alpha power, it is important as we go forward to carefully examine other frequency bands as these may provide additional information not reected in alpha power. 3.4. Bilateral variations in PFC are also important Extant methods of analysis featured in the articles in this Special Issue emphasize, for the most part, asymmetrical variations in frontal brain activity. The computation of an asymmetry metric, while serving an important function, also masks the opportunity to examine bilateral variations in frontal brain activity. The sole exception in this group of articles is the Nitschke et al. (this issue) report; they illustrate the predictive value of bilateral variations in alpha power. It is important to underscore the fact that even if condition or group differences are found in asymmetry, such differences could be superimposed upon bilateral variations in activity and these bilateral variations in activity may be functionally signicant (see e.g., Schmidt and Trainor, 2001). Future research in this area should include analysis methods that permit the systematic examination of bilateral variations in activation and assess the functional signicance of such bilateral variations. 3.5. Importance of comparative research Functional asymmetries are not the sole province of humans. Such asymmetries are found at many different levels of phylogeny (see Hugdahl and Davidson, 2003 for several reviews). And given the pervasive presence of affect at virtually all levels of phylogeny where behavior itself is observed, functional asymmetries associated with affective processes may be considerably more ubiquitous across the animal kingdom than asymmetries associated with language and communication. For example, there is a growing literature in rodents that is consistent with human data indicating a preferential role for regions of the right PFC in anxiety and withdrawal (see Berridge et al., 2003). In rhesus monkeys, we have recorded brain electrical activity from the scalp surface using procedures similar to those we use in humans and have found differences among animals in baseline levels of asymmetric prefrontal activity that are reliable over time and show systematic correlations with behavioral and biological measures that reect anxiety and stressful responding (e.g., Kalin et al., 1998, 2000). Such animal models provide a foundation for research where lesions in particular regions of PFC can be made to establish the differential causal role of left and right prefrontal regions in specic types of affective behavior.
4. Summary and conclusions This commentary provides some reections on the current state of affairs in research on EEG frontal asymmetries associated with emotion. The rst part addresses a series of conceptual issues and the second part some empirical and methodological issues. Of prime important for the future of EEG asymmetry research is that it make more meaningful contact with the growing corpus of literature on primate prefrontal function. One of the most lamentable characteristics of the current work on frontal EEG asymmetry is that it has evolved largely unconnected with the neuroscience literature on the structure and function
230
of the PFC. Scalp-recorded measures of prefrontal activity likely reect primarily dorsolateral PFC function, yet the PFC is anatomically and functionally heterogeneous. Careful consideration of this issue and developing hypotheses based upon ndings from the neuroscience literature should aid future progress. Another critical issue is the fact that the PFC is embedded in a complex circuit involving other cortical and subcortical structures and these other structures likely play a very important role in many of the affective processes featured in the articles in this Special Section. However, because the EEG measures that are utilized in these studies do not directly reect activity in other parts of the circuitry, these other neural components have, for the most part, been ignored by those investigators who study frontal EEG. Much of the research on frontal EEG asymmetry has examined the correlates of variations in asymmetry with self-report measures. While these reported associations have been interesting, they are typically not informative with regard to mechanism because the specic types of processes affected by prefrontal function are likely to themselves be opaque to self-report. Thus, while they inuence self-report (such as variations in the time to recovery following a negative event), they are not themselves consciously accessible and thus such self-report measures will ultimately be uninformative if we hope to build neurally-inspired theory. The approach-withdrawal model of frontal asymmetry (see e.g., Davidson, 1992) receives considerable support from numerous studies in this Special Issue, most especially from Harmon-Jones, with his ndings on anger. The provocative question of whether or not high levels of baseline left prefrontal activity, in light of the Harmon-Jones ndings, still should be considered to be a positive trait, is addressed, with a tentative afrmation, despite the ndings on anger. Finally, some discussion of what different sectors of the PFC might be doing in emotion is introduced. In the second section of this essay, several methodological issues are considered including problems with the linked ears or mastoids reference, the importance of examining frequency bands other than alpha, using analytic methods that permit bilateral variations in frontal activity to be assessed, and some comments on the importance of comparative research to address some of the fundamental outstanding questions in this line of research. The fact of this Special Issue is a testament to the enormous progress that has occurred in this area of research. It will be exciting to revisit this topic in 5 years since with the advent of new methods such as functional neuroimaging, particularly when they are combined with traditional psychophysiological measures (Davidson, 2003b), very rapid and novel progress can be expected.
Acknowledgements I wish to thank the many students, trainees and staff members in my laboratory who have enabled this research program and have been instrumental in it thriving. I particularly wish to thank Jack Nitschke, Alex Shackman, Carien van Reekum, Jim Coan, Tim Salamons and Tom Johnstone for their comments on an earlier draft of this manuscript. I also wish to thank John Allen for his comments on an earlier draft. I also wish to thank the consistent support of NIMH who has funded most of the research described in this commentary and is currently providing support through the following grants to RJD: NIMH grants MH43454, MH40747, P50-MH52354, P50-MH61083 and NIA grant PO1-AG021079.
231
References
Allen, J.J.B., Coan, J.A., Nazarian, M., this issue. Issues and assumptions on the road from raw signals to metrics of frontal EEG asymmetry in emotion. Biological Psychology. Aron, A.R., Fletcher, P.C., Bullmore, E.T., Sahakian, B.J., Robbins, T.W., 2003. Stop-signal inhibition disrupted by damage to right inferior frontal gyrus in humans. Nature Neuroscience 6, 115116. Berridge, C.W., Espana, R.A., Stalnaker, T.A., 2003. Stress and coping: asymmetry of dopamine efferents within prefrontal cortex. In: Hugdahl, K., Davidson, R.J. (Eds.), The Asymmetrical Brain. MIT Press, Cambridge, MA, pp. 69103. Cacioppo, J.T., this issue. Feelings and emotions: roles for electrophysiological markers. Biological Psychology. Clark, L., Manes, F., Antoun, N., Sahakian, B.J., Robbins, T.W., 2003. The contributions of lesion laterality and lesion volume to decision-making impairment following frontal lobe damage. Neuropsychologia 41, 1474 1483. Coan, J.A., 2003. The heritability of trait frontal EEG asymmetry and negative emotionality: sex differences and genetic nonadditivity (Doctoral dissertation, University of Arizona). Dissertation Abstracts International 64, 2382. Cook, I.A., OHara, R., Uijtdehaage, S.H., Mandelkern, M., Leuchter, A.F., 1998. Assessing the accuracy of topographic EEG mapping for determining local brain function, Electroencephalography. Electroencephalography and Clinical Neurophysiology 107 (6), 408414. Damasio, A.R., 1994. Descartes Error: Emotion, Reason, and the Human Brain. Avon Books, New York. Davidson, R.J., 1992. Emotion and affective style: hemispheric substrates. Psychological Science 3, 3943. Davidson, R.J., 2000. Affective style, psychopathology, and resilience: brain mechanisms and plasticity. American Psychologist 55, 11961214. Davidson, R.J., 2002. Anxiety and affective style: role of prefrontal cortex and amygdala. Biological Psychiatry 51, 6880. Davidson, R.J., 2003a. Seven sins in the study of emotion: correctives from affective neuroscience. Brain and Cognition 52, 129132. Davidson, R.J., 2003b. Affective neuroscience and psychophysiology: toward a synthesis. Psychophysiology 40, 655665. Davidson, R.J., Coe, C.C., Dolski, I., Donzella, B., 1999. Individual differences in prefrontal activation asymmetry predict natural killer cell activity at rest and in response to challenge. Brain, Behavior, and Immunity 13, 93 108. Davidson, R.J., Irwin, W., 1999. The functional neuroanatomy of emotion and affective style. Trends in Cognitive Science 3, 1121. Davidson, R.J., Jackson, D.C., Kalin, N.H., 2000a. Emotion, plasticity, context and regulation: perspectives from affective neuroscience. Psychological Bulletin 126, 890906. Davidson, R.J., Jackson, D., Larson, C., 2000b. Human electroencephalography. In: Cacioppo, J.T., Bernston, G., Tassinary, L. (Eds.), Principles of Psychophysiology, second ed. Cambridge University Press, New York, pp. 2752. Davidson, R.J., Pizzagalli, D., Nitschke, J.B., Kalin, N.H., 2003. Parsing the subcomponents of emotion and disorders of emotion: perspectives from affective neuroscience. In: Davidson, R.J., Goldsmith, H.H., Scherer, K. (Eds.), Handbook of Affective Sciences. Oxford University Press, New York, pp. 824. Davidson, R.J., Putnam, K.M., Larson, C.L., 2000c. Dysfunction in the neural circuitry of emotion regulationa possible prelude to violence. Science 289, 591594. Davidson, R.J., Schwartz, G.E., Saron, C., Bennett, J., Goleman, D.J., 1979. Frontal versus parietal EEG asymmetry during positive and negative affect. Psychophysiology 16, 202203. Fox, N.A., Davidson, R.J., 1988. Patterns of brain electrical activity during facial signs of emotion in ten month old infants. Developmental Psychology 24, 230236. Gainotti, G., 1972. Emotional behavior and hemispheric side of lesion. Cortex 8, 4155. Garavan, H., Ross, T.J., Stein, E.A., 1999. Right hemispheric dominance of inhibitory control: an event-related functional MRI study. Proceedings of the National Academy of Sciences of the United States of America 96, 83018306.
232
Goldman, R.I., Stern, J.M., Engel Jr., J., Cohen, M.S., 2002. Simultaneous EEG and fMRI of the alpha rhythm. Neuroreport 13, 24872492. Harmon-Jones, E., this issue. Contributions from research on anger and cognitive dissonance to understanding the motivational functions of asymmetrical frontal brain activity. Biological Psychology. Hornak, J., Bramham, J., Rolls, E.T., Morris, R.G., ODoherty, J., Bullock, P.R., Polkey, C.E., 2003. Changes in emotion after circumscribed surgical lesions or the orbitofrontal and cingulate cortices. Brain 126, 16911712. Hugdahl, K., Davidson, R.J. (Eds.), 2003. The Asymmetrical Brain. MIT Press, Cambridge, MA. Jackson, D.C., Malmstadt, J.R., Larson, C.L., Davidson, R.J., 2000. Suppression and enhancement of responses to emotional pictures. Psychophysiology 37, 515522. Jackson, D.C., Mueller, C.J., Dolski, I.V., Dalton, K.M., Nitschke, J.B., Urry, H.L., et al., 2003. Now you feel it, now you dont: frontal brain electrical asymmetry and individual differences in emotion regulation. Psychological Science 14, 612617. Kalin, N.H., Larson, C., Shelton, S.E., Davidson, R.J., 1998. Asymmetric frontal brain activity, cortisol, and behavior associated with fearful temperament in rhesus monkeys. Behavioral Neuroscience 112, 286292. Kalin, N.H., Shelton, S.E., Davidson, R.J., 2000. Cerebrospinal uid corticotropin-releasing hormone levels are elevated in monkeys with patterns of brain activity associated with fearful termerament. Biological Psychiatry 47, 579585. Kalin, N.H., Shelton, S.E., Davidson, R.J., Kelley, A.E., 2001. The primate amygdala mediates acute fear but not the behavioral and physiological components of anxious temperament. Journal of Neuroscience 21, 20672074. Kang, D.H., Davidson, R.J., Coe, C.L., Wheeler, R.W., Tomarken, A.J., Ershler, W.B., 1991. Frontal brain asymmetry and immune function. Behavioral Neuroscience 105, 860869. Kawasaki, H., Kaufman, O., Damasio, H., Damasio, A.R., Granner, M., Bakken, H., 2001. Single-neuron responses to emotional visual stimuli recorded in human ventral prefrontal cortex. Nature Neuroscience 4, 1516. Konishi, S., Nakajima, K., Uchida, I., Kikyo, H., Kameyama, M., Miyashita, Y., 1999. Common inhibitory mechanism in human inferior prefrontal cortex revealed by event-related functional MRI. Brain 122, 981991. Laufs, H., Kleinschmidt, A., Beyerle, A., Eger, E., Salek-Haddadi, A., Preibisch, C., Krakow, K., 2003. EEG-correlated fMRI of human alpha activity. Neuroimage 19, 14631476. LeDoux, J.E., 1996. The Emotional Brain. Simon and Schuster, New York. Lyons, D.M., Afarian, H., Schatzberg, A.F., Sawyer-Glover, A., Mosley, M.E., 2002. Experience-dependent asymmetric variation in primate prefrontal morphology. Behavioral and Brain Research 136, 5159. Miller, A., Tomarken, A.J., 2001. Task-dependent changes in frontal brain asymmetry: Effects of incentive cues, outcome expectancies and motor responses. Psychophysiology 38, 500511. Miller, E.K., Cohen, J.D., 2001. An integrative theory of prefrontal cortex function. Annual Review of Neuroscience 24, 167202. Minnix, J.A., Kline, J.P., Blackhart, G.C., Pettit, J.W., Perez, M., Joiner, T.E., this issue. Relative left frontal activity is associated with increased depression in high reassurance-seekers. Biological Psychology. Morgan, M.A., Schulkin, J., LeDoux, J.E., 2003. Ventral medial prefrontal cortex and emotional perseveration: the memory for prior extinction training. Behavioral and Brain Research 146, 121130. Murphy, F.C., Nimmo-Smith, I., Lawerence, A.D., 2003. Functional neuroanatomy of emotions. A meta-analysis. Cognitive, Affective and Behavioral Neuroscience 3, 207233. Myers, R.E., 1972. Role of prefrontal and anterior temporal cortex in social behavior and affect in monkeys. Acta Neurobiologiae Experimentalis (Warsz) 32, 567579. Nauta, W.H., 1971. The problem of the frontal lobe: a reinterpretation. Journal of Psychiatric Research 8, 167187. Nitschke, J.B., Heller, W., Etienne, M.A., Miller, G.A., this issue. Prefrontal cortex activity differentiates processes affecting memory in depression. Biological Psychology. Nitschke, J.B., Nelson, E.E., Rusch, B.D., Fox, A.S., Oakes, T.R., Davidson, R.J., 2004. Orbitofrontal cortex tracks positive mood in mothers viewing pictures of their newborn infants. NeuroImage 21, 583592. Oakes, T.R., Pizzagalli, D.A., Hendrick, A.M., Horras, K.A., Larson, C.L., Abercrombie, H.C., Schaefer, S.M., Koger, J.V., Davidson, R.J., 2004. Functional coupling of simultaneous electrical and metabolic activity in the human brain. Human Brain Mapping 21, 257270. Ochsner, K.N., Bunge, S.A., Gross, J.J., Gabrieli, J.D., 2002. Rethinking feelings: an fMRI study of the cognitive regulation of emotion. Journal of Cognitive Neuroscience 14, 12151229. Panksepp, J., 2003. At the interface of the affective, behavioral and cognitive neurosciences: decoding the emotional feelings of the brain. Brain and Cognition 52, 414.
233
Pizzagalli, D.A., Greischar, L.L., Davidson, R.J., 2003. Spatio-temporal dynamics of brain mechanisms in aversive classical conditioning: high-density event-related potential and brain electrical tomography analyses. Neuropsychologia 41, 184194. Pizzagalli, D.A., Nitschke, J.B., Oakes, T.R., Hendrick, A.M., Horras, K.A., Larson, C.L., Abercrombie, H.C., Schaefer, S.M., Koger, J.V., Benca, R.M., Pascual-Marqui, R.D., Davidson, R.J., 2002. Brain electrical tomography in depression: the importance of symptom severity, anxiety, and melancholic features. Biological Psychiatry 52, 7385. Pizzagalli, D., Pascual Marqui, R.D., Nitschke, J.B., Oakes, T.R., Larson, C.L., Abercrombie, H.C., Schaefer, S.M., Koger, J., Benca, R.M., Davidson, R.J., 2001. Anterior cingulate activity predicts degree of treatment response in major depression: Evidence from brain electrical tomography analysis. The American Journal of Psychiatry 158, 405415. Quirk, G.J., Likhtik, E., Pelletier, J.G., Pare, D., 2003. Stimulation of medial prefrontal cortex decreases the responsiveness of central amygdala output neurons. Journal of Neuroscience 23, 88008807. Reuter-Lorenz, P., Davidson, R.J., 1981. Differential contributions of the two cerebral hemispheres to the perception of happy and sad faces. Neuropsychologia 19, 609613. Rolls, E.T., 1999. The Brain and Emotion. Oxford University Press, New York. Rolls, E.T., Kringelbach, M.L., de Araujo, I.E., 2003. Different representations of pleasant and unpleasant odours in the human brain. European Journal of Neuroscience 18, 695703. Rosenkrantz, M.A., Jackson, D.C., Dalton, K.M., Dolski, I., Ryff, C.D., Singer, B.H., Muller, D., Kalin, N.H., Davidson, R.J., 2003. Affective style and in vivo immune response: neurobehavioral mechanisms. Proceedings of the National Academy of Sciences of the United States of America 100, 1114811152. Schmidt, L.A., Trainor, L.J., 2001. Frontal brain electrical activity (EEG) distinguishes valence and intensity of musical emotions. Cognition and Emotion 15, 487500. Scott, T.R., Yan, J., Rolls, E.T., 1995. Brain mechanisms of satiety and taste in macaques. Neurobiology 3, 281292. Senulis, J.A., Davidson, R.J., 1989. The effect of linking the ears on the hemispheric asymmetry of the EEG. Psychophysiology 26, S55. Shackman, A.J., Maxwell, J.S., Skolnick, A.J., Schaefer, H.S., Davidson, R.J., 2003. Exploiting individual differences in the prefrontal asymmetry of approach-related affect: hemodynamic, electroencephalographic, and psychophysiological evidence. Program No. 444.6. 2003 Abstract Viewer/Itinerary Planner. Society for Neuroscience, online, Washington, DC. Skolnick, A.J., Davidson, R.J., 2002. Affective modulation of eyeblink startle with reward and threat. Psychophysiology 39, 835850. Sobotka, S.S., Davidson, R.J., Senulis, J.A., 1992. Anterior brain electrical asymmetries in response to reward and punishment. Electroencephalography and Clinical Neurophysiology 83, 236247. Sutton, S.K., Davidson, R.J., 1997. Prefrontal brain asymmetry: a biological substrate of the behavioral approach and inhibition systems. Psychological Science 8, 204210. Tomarken, A.J., Davidson, R.J., Wheeler, R.E., Doss, R.C., 1992. Individual differences in anterior brain asymmetry and fundamental dimensions of emotion. Journal of Personality and Social Psychology 62, 676687. Tomarken, A.J., Dichter, G.S., Garber, J., Simien, C., this issue. Resting frontal brain activity: linkages to maternal depression and socioeconomic status among adolescents. Biological Psychology. Tranel, D., Bechara, A., Denburg, N.L., 2002. Asymmetric functional roles of right and left ventromedial prefrontal cortices in social conduct, decision-making and emotional processing. Cortex 38, 589612. Urry, H.L., Nitschke, J.B., Dolski, I., Jackson, D.C., Dalton, K.M., Mueller, C.J., Rosenkranz, M.A., Ryff, C.D., Singer, B.H., Davidson, R.J., in press. Making a life worth living: neural correlates of well-being. Psychological Science. Wager, T.D., Phan, K.L., Liberzon, I., Taylor, S.F., 2003. Valence, gender, and lateralization of functional brain anatomy in emotion: a meta-analysis of ndings from neuroimaging. Neuroimage 19, 513531. Wallis, J.D., Miller, E.K., 2003. Neuronal activity in primate dorsolateral and orbital prefrontal cortex during performance of a reward preference task. European Journal of Neuroscience 18, 20692081. Wang, J., Aguirre, G.K., Kimberg, D.Y., Roc, A.C., Li, L., Detre, J.A., 2003. Arterial spin labeling perfusion fMRI with very low task frequency. Magnetic Resonance in Medicine 49, 796802. Zinser, M.C., Fiore, M., Davidson, R.J., Baker, T., 1999. Manipulating smoking motivation: impact on an electrophysiological index of approach motivation. Journal of Abnormal Psychology 108, 240254.
Commentary
Feelings and emotions: roles for electrophysiological markers

John T. Cacioppo
Department of Psychology, University of Chicago, 5848 S. University Avenue, Chicago, IL 60637, USA
Abstract Asymmetrical electroencephalogram (EEG) alpha activity over anterior regions of the scalp predicts a variety of outcome measures of interest to emotion researchers. This vast and diverse literature is examined from three different viewpoints. First, the organization of this vast literature is contrasted from theoretical and statistical perspectives, and the advantages and disadvantages of each perspective are considered. Second, the correlates of EEG asymmetry are sometimes treated as criterion (dependent) measures and at other times treated as predictor (independent) measures. Differences in the interpretation of each are surveyed, and the need for attention to whether EEG asymmetry is a simple correlate, mediator, or moderator of the associated affective measures is noted. Finally, the studies of EEG asymmetry and emotion that adopt a psychological perspective are contrasted with those that adopt a neurophysiological perspective, and the import of each for theory, experimental design, and analytic strategy is discussed. 2004 Elsevier B.V. All rights reserved.
Keywords: Electroencephalogram (EEG); Criterion; Predictor
1. Introduction For centuries scholars regarded feelings (e.g. sadness, fear, anger, disgust, dysphoria) as the most (if not only) appropriate data for the study of emotion. The reasoning for this position, on the surface, is straightforward. When one nds oneself teetering dangerously on the edge of a cliff, positioned in the path of a charging beast, or trapped on the top oors of a burning skyscraper, the consequences are not subtle: the unmistakable feeling of fear commands center stage, the expression of terror is etched involuntarily across ones face, attention is narrowed to a focal set of stimuli, actions are driven by attempts to avoid
E-mail address: cacioppo@uchicago.edu (J.T. Cacioppo). 0301-0511/$ see front matter 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.biopsycho.2004.03.009
236
J.T. Cacioppo / Biological Psychology 67 (2004) 235243
the imminent peril, and metabolic resources are mobilized to support these actions. When the same set of responses is elicited by normal circumstances, such as when an individual approaches a podium to deliver a speech, encounters a nonpoisonous snake, or prepares to board an airplane, normal life is disrupted and individuals complain about the horric feelings that hijack their will and their way. Feelings represent a special class of outcomes that are important to explain in theories of emotion, to be sure, but it does not follow from their position in center stage that the most appropriate data for constructing or testing theories of emotion are limited to feelings or self-reports (cf. Cacioppo et al., 1997, 1999). An important contribution of both the cognitive revolution in psychology and animal research on emotion is an appreciation that feelings provide an incomplete glimpse of their underlying information processing structures and operations. LeDoux (2000), in his recent survey of the eld in the Annual Review of Neuroscience, echoed this point: It is widely recognized that most cognitive processes occur unconsciously, with only the end products reaching awareness, and then only sometimes. Emotion researchers, though, did not make this conceptual leap. They remained focused on subjective emotional experience. . . . The main lesson to be learned. . . is that emotion researchers need to gure out how to escape from the shackles of subjectivity if emotion research is to thrive (LeDoux, 2000, p. 156). How are we to plumb noninvasively the structures and processes of emotion in humans if not through self-reports? The work summarized in this special issue suggests that measures of frontal electroencephalogram (EEG) asymmetry have much to offer in this regard. My aim in this commentary is to highlight a few themes in this literature that I hope will foster this effort. 1.1. Theoretical and statistical perspectives Research on EEG frontal asymmetry and emotion is one of the most promising and fertile in the eld. The assortment of variables associated with EEG frontal asymmetry, for instance, ranges from child temperament (see Fox et al., 2001), the maternal depression of infants (Jones et al., 2004) and adolescents (Tomarken et al., 2004), self-report measures of affect and personality (e.g. Tomarken and Davidson, 1994), shyness and social anxiety (Schmidt, 1999), socioeconomic status (Tomarken et al., 2004), basal cortisol levels (Kalin et al., 1998), and immune function (e.g. Kang et al., 1991) to cognitive dissonance arousal (Harmon-Jones, 2004), the emotional content of dreams upon awakening (Donzella et al., 1994), the memory recognition of sad narrative material (Nitschke et al., 2004), and reassurance seeking (Minnix et al., 2004). The number of associations between EEG frontal asymmetry and various aspects of emotion is sufciently large, and the measures that are associated with EEG frontal asymmetry are sufciently diverse, that it can be difcult to see beyond these data to underlying organizations, structures, and processes. One way of dealing with such a complex array of associations is to rely on a theoretical formulation to guide ones analysis and interpretation. The most inuential is Davidsons (e.g. 1993, 2004) model, which species that high levels of relative left frontal activity are associated with the expression and experience of positive,
237
approach-related emotions, and high levels of relative right frontal activity are associated with the experience and expression of negative, withdrawal-related emotions. Harmon-Jones (2004) has argued that this model includes both motivational and valence components, which can be distinguished conceptually and empirically. In doing so, he has identied a valence model of EEG asymmetry in which high levels of relative left frontal activity are associated with the expression and experience of positive emotions and high levels of relative right frontal activity are associated with the experience and expression of negative emotions; and a motivational direction model in which high levels of relative left frontal activity are associated with the expression of approach-related emotions and high levels of relative right frontal activity are associated with the expression of withdrawal-related emotions. Although positive emotions are typically associated with motivations to approach and negative emotions are typically associated with motivations to withdraw, there are notable exceptions (e.g. anger-out). Through the development and tests of competing hypotheses, Harmon-Jones (2004) and colleagues have pursued the goal of specifying more precisely what the emotional and motivational functions of asymmetrical frontal brain activity might be. In the pursuit of the simplest single theoretical construct that might account for these assorted ndings, it may be worthwhile to ask whether the various correlates of EEG asymmetry are themselves sufciently correlated that they likely result from a unitary underlying faculty or mechanism. Indeed, philosophers of science have long noted the benecial effect of theory testing on scientic progress (see Brazier, 1959), but in the past century there has been an increasing recognition of the ways in which a theoretical focus can also hinder scientic progress (e.g. by biasing the selection of measurement instruments, Eddington, 1939, or the construction or interpretation of crucial tests, Greenwald et al., 1986). A multivariate statistical approach to investigate the structure of the correlates of EEG asymmetry may also be informative by pointing to possible oversimplications in existing theoretical formulations and, relatedly, by suggesting new theoretical hypotheses that can be tested empirically. Consider a thought experiment in which all of the known correlates of EEG asymmetry (e.g. see Tables 14 in Coan and Allen, this issue) were measured in a single study with a large sample size. Would one expect EEG asymmetry to predict each of these outcome measures? The answer is yes, of course, assuming measurement error was negligible and the ndings reliable. So far, this thought experiment simply constitutes a replication of prior research. However, because all of the outcome measures in this thought experiment are obtained from the same individuals, it is possible to use multivariate statistical techniques to explore the interrelationships among and the statistical coherences among these measures. It would be possible, for instance, to submit the data from half of the participants to an exploratory factor analysis using an oblique rotation to examine the statistical structure of the multivariate dataset. The data from the other half of the participants could then be subjected to a conrmatory factor analysis to validate this structure. If the results conrmed these measures were sufciently coherent and they represented a single dimension, then not only would this provide additional evidence for their being a unitary faculty or single mechanism underlying their production, but also the factor loadings of the measures should bear on the core features of this unidimensional construct. If, on the other hand, the results of this thought experiment indicated a single dimension was insufcient to represent the correlates of EEG asymmetry, the nature of the multidimensional representation and the factor loadings on these dimensions could have considerable
238
theoretical import, as they would lead to new and testable hypotheses about what precisely EEG asymmetry is predicting in specic contexts. The point of this thought experiment is not to prejudge what the outcome might be, but rather to underscore the complementary role of theoretical and statistical perspectives on a literature as complex and diverse as that on the associations of anterior EEG asymmetry. The concept of arousalonce a popular construct in the eld of emotion, was undermined not so much by its lack of empirical support as by its oversimplication of the accumulated empirical data. Arousal came to stand for such a diverse set of outcomes, many of which were only so poorly correlated, that the formulation lost its predictive power and, thus, its scientic import. The same fate need not befall an area of work as fertile as EEG asymmetry and emotion if both statistical and theoretical approaches are used to complement each other. 1.2. Associations between EEG frontal asymmetry and emotion Coan and Allen (2004) classied studies of EEG asymmetry and emotion as falling under one of four categories: (a) research on trait frontal EEG asymmetry and other trait-like measures, (b) studies of trait frontal asymmetry as a predictor of state dependent changes, (c) studies of trait frontal EEG asymmetry and measures of psychopathology, and (d) research on frontal EEG activation asymmetry as a state measure. This categorization leads to several quick observations. First, EEG asymmetry is not itself a mechanism but rather it is a marker of some underlying neural processes. Although the rst three categories of research concern trait frontal EEG asymmetry as it relates to traits, state changes, and psychopathology, there appears to be a paucity of research on the neural structures and processes that are responsible for these associations (cf. Davidson, 2003). I return to this issue in Section 1.3. Second, it is judicious to ask for each new association whether EEG asymmetry, or more specically the neural mechanism responsible for frontal EEG asymmetry, plays a causal role in the outcome measure. Although the studies tend to be treated similarly, some studies of frontal EEG asymmetry are designed to examine behavioral correlates, other studies treat frontal EEG asymmetry as an independent (blocking) variable and behavioral variables as the dependent measure, and still others treat one or more behavioral measures as the independent (blocking) variable and frontal EEG asymmetry as the dependent measure. These distinctions are signicant. Studies of frontal EEG asymmetry designed to examine behavioral correlates speak only the probability of the co-occurrence of these measures (i.e. P(EEG asymmetry, behavioral measure)), studies that treat frontal EEG asymmetry as an independent (blocking) variable and behavioral variables as the dependent measure speak the probability of the behavioral variable given frontal EEG asymmetry (i.e. P(behavioral measure/EEG asymmetry)), and studies that treat one or more behavioral measures as the independent (blocking) variable and frontal EEG asymmetry as the dependent measure speak the probability of frontal EEG asymmetry given the behavioral variable differs (i.e. P(EEG asymmetry/behavioral measure)). We have dealt elsewhere with the different implications for the interpretation of empirical ndings for each of these classes of studies (e.g. Cacioppo and Tassinary, 1990; Sarter et al., 1996). Sufce it to say that statements about the causal role of frontal EEG asymmetry in producing a particular emotional state or predisposition are inappropriate in studies that treat frontal EEG asymmetry as the dependent measure and a behavioral measure (e.g. depression) as the independent (blocking) variable.
239
Studies that treat frontal EEG asymmetry as a blocking variable and behavioral variables as the dependent measure are better suited to such interpretations, but alternative interpretations still need to be considered. Allen et al. (2001), for instance, recognized that most of the extant evidence about the possible role of frontal EEG asymmetry and emotional responses is based on correlational evidence. To examine the potential causal role of frontal EEG asymmetry in the production of emotional responses, they used biofeedback training to increase either relative right or relative left frontal EEG activity. Manipulation checks conrmed that 5 consecutive days of biofeedback training produced the expected differences in EEG asymmetry, and self-reported responses to emotionally evocative lm clips indicated that individuals trained to increase left frontal activity reported more positive affect to a happy lm clip than individuals trained to increase right frontal activity. This is a laudable effort on a number of fronts, including the attempt to test the causal role played by the neural processes marked by frontal EEG asymmetry, the extensive biofeedback training provided to participants over 5 consecutive days in an effort to manipulate frontal activity, and the treatment of EEG asymmetry as a blocking variable and emotional responses as the dependent variable. It is possible that the biofeedback training operated directly to increase right or left frontal activity which then in turn led to more positive or negative affective responses to happy lm (see Harmon-Jones, 2004), but the opposite is also possible. For instance, it is conceivable that participants tried different mental strategies to achieve reinforcement in the biofeedback training task. If participants discovered during biofeedback training that thinking happier thoughts led to reinforcement for producing relative right frontal activity and that thinking less happy thoughts led to the opposite, then the frontal EEG asymmetry may not have been the cause of the affective responses they observed. Techniques such as transcranial magnetic stimulation, which would avoid this alternative interpretation, may prove a valuable additional tool for investigating questions of this form (e.g. see Nahas et al., 2003). As Coan and Allen (2004) suggest, experimental designs and statistical tests to determine whether EEG asymmetry may be serving as a moderator or a mediator of emotional processes can be especially helpful at this point in time to advance theory and research in the area. The frontal EEG asymmetry marks some underlying neural process or processes. Continued efforts are needed to identify the neural processes that underlie the correlation between resting EEG alpha asymmetry and emotional style and the neural processes that underlie differential EEG alpha activation in emotionally evocative situations. Nevertheless, EEG asymmetry as a marker of these underlying events can be examined using the procedures outlined by Coan and Allen (2004) to determine whether the processes it marks are likely to be serving as a mediator, moderator, or simple correlate in a given study. The frontal EEG asymmetry is conceptualized as a mediator when the neural processes underlying the differential frontal activation is thought to be instrumental in the production of tonic affective states or state changes (e.g. more or less pleasant or unpleasant feelings about a stimulus) or approach- or withdrawal-motivational tendencies, whereas frontal EEG asymmetry is conceptualized as a moderator when the neural processes underlying it are thought to dampen or augment the processes instrumental in the production of tonic affective states or state changes. Allen and Coans (2004) description of the methodological and statistical conditions needed to differentiate these possible roles is a valuable contribution to the eld as it may explain some apparent inconsistencies in results (e.g. see Reid et al., 1998).
240
1.3. Psychological and physiological perspectives Whether frontal EEG asymmetry is treated as an individual difference variable or a state-dependent concomitant, its association to emotion and emotion-related constructs has generally reected a search for functional relationships to other behavioral constructs rather than a search for underlying neural mechanisms. The studies comprising this special issue are a case in point, and they nicely illustrate the theoretical value of this psychological approach to the study of frontal EEG asymmetry and emotion. Indeed, Coles et al. (1987) noted in their discussion of event-related brain potentials (ERPs) that knowledge of the neural substrates of ERPs is logically neither necessary nor sufcient to ascribe psychological meaning to ERP components, and this is true for frontal EEG asymmetry, as well. The ascription of psychological meaning to the ERP components, frontal EEG asymmetry measures, or any other such marker ultimately resides in the quality of the experimental design and the psychometric properties of the measurements. Although numerous aspects of the physiological bases of ERPs remain uncertain, for instance, functional relationships within specic paradigms have been established between elementary cognitive operations and components of these potentials by systematically varying one or more of the former and monitoring changes in the latter (e.g. Coles et al., 1986). This literature has also revealed that the psychological interpretation of an ERP measure may differ across paradigmsa nding that should serve as a cautionary note to those who wish to interpret EEG asymmetry measures as necessarily reecting the same substrates, predispositions, or information processing operations across different paradigms or methods of measuring frontal asymmetries (cf. Hagemann et al., 1998; Reid et al., 1998). Continued efforts are also needed, however, to further specify the neural processes or processes that are responsible for the pattern of resting EEG alpha asymmetry associated with specic facets of emotional style and those that are responsible for differential EEG alpha activation in emotionally evocative situations (cf. Davidson, 2003, 2004). Among the major evolutionary advances in humans is the striking development of the human cerebral cortex, especially the frontal region. The cerebral cortex is a mantle of between 2.6 and 16 billion neurons with each neuron receiving 10,000100,000 synapses in their dendritic trees (e.g., Pakkenberg, 1966). The frontal lobes constitute approximately 32% of this cerebral mantle. The vast expansion of the frontal regions in the human brain is largely responsible for the human capacity for reasoning, planning, and performing mental simulations, and an intact frontal region contributes to the human ability to reason, remember and work together. The neocortex in particular is a recent development in evolutionary time, and the means for guiding behavior through the environment, albeit in a more rigid and stereotyped fashion, emerged prior to neocortical expansion. These evolutionarily older systems likely also play a role in human information processing, emotion, and behavior, and ample anatomical connections between frontal and limbic regions are documented (Davidson and Irwin, 1999). The specication of the neural substrates and processes underlying specic measures of EEG asymmetry in a given paradigm can constrain and inspire theoretical hypotheses, foster experimental tests of otherwise indistinguishable theoretical explanations, and increase the comprehensiveness and relevance of theories of emotion. In addition, the measurement
241
characteristics and psychometric properties of frontal EEG asymmetry have received careful attention (e.g. see Allen et al., 2004; Hagemann, 2004), but knowledge of the specic neural structures and processes underlying tonic and phasic EEG asymmetry measures may contribute further to the accuracy of its measurement in particular paradigms and in specic populations. The point is not that either a neurophysiological or the psychological perspective is preeminent, but rather that both are fundamental to and complementary in studies of EEG asymmetry and emotion (see also, Hagemann, 2004). Inattention to the logic underlying psychophysiological inferences simply because one is dealing with observable electrocortical responses is likely to lead to simple and restricted descriptions of empirical relationships and/or to erroneous interpretations of relationships involving these responses. An entirely aphysiological attitude, in contrast, ignores relevant information and therefore increases the chances of misinterpretations of the empirical relationships that are found between EEG asymmetry measures and emotional processes or states. It is the joint consideration of physiological and psychological perspectives that enriches theory and research on human emotion by reducing errors of operationalization, measurement, and inference. The eld of research on frontal EEG asymmetry appears well positioned for such a multi-level integrative approach to the study of emotion.
2. Summary The frontal regions of the neocortex have led to dramatic expansions in emotional, behavioral, and social capacities. Perhaps it should not be surprising that research on frontal EEG asymmetry and emotion has uncovered such a diverse set of associations, therefore. Rather than reducing this literature to its simplest theme, I have discussed ways investigators might investigate whether frontal EEG asymmetry bears on separable but ultimately related facets of affective information processing, expression, and response. Metabolic and electrophysiological images of brain activity hold the promise of examining when and where affective information processing is unfolding in the brain, from which investigators will be in a better position to infer the nature of the information processing operation being performed moment by moment in the brain.
Acknowledgements Preparation of this paper was supported by National Institute of Mental Health Grant Number P50MH52384-01A1.
References
Allen, J.J.B., Coan, J.A., Nazarian, M., 2004. Issues and assumptions on the road from raw signals to metrics of frontal EEG asymmetry in emotion. Biological Psychology.
242
Allen, J.B., Harmon-Jones, E., Cavender, J.H., 2001. Manipulation of frontal EEG asymmetry through biofeedback alters self-reported emotional responses and facial EMG. Psychophysiology 38, 685693. Brazier, M.A., 1959. The historical development of neurophysiology. In: J. Field (Ed.), Handbook of Physiology. Section I: Neurophysiology, Vol. I. American Physiological Society, Washington, DC, pp. 158. Cacioppo, J.T., Gardner, W.L., Berntson, G.G., 1997. Beyond bipolar conceptualizations and measures: the case of attitudes and evaluative space. Personality and Social Psychology Review 1, 325. Cacioppo, J.T., Gardner, W.L., Berntson, G.G., 1999. The affect system has parallel and integrative processing components: form follows function. Journal of Personality and Social Psychology 76, 839855. Cacioppo, J.T., Tassinary, L.G., 1990. Inferring psychological signicance from physiological signals. American Psychologist 45, 1628. Coan, J.A., Allen, J.J.B., 2004. EEG asymmetry as a moderator and mediator of emotion. Biological Psychology. Coles, M.G.H., Donchin, E., Porges, S.W., 1986. Psychophysiology: Systems, Processes, and Applications. New York, Guilford Press. Coles, M.G.H., Gratton, G., Gehring, W.J., 1987. Theory in cognitive psychophysiology. Journal of Psychophysiology 1, 1316. Davidson, R.J., 1993. Cerebral asymmetry and emotion: conceptual and methodological conundrums. Cognition and Emotion 7, 115138. Davidson, R.J., 2003. Affective neuroscience and psychophysiology: toward a synthesis. Psychophysiology 40, 655665. Davidson, R.J., 2004. Biological Psychology, this volume. Davidson, R.J., Irwin, W., 1999. The functional neuroanatomy of emotion and affective style. Trends in Cognitive Sciences 3, 1121. Donzella, B., Davidson, R.J., Stickgold, R., Hobson, J.A., 1994. Waking electrophysiological measures of prefrontal activation asymmetry predict the emotional content of dreams. Sleep Research 23, 183. Eddington, A., 1939. The Philosophy of Physical Science. New York, Macmillan. Fox, N.A., Henderson, H.A., Rubin, K.H., Calkins, S.D., Schmidt, L.A., 2001. Continuity and discontinuity of behavioral inhibition and exuberance: psychophysiological and behavioral inuences across the rst 4 years of life. Child Development 72, 121. Greenwald, A.G., Pratkanis, A.R., Leippe, M.R., Baumgardner, M.H., 1986. Under what conditions does theory obstruct research progress? Psychological Review 93, 216229. Hagemann, D., 2004. Individual differences in anterior EEG-asymmetry: methodological problems and solutions. Biological Psychology. Hagemann, D., Naumann, E., Becker, G., Maier, S., Bartussek, D., 1998. Frontal brain asymmetry and affective style: a conceptual replication. Psychophysiology 35, 372388. Harmon-Jones, E., 2004. Contributions from research on anger and cognitive dissonance to understanding the motivational functions of asymmetrical frontal brain activity. Biological Psychology. Jones, N.A., McFall, B.A., Diego, M.A., 2004. Patterns of brain electrical activity in infants of depressed mothers who breastfeed and bottle feed: the mediating role of infant temperament. Biological Psychology. Kalin, N.H., Larson, C.L., Shelton, S.E., Davidson, R.J., 1998. Asymmetric frontal brain activity, cortisol, and behavior associated with fearful temperament in Rhesus monkeys. Behavioral Neuroscience 112, 286292. Kang, D.H., Davidson, R.J., Coe, C.L., Wheeler, R.E., Tomarken, A.J., Ershler, W.B., 1991. Frontal brain asymmetry and immune function. Behavioral Neuroscience 105, 860869. LeDoux, J.E., 2000. Emotion circuits in the brain. Annual Review of Neuroscience 23, 155184. Minnix, J.A., Kline, J.P., Blackhart, G.C., Pettit, J.W., Perez, M., Joiner, T.E., 2004. Relative left frontal activity is associated with increased depression in high reassurance-seekers. Biological Psychology. Nitschke, J.B., Heller, W., Etienne, M.A., Miller, G.A., 2004. Prefrontal cortex activity differentiates processes affecting memory in depression. Biological Psychology. Nahas, Z., Kozel, F.A., Li, X., Anderson, B., George, M.S., 2003. Left prefrontal transcranial magnetic stimulation (TMS) treatment of depression in bipolar affective disorder: a pilot study of acute safety and efcacy. Bipolar Disorders 5 (1), 4047. Pakkenberg, H., 1966. The number of nerve cells in the cerebral cortex of man. Journal of Comparative Neurology 128, 1720. Reid, S.A., Duke, L.M., Allen, J.B., 1998. Resting frontal electgroencephalographic asymmetry in depression: inconsistencies suggest the need to identify mediating factors. Psychophysiology 35, 389404.
243
Sarter, M., Berntson, G.G., Cacioppo, J.T., 1996. Brain imaging and cognitive neuroscience: toward strong inference in attributing function to structure. American Psychologist 51, 1321. Schmidt, L.A., 1999. Frontal brain electrical activity in shyness and sociability. Psychological Science 10, 316320. Tomarken, A.J., Davidson, R.J., 1994. Frontal brain activation in repressors and non-repressors. Journal of Abnormal Psychology 103, 339349. Tomarken, A.J., Dichter, G.S., Garber, J., Simien, C., 2004. Resting frontal brain activity: linkages to maternal depression and socioeconomic status among adolescents. Biological Psychology.

Biol Psychol 2004

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Biol Psychol 2004

Uploaded by

Copyright:

Available Formats

Biological Psychology 67 (2004) 15

Editorial / Biological Psychology 67 (2004) 15

Editorial / Biological Psychology 67 (2004) 15

Editorial / Biological Psychology 67 (2004) 15

Editorial / Biological Psychology 67 (2004) 15

Biological Psychology 67 (2004) 749

Frontal EEG asymmetry as a moderator and mediator of emotion

J.A. Coan, J.J.B. Allen / Biological Psychology 67 (2004) 749

J.A. Coan, J.J.B. Allen / Biological Psychology 67 (2004) 749

J.A. Coan, J.J.B. Allen / Biological Psychology 67 (2004) 749

J.A. Coan, J.J.B. Allen / Biological Psychology 67 (2004) 749

Table 1 Trait frontal EEG asymmetry and other trait-like measures

Buss et al., 2003

From 85 infants samples sizes for specic analyses vary

250 (132 female)

J.A. Coan, J.J.B. Allen / Biological Psychology 67 (2004) 749

Coan and Allen, 2003 Davidson et al., 1999

RFA, NK (rest) RFA, NK (exam) LFA, NK (pos lm clip)

Ehlers et al., 2001

134 (68 female)

Field et al., 2002

48 infants (29 female)

Infant EEG at F3/4, P3/4

Fox et al., 1995

4962 months Mean = 24.7

EEG at F3/4, P3/4 and O1/2

J.A. Coan, J.J.B. Allen / Biological Psychology 67 (2004) 749

RFA, SC LFA, SC NA, LATA

Hagemann et al., 1999

37 females 26 (11 female) 40 males 47 (30 female)

No information Mean = 13 years 1853 years 5760 years

J.A. Coan, J.J.B. Allen / Biological Psychology 67 (2004) 749

Kang et al., 1991

Extreme LFA and RFA groups

Kalin et al., 2000

49 (23 female; rhesus monkeys)

Longitudinal data at 4, 8, 14, 40 and 52 months 1733 years

Extreme LFA and RFA groups in monkeys

Kline et al., 1998

Linked ears (LE)

Defensive coping (EPQ-L scale)

Mean = 20.4 Mean = 20.4

235 (141 females)

Moss et al., 1985

12 (Japanese), 12 (Western), all Female

J, mean = 32.6 years; W, mean = 29.1 years

EEG at T3/4 and P3/4

Merckelbach et al., 1996 Schmidt, 1999

2238 years M = 20.97 years

EEG at F3/4 and P3/4 EEG at F3/4, P3/4 and O1/2

Schmidt and Fox, 1994

EEG at F3/4, P3/4, A1/2 and O1/2

J.A. Coan, J.J.B. Allen / Biological Psychology 67 (2004) 749

Stough et al., 2001

Sutton and Davidson, 1997

Tomarken and Davidson, 1994 Tomarken et al., 1992a

No information 1721 years

Urry et al., in press

LFA in FC3/4, EWB LFA in FC3/4, HWB

J.A. Coan, J.J.B. Allen / Biological Psychology 67 (2004) 749

J.A. Coan, J.J.B. Allen / Biological Psychology 67 (2004) 749

J.A. Coan, J.J.B. Allen / Biological Psychology 67 (2004) 749