You are on page 1of 16

PERSONALITY AND SOCIAL PSYCHOLOG Y BULLETIN

Agnew et al. / PROXIMAL AND DISTAL INFLUENCES ON PREJUDICE

Incorporating Proximal and Distal


Influences on Prejudice: Testing a
General Model Across Outgroups

Christopher R. Agnew
Purdue University
Vaida D. Thompson
University of North Carolina at Chapel Hill
Stanley O. Gaines, Jr.
Pomona College

The present research integrates various social psychological (Tajfel & Turner, 1985). Although these past research
approaches to understanding the causes of traditional prejudice. efforts have added to our understanding of intolerant
The authors examined (a) whether conceptually distinct vari- attitudes, it is generally agreed that no individual factor
able sets shown previously to predict prejudice could be modeled or approach provides a comprehensive explanation of
collectively within a proximal-distal framework and (b) whether prejudice (Duckitt, 1992a). Moreover, continued pur-
different outgroups could be modeled collectively within this suit of single-factor explanations is unlikely to generate a
framework. The authors developed and tested a model that useful theoretical framework capable of integrating the
included four sets of explanatory factors derived from past considerable amount of knowledge that currently exists
research: Family Status, Contextual Exposures, Beliefs, and Per- regarding prejudice.
sonality. It was hypothesized that the influence of these factors In this article, we examine the multiple roots of overt,
could be represented in a causal sequence such that (a) the distal negative attitudes toward members of one’s outgroups.
factors (i.e., Family Status and Contextual Exposures) would In the more recent literature, this type of prejudice is
lead to both proximal factors (i.e., Beliefs and Personality) and sometimes labeled “traditional” or “old-fashioned” to
(b) the proximal factors would lead directly to Negative Attitudes distinguish it from more modern and covert expressions
Toward Outgroups. Structural equation analyses of data of outgroup rejection (Dovidio & Gaertner, 1991).
obtained from two independent samples generally supported the
Authors’ Note: This research was supported by a summer research
model—the impact of distal factors on prejudice was mediated grant to the first and second authors from the Institute for Research in
largely by proximal factors. Social Science (IRSS) at the University of North Carolina at Chapel
Hill, a National Institute of Child Health and Human Develop-
ment–National Research Science Award (NICHD-NRSA) predoctoral

W hat causes prejudice? In searching for causal expla-


fellowship from the Carolina Population Center to the first author, and
a postdoctoral fellowship from the University of North Carolina at
nations of prejudice toward various outgroups, past Chapel Hill and institutional funds from Pomona College to the third
research has addressed and supported a number of author. We thank David Currey, John Dovidio, Chet Insko, Abigail Pan-
ter, Bev Wiggins, and several anonymous reviewers for thoughtful com-
single-factor solutions. The factors explored have been
ments on previous drafts of this article, and we thank Eliot Smith and
diverse, including sociodemographic factors (e.g., an Howard Weiss for comments concerning analysis issues. We also thank
association between urbanism and tolerance for various the administration and staff of IRSS for assistance with various phases
outgroups) (Abrahamson & Carter, 1986), family sociali- of this research and Dustin Walling for assistance with Sample 2 data
zation processes (e.g., prejudice stemming from domi- collection. Correspondence concerning this article should be sent to
Christopher R. Agnew, Department of Psychological Sciences, Purdue
neering parents) (Harris, Gough, & Martin, 1950), per-
University, West Lafayette, IN 47907-1364; e-mail: agnew@psych.pur-
sonal orientations (e.g., authoritarianism) (Adorno, due.edu.
Frenkel-Brunswik, Levinson, & Sanford, 1950), and
PSPB, Vol. 26 No. 4, April 2000 403-418
cognitive-motivational processes (e.g., social identity) © 2000 by the Society for Personality and Social Psychology, Inc.

403
404 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

Recent years have witnessed a decline in research theoretical model (Agnew, Thompson, Smith,
emphasis on overt prejudice despite strong evidence for Gramzow, & Currey, 1993). This research, focusing on
its continued presence in the world (e.g., the ethnic negativity toward homosexuals, followed from the think-
cleansing in the former Yugoslavia) (see Jones, 1997). ing of Allport and was heuristically guided by conceptu-
We believe it remains important to focus research atten- alizations of Fishbein and Ajzen pertaining to predic-
tion on the social psychological mechanisms that under- tions of behavioral intentions and behavior. As with
lie traditional prejudice. Such attention is necessary Allport’s (1954) position and the reasoned action model
(although perhaps not sufficient) for progress to be (Ajzen & Fishbein, 1980), we theorized that a number of
made toward eliminating such negativity. variables, representing varying degrees of proximity to
Past social psychological searches for the cause(s) of the attitude, should predict to prejudicial attitudes. We
traditional prejudice often appear to have been guided used the terms “proximal” and “distal” to indicate the
more by changing fashions and historical events than by theorized degree of immediate influence of a variable
empirical data and are decidedly focused in their causal class on a person’s current attitudes. We used the term
emphasis (see Duckitt, 1992a; Gaines & Reed, 1995; “distal” to denote those factors that are preexisting, con-
Stroebe & Insko, 1989). This is unfortunate given the ceptualizing those family factors that existed at the time
caveats offered over the years by leading thinkers in the of birth as being the most remote from current attitudes,
area. For example, Pettigrew (1958) argued that the intentions, and behavior. We used the term “proximal”
major psychological and sociological processes routinely to denote factors that are more immediate to behavior,
addressed separately by researchers are, in fact, “elabo- conceptualizing such factors as concomitant percep-
rately intertwined and . . . both are essential to provide an tions as being most proximal to current attitudes, inten-
adequate theoretical framework for this complex phe- tions, and behavior. We reasoned that the more theoreti-
nomenon” (p. 29). More recently, Hamilton and Trolier cally proximal the variable type, the stronger should be
(1986) point out that its prediction of the attitude. The most distal variables
should predict to the attitude but each more proximal
any particular form of stereotyping or prejudice, such set of variables should yield a stronger prediction.
as racism, is in all likelihood multiply determined by Why should proximal variables be more predictive
cognitive, motivational, and social learning processes,
than distal variables of current prejudicial attitudes?
whose effects combine in a given social context to
produce specific judgmental and behavioral manifesta-
Although preexisting family factors and past experi-
tions. Therefore, any attempt to understand such ences are necessary and important precursors of current
phenomena as a product of one process alone is proba- behaviors, behavior is primarily determined by the indi-
bly misguided. (p. 153) vidual’s immediate perceptions and situational influ-
ences. For example, whereas early family environment
Although theoretical and empirical pursuit of single influences what kind of person one becomes, it is cur-
explanations for prejudice may have hampered the rent situational pressures and one’s perception of these
development of more comprehensive models, there are forces that influence how one will act at the present time.
models that include multiple causal determinants, As one develops, one’s earliest experiences become
including societal, familial, individual, and situational increasingly removed as predictors of present behavior,
variables. Notably, early writings by Allport (1954) sug- whereas one’s immediate psychological state and the
gest that prejudice involves a multilevel, multifactor set immediate situation become increasingly prominent as
of causes, ranging from those that are remote to those predictors of behavior. Extrapolating from the reasoned
that are immediate. Describing factors varying in prox- action model, this same logic pertains to the prediction
imity to prejudice, Allport cited historical factors as the of current attitudes, themselves a relatively proximal
most remote influence, with sociocultural, situational, predictor of behavior. Distal factors such as those exist-
personality, phenomenological, and stimulus object fac- ing in the family of origin should not be as predictive of
tors becoming, in turn, increasingly more immediate. current attitudes as are one’s current beliefs and
Others (e.g., Ashmore, 1970; Duckitt, 1992a; Simpson & personality.
Yinger, 1985) have also proposed that similar sets of vari- The proximal-distal model that we developed for the
ables are influential. Unfortunately, tests providing investigation of attitudes toward homosexuals (Agnew
empirical support for such multivariate frameworks are et al., 1993) obtained substantial empirical support. The
lacking. model contained five broad sets of variables that were
theorized to be increasingly proximal to attitudinal
A Proximal-Distal Approach
negativity toward homosexuals: distal family factors
We have previously investigated prejudice toward a (e.g., parental education), distal individual factors (e.g.,
specific outgroup using a multivariate proximal-distal dogmatism), proximal family factors (e.g., parental atti-
Agnew et al. / PROXIMAL AND DISTAL INFLUENCES ON PREJUDICE 405

tude toward homosexuality), proximal individual fac- toward criminals, delinquents, prostitutes, the mentally
tors (e.g., current religious beliefs), and proximal ill, and drug addicts. Bierly (1985) also reported strong
individual-situational factors (e.g., acquaintance with empirical relations between attitudes toward homosexu-
homosexuals). When entered into separate analyses, als, African Americans, women, and old people. Consis-
each of these five variable sets was significantly predictive tent with these findings, Weigel and Howes (1985) stud-
of attitudes toward homosexuals. We also found that the ied a sample of White, New England residents and
more proximal variable sets tended to be increasingly showed significant intercorrelations among measures of
stronger predictors of such attitudes. Furthermore, we prejudice toward African Americans, homosexuals, and
found that the proximal variables not only accounted for the elderly. Finally, Crandall (1994) recently reported a
more variance in prejudice but also subsumed the effects significant association between antifat prejudice and
of the most distal variables when both were included in a measures of both traditional and modern racism.
hierarchical regression model. In contrast to the above findings, Ray and Lovejoy
(1986) obtained empirical support for generalizability
Focusing on Prejudice
of outgroup negativity toward Asians, Blacks, Southern
Toward Multiple Outgroups
Europeans, and Jews but found that different factors
To extend and complement our past research, we contributed to rejection of different groups. However, as
believe it is necessary to direct greater attention to preju- in most of the extant research, predictor constructs used
dice pertaining across outgroups rather than to focus on were limited in nature and number, rendering conclu-
the causes of prejudice toward a specific outgroup. This sions about the source of specific or generalized rejec-
is particularly important in relation to any theoretical tion speculative. Similarly, Stangor, Sullivan, and Ford
speculation about the primacy of a specific factor (or fac- (1991) found what appeared to be generalizability of
tors) in influencing prejudice because focus on preju- prejudice whereby emotional responses were “a more
dice toward a distinct group may not confirm whether consistent and stronger predictor of attitudes and social
prejudice toward other groups could be explained by distance than were social stereotypes” (p. 359) in rela-
the same factor(s). It seems reasonable to assume that tion to various outgroups. However, affective responses
prejudice toward one group indicates an increased ten- to different groups were not similar, suggesting some
dency to harbor prejudice toward other groups. Such a degree of nongeneralizability. Thus, throughout the lit-
position is consistent with Allport’s (1954) view that erature there is evidence for both generality and non-
“people who reject one out-group will tend to reject generality of prejudice across outgroups, but predictor
other out-groups. If a person is anti-Jewish, he is likely to variables employed have been limited in nature.
be anti-Catholic, anti-Negro, anti- any out-group” (p. 68).
The Present Research
This point also has been made more recently by Ray and
Lovejoy (1986), who hold that “to be ethnocen- Given the above observations, there is a need for
tric . . . implies that one will dislike all out-groups. To dis- empirical research that examines both the multivariate
like just one out-group would be incoherent” (p. 563). nature of prejudice and the potential generality of rejec-
However, research exploring the generalizability of tion across outgroups. Such research requires both a
prejudice rarely includes both a full set of predictor vari- complete set of predictors and a reasonably varied set of
ables and multiple, distinctive outgroup targets suffi- distinctive outgroups toward which negative attitudinal
cient to provide a definitive answer to the generalizabil- responses can be explored. As noted earlier, our past
ity question. As a result, research may demonstrate the research (Agnew et al., 1993) examined only one out-
presence or absence of generalizability but fail to pro- group (homosexuals), with no conclusions necessarily
vide a full causal picture for either. applicable across different outgroups. Moreover, we did
Past empirical findings concerning generalizability of not examine the linkages that may exist between proxi-
prejudice are mixed but appear to support the notion of mal and distal variable sets.1 In the present research, we
generality. The classic authoritarian personality research examine these linkages and address two distinct research
by Adorno and colleagues (Adorno et al., 1950) pro- questions: (a) Can the multiple antecedents of tradi-
vided early evidence that particular individuals are intol- tional prejudice be modeled within a multivariate
erant of a wide variety of outgroups. Butler’s (1969) find- explanatory model, one that incorporates simultane-
ings also suggest generality of negativity toward racial, ously both proximal and distal influences? and (b) Can a
class, and age outgroups. Similarly, in a review of a large generalizable model of prejudice (that includes multi-
number of studies, Batson and Ventis (1982) found ple outgroup targets) be supported by this single causal
strong evidence that particular religious orientations framework? We were interested especially in testing both
and behaviors were associated with prejudice toward an “indirect effects” model, in which the influence of dis-
African Americans as well as with being more punitive tal variables on prejudice is mediated by the proximal
406 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

variables, and a “direct effects” model, in which the distal


variables exert a direct influence on prejudice above and
beyond that which would be obtained solely by mediator
(i.e., proximal) variables.
Hypothesized Proximal-Distal
Model of Prejudice Across Outgroups
An adequate multivariate model of prejudice should
be both parsimonious and reflective of the major antece-
dents of prejudice included in earlier frameworks. At a
minimum, four sets of explanatory factors should be
included:2 (a) Family Status, representing historical,
socioeconomic precursors to an individual’s current
social attitudes; (b) Contextual Exposures representing
early social and environmental conditions that affect the
formation of an individual’s outgroup attitudes; (c)
Beliefs reflecting the current general cognitive orienta-
tions of the individual that inform social attitudes; and Figure 1 Proximal-distal model of prejudice, with Paths 1 to 8 repre-
(d) Personality, reflecting individual differences in the senting the indirect effects model (Model 1) and the addi-
tion of Paths 9 and 10 representing the direct effects model
propensity to currently possess prejudicial attitudes. (Model 2).
Two of these sets, Family Status and Contextual Expo-
sures, can be considered to be distal, reflecting struc- Contextual Exposures are seen as distal factors that tap
tural factors that shape the nature of exposure to out- early influences hypothesized to relate to later proximal
groups early in life. The other two sets, Beliefs and factors.
Personality, are proximal, reflecting current defining Consistent with past research, conservative beliefs
aspects of the individual. and tolerant personality traits are hypothesized to be the
We hypothesized that the influence of these four fac- most proximal and direct predictors of prejudice. Beliefs
tors could be represented in a causal sequence such that have been viewed traditionally as a component of atti-
the two distal factors would lead to and shape the two tude (Fishbein & Ajzen, 1975; Insko & Schopler, 1967;
proximal factors and the two proximal factors would McGuire, 1985), and conservative beliefs in particular
lead directly to Negative Attitudes Toward Outgroups. have been found to relate to prejudiced attitudes (for a
Note that the effects of the two distal factors on prejudice review, see Duckitt, 1992b). Personality traits also have
are hypothesized to be indirect in that their effect on been shown to relate to prejudice (Adorno et al., 1950;
prejudice is predicted to be only through the two inter- Duckitt, 1992b; Simonton, 1990). In the hypothesized
vening proximal factors. Our hypothesized theoretical model shown in Figure 1, note that we suggest that these
model (Model 1) is presented in Figure 1 (Paths 1 to 8). two proximal factors are intercorrelated, with neither
As can be seen in Figure 1, Family Status is posited as factor posited as necessarily causing the other.
affecting the Contextual Exposures that the individual To test the generality of the proposed model, it was
confronts early in life as well as influencing the posses- necessary to measure attitudes toward several distinct
sion of particular Beliefs and Personality traits in later outgroups simultaneously. Furthermore, we reasoned
life. Consistent with past sociological theorizing and that the groups should be conceptually dissimilar from
findings (e.g., Wilson, 1985; Wirth, 1938), it is reasoned one’s own in more than one way, for instance, people of a
that individuals raised in families of higher status (with different race and people of a different sexual orienta-
more education and income) are more likely to have tion. Several such outgroups were identified and are
opportunities to experience environments (both physi- included in our model tests.
cal and social) that allow for increased contact with a het-
An Alternative Model:
erogeneous population. Increased contact is seen as fos-
Direct Effects of Distal Factors
tering the formation of a more open belief system and a
more tolerant personality.3 Furthermore, higher status It is possible that one or more alternative models pro-
families are less likely to espouse particular religious ten- vide a better theoretical explanation of the underpin-
ets in the home or regularly attend religious services, nings of prejudice than does our hypothesized model
behaviors that have been shown (paradoxically) to lead (see MacCallum, Wegener, Uchino, & Fabrigar, 1993).
to an increase in prejudice (Batson & Ventis, 1982; One such model, which we call a direct effects model,
Braithwaite & Scott, 1991). Thus, both Family Status and includes direct causal paths from the two distal factors to
Agnew et al. / PROXIMAL AND DISTAL INFLUENCES ON PREJUDICE 407

prejudice (i.e., from Contextual Exposures to Negative Family status. Three individual CAPS variables were
Attitudes Toward Outgroups [Path 9 in Figure 1] and used to assess the general socioeconomic status of each
from Family Status to Negative Attitudes Toward Out- participant’s family while he or she was growing up.
groups [Path 10 in Figure 1]), which were not included These were father’s education, father’s occupational
in the hypothesized indirect effects model. The plausi- status, and social class of the family.
bility of the direct effects model rests on a crucial
assumption from the standpoint of sociologically ori- Social and environmental contextual exposures. Four
ented social psychological theory: Above and beyond measures were used to assess the contextual exposures
the mediated influence of Family Status and Contextual experienced by each participant early in life. Two of
Exposures on prejudice, both factors may be expected to these represented early exposure to religion (religious
exert direct influences on prejudice (Stryker & Statham, exposure outside the home during high school and
1985). Given the potential importance and plausibility moral/religious emphases in the home while growing
of such additional direct causal paths, we evaluated the up) and two represented early environmental condi-
goodness of fit of a direct effects model in absolute terms tions (development of the community and community
and in comparison to the hypothesized indirect effects size during high school).
model. Conservative beliefs. Three standard measures were
To test the hypothesized model as well as possible selected to assess the current level of conservative
alternatives, we used data collected from two independ- beliefs. These were orthodox religious beliefs, general
ent samples. conservatism, and traditionality in gender-role beliefs.

SAMPLE 1 Tolerant personality. Three scales from widely used per-


sonality inventories were used to measure current per-
Method sonality traits associated with prejudice: openness (from
the NEO Personality Inventory) (Costa & McCrae, 1985)
Study database and model variables. The first sample was
and flexibility and tolerance (from the California Per-
derived from a unique, extant database, the same data-
sonality Inventory [CPI]) (Gough, 1956). By proxy,
base that we used in our earlier research on attitudes
these measures can be thought to tap tendencies toward
toward homosexuals (Agnew et al., 1993). Specifically,
authoritarianism (Adorno et al., 1950).
data were collected by the Institute for Research in Social
Science at the University of North Carolina at Chapel Negative attitudes toward outgroups. Three composite
Hill (UNC-CH) from randomly selected samples of male attitude scales were used to assess prejudice toward out-
and female undergraduate students over a 5-year period groups. As the alphas reported in the appendix show,
(1983-1988) as part of the Computer Assisted Panel these scales reliably measured attitude toward homo-
Study (CAPS) (Latané, 1989). Over the course of a given sexuals, toward other races, and toward elderly persons.
9-month academic year, CAPS participants completed
anonymously a number of different measures each week Participants. The study included 288 undergraduates at
via computer. By the end of their participation, most UNC-CH who participated in CAPS over the 3-year period
CAPS participants had completed more than 200 sepa- from 1986 to 1988. These students, 96 of whom partici-
rate measures. Thus, the CAPS database contains a wide pated in each of the 3 years, were selected randomly from
variety of variables, including (a) family status variables admission records to participate. Because only measures
(largely demographic information); (b) contextual that were collected during each of the 3 years were
exposure variables (including early lifetime social and employed in our model, we merged the data from the 3
environmental conditions); (c) beliefs, attitudes, and years to create a combined data set for analyses.4
values; (d) measures of personality; and (e) three meas- To test the proposed generalized outgroup model, it
ures of prejudiced attitudes (specifically, attitudes was necessar y to exclude those participants who
toward homosexuals, other races, and the elderly). reported that they were homosexual (n = 7) or bisexual
Based on the array of data available and an examination (n = 3), who declined to state their sexual orientation (n =
of the psychometric properties of the measures, we 8), or who failed to complete all model measures (n =
selected the subset of measures detailed in the appendix 65). This reduced the sample size to 205, 101 males and
as most conceptually consistent with the factors in the 104 females. Mean comparisons between this reduced
hypothesized model. The possible scale range, mean, sample and the 65 “missing data” participants indicated
standard deviation, and brief description of each meas- no significant differences on those variables for which
ure are provided in the appendix. Individual variables data were not missing. Age ranged from 17 to 24 years,
chosen to measure each model factor are reviewed with a mean age of 20. The reported ethnicity of the sam-
briefly below. ple closely matched that of the university: 85% White,
408 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

10% African American, 3% Asian American, and 2% females. Age ranged from 18 to 35 years, with a mean age
other (or unreported) ethnicity. of 22.
Procedure. Participants sequentially completed five
SAMPLE 2 separate packets of questionnaires, each containing a
We collected data from a second sample to extend number of different self-report measures. As detailed
and complement the data obtained from the first sam- below, the first four packets contained the measures of
ple. There were several notable methodological differ- Family Status, Contextual Exposures, Beliefs, and Per-
ences. First, the extant nature of the Sample 1 data did sonality. The fifth packet contained the outgroup atti-
not permit us to operationalize all conceptual variables tude measures. The order of these outgroup attitude
in the most direct fashion (unfortunately, such problems measures was counterbalanced to determine the pres-
are often encountered in research that uses secondary ence of possible order effects (subsequent analyses
data) (Dale, Arber, & Procter, 1988). That is, we found it revealed no such effects).
necessary to adopt measures that fit the model compo- Model variables. The appendix lists all measures com-
nents conceptually but that had not been designed spe- pleted by Sample 2. Notable measurement differences
cifically to test the temporal ordering implicit in the with the measures completed by Sample 1 will be high-
hypothesized model. It would have been preferable to lighted here. In the measurement of Family Status, par-
have distal measures that specified precise temporal ticipants were asked to provide both parents’ education
periods (e.g., asking participants to provide answers to level and occupational status. An averaged parental
distal items with respect to the period “while you were measure of education and occupational status was then
growing up”). computed and used in analyses. In the measurement of
In addition, the measurement of Contextual Expo- social and environmental contextual exposures, partici-
sures in the first sample relied heavily on religion- pants were asked about the racial composition of the
oriented measures, such as early religious service atten- neighborhood in which they grew up. In addition, the
dance and family religiosity. Because the measurement amount of contact with the four outgroups under investi-
of Conser vative Beliefs also contained a religion- gation was assessed and used as an indicator of early life
oriented variable (religious orthodoxy), it is possible exposure to outgroup members. In the measurement of
that any latent association between Contextual Expo- Negative Attitudes Toward Outgroups, we used three
sures and Conservative Beliefs is due to this shared meas- published scales. As the alphas reported in the appendix
urement base and not due to an association at the show, these scales reliably measured attitudes toward
hypothesized latent level. Accordingly, with Sample 2 we homosexuals, toward other races, and toward elderly
administered several new variables to operationalize persons. Because we could find no standard scale assess-
both Family Status and Contextual Exposures. These ing prejudice toward foreigners, we created an attitude
variables more closely capture our intended conceptu- toward foreigners scale by adopting McConahay’s
alization of Contextual Exposures. (1986) measure of traditional racism. Alpha analyses
Third, to explore the model’s applicability to an even confirmed the internal reliability of this derived attitude
broader set of outgroups, we included a fourth group toward foreigners measure (see the appendix).
(foreigners) in our measurement of Negative Attitudes
Toward Outgroups. Moreover, with our second sample
ANALYTIC STRATEGY
we used previously published prejudice scales where pos-
sible, benefiting from the past measurement research on In our analyses, we followed a two-step analytic strat-
racism by McConahay (1986); on ageism by Fraboni, egy suggested by Schumacker and Lomax (1996) based
Saltstone, and Hughes (1990); and on antigay prejudice on earlier work by Anderson and Gerbing (1988) as well
by Kite and Deaux (1986). as James, Muliak, and Brett (1982). These researchers
emphasize the desirability of analyzing two conceptually
Method
distinct confirmatory models: (a) a measurement model
Participants. The study included 225 Purdue Univer- and (b) a structural model. The measurement model,
sity undergraduates who completed self-report question- tested via confirmatory factor analysis, specifies the asso-
naires in partial fulfillment of a psychology course ciations between measured variables underlying
requirement. To test the proposed generalized out- hypothesized latent factors and provides an assessment
group model, it was necessary to exclude those partici- of convergent, discriminant, and construct validity.
pants who reported that they were not White (n = 26), Assuming an adequate measurement model, one then
heterosexual (n = 6), or an American citizen (n = 13), or proceeds to the structural model. The structural model,
who failed to complete all model measures (n = 3). This tested via structural equation analysis, specifies the
reduced the sample size to 189, 118 males and 71 hypothesized associations among the latent factors and
Agnew et al. / PROXIMAL AND DISTAL INFLUENCES ON PREJUDICE 409

TABLE 1: Correlations Among Measured Variables for Each Hypothesized Latent Factor and Standardized Loadings on Latent Factors,
Samples 1 and 2

Loadings
Measured Variable 1 2 3 4 5 6 S1 S2

Family status
1. Father’s/parents’ education 1.00 .46*** .48*** — — — .38 1.00
2. Father’s/parents’ occupational status .39*** 1.00 .44*** — — — .42 .60
3. Social class of family .32*** .31*** 1.00 — — — 1.00 .69
Social and environmental contextual exposures
1. Religious exposure outside the home 1.00 — — — — — .75 —
2. Moral/religious emphases in home .61*** 1.00 — — — — 1.00 —
3. Early outgroup contact — — 1.00 .39*** .22** .25*** — .41
4. Early neighborhood racial mix — — — 1.00 .21** .37*** — .47
5. Development of community .16* .22** — — 1.00 .51*** .27 .60
6. Size of community .29*** .25** — — .37*** 1.00 .41 1.00
Conservative beliefs
1. Orthodox religious beliefs 1.00 .43** .45*** — — — .84 .51
2. General conservatism .50*** 1.00 .44*** — — — 1.00 .66
3. Traditionality in gender-role beliefs .14* .33*** 1.00 — — — .44 1.00
Tolerant personality
1. Openness 1.00 .52*** .29*** — — — 1.00 .72
2. Flexibility .44*** 1.00 .38*** — — — .63 1.00
3. Tolerance .35*** .40*** 1.00 — — — .75 .54
Negative attitudes toward outgroups
1. Attitude toward homosexuals 1.00 .37*** .32*** .48*** — — 1.00 .68
2. Attitude toward other races/Blacks .51*** 1.00 .37*** .62*** — — .91 .72
3. Attitude toward elderly persons .18** .22** 1.00 .36*** — — .31 .47
4. Attitude toward foreigners — — — 1.00 — — — 1.00

NOTE: Sample 1 (S1) correlations are displayed below the diagonal. Sample 2 (S2) correlations are displayed above the diagonal.
*p < .05. **p < .01. ***p < .001.

can be thought of as providing an assessment of ship between the latent factors but not necessarily in the
nomological validity. precise manner specified in our hypothesized model.
Following this two-step approach, we began by assess- Next, to test our hypothesized structural model and
ing the measurement model, determining whether each potential alternative structural models, we analyzed the
set of measured variables hypothesized to measure a data from both samples simultaneously by conducting
given second-order factor (shown in Figure 1) actually two-group (also referred to as multiple group) analyses
measured that latent factor in a valid and reliable man- using LISREL. Two-group analyses test whether the
ner within each sample (see Bernstein, Garbin, & Teng, obtained pattern of associations observed between
1988; Nunnally, 1967). To accomplish this, we used latent factors differs significantly between groups (or
LISREL to conduct two separate confirmatory factor samples). Besides providing a parsimonious manner of
analyses, one for each sample, in which we tested model testing, two-group structural analyses are particu-
whether each measure loaded significantly on the larly useful in cases in which the groups have an unequal
hypothesized latent factor. As part of the confirmatory number of measured variables. Given that Sample 1 had
factor analyses, we tested whether an orthogonal or a total of 16 measured variables and Sample 2 had 17
oblique solution provided a significantly better model fit measured variables (including the additional foreigner
to the sample data. This type of model comparison indi- measure), two-group analyses provides an excellent
cates whether the latent factors postulated to underlie a method for testing the equality of factor structure across
set of measured variables are or are not intercorrelated the two samples.
with another given the data (Schumacker & Lomax,
1996). Support for an orthogonal model would argue RESULTS
against our hypothesized model in that such a model
Confirmatory Factor Analyses
would hold that there are no relationships between
latent factors. Support for an oblique model would be Table 1 presents the simple correlations between vari-
consistent with our hypothesized model in that such a ables hypothesized to comprise each latent factor for
model would indicate that there is at least some relation- Samples 1 and 2. As can be seen in the table, the variables
410 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

postulated to measure each construct were significantly Test of the indirect effects model (Model 1, Paths 1 to 8). As
intercorrelated. In addition, the loadings derived from Table 2 reveals, seven of the eight paths hypothesized to
the separate confirmatory factor analyses of the meas- be significant were significant, with the valences for
ured variables are displayed in Table 1. In conducting those paths in the expected directions. The exception
these confirmatory analyses, note that (a) maximum involves the direct path between Family Status and Con-
likelihood estimation methods were employed, a servative Beliefs. Essentially, no linear relation was found
method preferred due to its desirable psychometric between these two latent factors (standardized beta =
properties (Jöreskog & Sörbom, 1989); (b) one variable –.04). As for overall model fit, results of the structural
in each measured set was set to 1 (often referred to as the equation analysis indicated that the indirect effects
reference variable) (Jöreskog & Sörbom, 1993) and the model provided a satisfactory fit to the data, χ2(17) =
others in the set allowed to vary; and (c) LISREL’s ridge 24.03, ns, Goodness-of-Fit Index (GFI) = .98, standard-
option was employed, allowing estimations in the pres- ized Root Mean Square Residual (RMR) = .07, with the
ence of multicollinearity. For both samples, the obtained chi-square/degrees-of-freedom ratio (i.e., 1.41) well
factor loadings indicate that the measured variables within the range of values commonly viewed as accept-
hypothesized to load on a given latent factor did so gen- able (i.e., 1.0 to 2.0) (Loehlin, 1992). The total amount
erally at a moderate to high level. of variance in the criterion variable (i.e., Negative Atti-
In addition, model comparisons of orthogonal versus tudes Toward Outgroups) accounted for by the model
oblique solutions suggest support for an oblique solu- was 57%.
tion. For Sample 1, the obtained chi-square value was
Test of the direct effects model (Model 2, Paths 1 to 10).
175.23 (df = 109) for an orthogonal solution and 77.44
Although the a priori hypothesized indirect effects
(df = 99) for an oblique solution. The chi-square differ-
model could not be rejected in terms of goodness-of-fit
ence test for significance indicated that the oblique solu-
criteria, it is possible that an alternative direct effects
tion provided a significantly better fit to the data than
model would provide an even better fit to the data. Such
did the orthogonal solution, χ2 difference (10) = 97.79,
a model, as described in the Introduction, adds direct
p < .01. Similarly, for Sample 2, the obtained chi-square
causal paths (a) from Contextual Exposures to Negative
value was 101.64 (df = 124) for an orthogonal solution
Attitudes Toward Outgroups (Path 9 in Figure 1) and (b)
and 61.65 (df = 114) for an oblique solution, with the
from Family Status to Negative Attitudes Toward Out-
oblique solution providing a significantly better fit, χ2
groups (Path 10 in Figure 1). The results for the direct
difference (10) = 39.99, p < .01. On the basis of the corre-
effects model (Model 2) are shown in Table 2. In and of
lational findings, the confirmatory factor analyses, and
itself, the direct effects model provided a marginal fit to
the oblique versus orthogonal model comparisons, we
the data, χ2(15) = 23.90, p <.07, GFI = .98, standardized
concluded that our measurement models were adequate
RMR = .07. Of importance, as shown in Table 2, neither
in both samples.
added direct path approached statistical significance.
Structural Equation Analyses
Comparing Model 2 with Model 1. We then compared the
Next, to assess the structural model, we performed overall fit of Model 2 with Model 1 by computing the dif-
two-group structural equation analyses in which the ferences between the chi-squares and degrees-of-
goodness of fit of Model 1 (shown in Figure 1, Paths 1 to freedom associated with each model (Loehlin, 1992). To
8) was assessed. All parameters were estimated using the support Model 2 (the direct effects model) the loss in
maximum likelihood method, with ridge option. With degrees of freedom corresponding to the inclusion of
the exception of the correlation between Conservative the two additional paths in that model would have to be
Beliefs and Tolerant Personality (which was predicted to offset by a significant reduction of chi-square value from
be negative), all off-diagonal elements of the variance- Model 1 (the indirect effects model). Otherwise, adop-
covariance matrix were fixed at 0. Moreover, all elements tion of Model 2 would amount to sacrificing theoretical
in the beta weight matrix corresponding to paths in Fig- and statistical parsimony for the sake of a negligible
ure 1 were freed. All measurement error terms were esti- change in chi-square (Loehlin, 1992; see also MacCal-
mated by LISREL but were constrained to be equal lum et al., 1993). The chi-square test for significance
across factors and across samples. Obtained latent factor- indicated that Model 2 did not provide a significantly
pooled, within-group correlations ranged from .23 to .67 better fit than Model 1, χ2 difference (2) = .13, ns. There-
(all ps < .001). The resulting standardized beta weights, fore, we concluded that the direct effects model fit the
correlation, and residuals associated with the elements correlational data no better than the hypothesized indi-
of the hypothesized structural model are shown in Table 2. rect effects model and was less parsimonious in its
Agnew et al. / PROXIMAL AND DISTAL INFLUENCES ON PREJUDICE 411

TABLE 2: Two-Group Structural Results for Indirect- and Direct-Effects Models, Across Outgroups and by Outgroup

Overall Model Fit Indices Standardized Loadings by Path


2
c df GFI RMR 1 2 3 4 5 6 7 8 9 10

Across outgroups
Indirect Effects Model (Model 1) 24.03 17 .98 .07 –.04 .44** .28* –.43** .20* –.36** .46** –.40** — —
Direct Effects Model (Model 2) 23.90 15 .98 .07 –.03 .44** .27* –.44** .21* –.36** .48** –.39** .04 –.04
Residuals (ε) for Model 1/Model 2: Contextual Exposures = .81/.81, Conservative Beliefs = .80/.80,
Tolerant Personality = .84/.84, Negative Attitudes Toward Outgroups = .44/.43

Homosexuals
Indirect Effects Model (Model 1) 15.93 17 .98 .12 –.03 .42** .25* –.40** .17 –.33** .40** –.29* — —
Direct Effects Model (Model 2) 15.10 15 .98 .11 –.02 .43** .25* –.43** .19 –.32** .47** –.29* .13 –.03
Residuals (ε) for Model 1/Model 2: Contextual Exposures = .82/.81, Conservative Beliefs = .83/.80,
Tolerant Personality = .87/.86, Negative Attitudes Toward Outgroups = .66/.62

Other races/Blacks
Indirect Effects Model (Model 1) 28.28 17 .97 .16 –.02 .44** .25* –.44** .20 –.32** .41** –.30* — —
Direct Effects Model (Model 2) 28.11 15 .97 .16 –.03 .43** .25* –.42** .20 –.32** .38* –.29** –.06 .01
Residuals (ε) for Model 1/Model 2: Contextual Exposures = .81/.81, Conservative Beliefs = .80/.81,
Tolerant Personality = .85/.86, Negative Attitudes Toward Outgroups = .63/.65

The elderly
Indirect Effects Model (Model 1) 17.95 17 .98 .11 –.03 .43** .25* –.43** .18 –.32** –.00 –.28* — —
Direct Effects Model (Model 2) 17.17 15 .98 .11 –.03 .43** .25* –.43** .18 –.31** .06 –.29* .12 –.01
Residuals (ε) for Model 1/Model 2: Contextual Exposures = .82/.81, Conservative Beliefs = .81/.80,
Tolerant Personality = .87/.86, Negative Attitudes Toward Outgroups = .92/.91

NOTE: GFI = Goodness-of-Fit index, RMR = standardized Root Mean Square Residual.
*z > 1.96, p < .05. **z > 2.58, p < .01.

explanation of the pattern of correlations. This conclu- hypothesized indirect effects model in its explanation of
sion is particularly warranted when one considers the the observed pattern of correlations.5
lack of significance obtained for the two additional
Exploring Mediation
direct paths.
We also examined the degree to which the effects of
Test of a revised indirect effects model (Model 3, Paths 2, 4, 5, the two distal factors (Family Status and Contextual
6, 7, and 8). Given the lack of relationship between Fam- Exposures) on Negative Attitudes Toward Outgroups
ily Status and Conservative Beliefs, we also tested a were mediated by the two proximal factors (Conserva-
revised version of our indirect effects model, which pos- tive Beliefs and Tolerant Personality). To that end, we
ited no direct association between Family Status and conducted two separate analyses. First, consistent with
either of the proximal factors. In this revised indirect an approach suggested by Schumacker and Lomax
effects model, Family Status is seen as affecting Beliefs (1996), we computed the total effect size obtained from
and Personality only through its relationship to Contex- Model 1 (which includes no direct effect paths for the
tual Exposures. The revised indirect effects model pro- two distal factors) and compared it with the total effect
vided a marginally satisfactory fit to the data, c2(19) = size obtained from Model 2 (which includes the two
28.93, p < .10, GFI = .97, standardized RMR = .08. All six direct effects paths) using GPOWER (Erdfelder, Faul, &
paths remain significant at the .01 level. We then com- Buchner, 1996; Faul & Erdfelder, 1992). Results indi-
pared the overall fit of Model 3 with Model 1 by comput- cated no difference in effect size between the two models
ing the differences between the chi-squares and degrees (effect size = 1.44 in both models), indicating that the
of freedom associated with each model. The chi-square distal factors do not account for additional variance in
test for significance indicated that Model 3 provided a prejudice beyond that provided by proximal factors. Sec-
marginally worse fit than Model 1, χ2 difference (2) = ond, we conducted stepwise regression analyses using
4.90, p < .10. Therefore, we concluded that the revised the obtained latent factor scores and found that the two
indirect effects model fit the data less well than did the distal factors did not contribute significantly to a model
412 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

that contained the two proximal factors (R 2 = .57 in both model provide a superior fit to the data; for attitude
models). In contrast, the two proximal factors did con- toward homosexuals, χ2 difference (2) = 0.83, ns; for atti-
tribute significantly to a model containing only the two tude toward other races, χ2 difference (2) = 0.17, ns; and
proximal factors. Thus, the two proximal factors can be for attitude toward elderly persons, χ2 difference (2) =
said to meaningfully mediate the association of the two 0.78, ns. Although there were some differences in magni-
distal factors on Negative Attitudes Toward Outgroups. tude or significance for individual path loadings (e.g.,
In addition, we explored the manner in which the two Path 5 was only marginally significant when considered
distal factors exert their indirect effects on Negative Atti- within each of the separate outgroups; Path 7 was not sig-
tudes Toward Outgroups via the two proximal factors. To nificant for attitude toward the elderly), the overall ade-
that end, we calculated the total indirect effects of each quacy and superiority of the indirect effects model was
distal factor on Negative Attitudes Toward Outgroups via not influenced significantly by the particular outgroup
each proximal factor. Family Status was found to have an in question.
indirect effect size of .147 via Tolerant Personality and To more fully examine mediation, we also conducted
.105 via Conservative Beliefs. Contextual Exposures was two additional analyses for each outgroup. First, using
found to have an indirect effect size of .080 via Tolerant the procedure outlined above, we computed the total
Personality and .198 via Conservative Beliefs. Although effect size obtained from Model 1 and compared it with
descriptive in nature, the differences in the relative sizes the total effect size obtained from Model 2 for each out-
of these indirect effects suggest that the two distal factors group. Results indicated no difference in effect size
influence general prejudice via slightly different proxi- between the two models for attitude toward other
mal routes, with Family Status exerting its effect through races/Blacks (effect size = .61 in both models) and small
both Personality and Beliefs and Contextual Exposures differences in effect size for attitude toward the elderly
exerting its effect mainly through Beliefs. (effect size = .09 and .10 for Models 1 and 2, respec-
tively), attitude toward homosexuals (effect size = .56
Testing Single Outgroup Models and .64 for Models 1 and 2, respectively), and attitude
Although we were primarily interested in determin- toward foreigners (effect size = .18 and .23 for Models 1
ing whether there was general support for an indirect and 2, respectively). By convention, the magnitude of
effects versus a direct effects model across outgroups, we such effect size differences is considered to be small
also recognize that it may be of interest to know how the (Lipsey, 1990). Second, we conducted stepwise regres-
sion analyses for each outgroup using the single out-
models fare for each of the separate outgroups. To that
group prejudice measure and the obtained latent factor
end, we conducted ancillary analyses (a) to determine
scores derived from the proximal-distal measures. We
whether there was general support for the indirect
found that the two distal factors did not contribute sig-
effects model over the direct effects model for each of
nificantly to a model that contained the two proximal
the separate outgroups, (b) to examine the path load-
factors for attitude toward other races/Blacks (R2 = .38 in
ings obtained for each outgroup in each model, and (c)
both models), did contribute minimally for attitude
to explore the degree to which the effects of the two dis-
toward the elderly (R2 = .08 for Model 1 and .09 for
tal factors on Negative Attitudes Toward Outgroups were
Model 2), and did contribute significantly for attitude
mediated by the two proximal factors for each outgroup.
toward homosexuals (R2 = .36 for Model 1 and .39 for
Table 2 contains the two-group structural results for
Model 2) and attitude toward foreigners (R 2 = .15 for
both indirect and direct effects models computed sepa-
Model 1 and .19 for Model 2). In contrast, the two proxi-
rately for each of the outgroups (note that two-group
mal factors did contribute significantly to a model con-
analyses were not possible for attitude toward foreigners
taining only the two proximal factors for each outgroup.
because this attitude was measured only in Sample 2). As
Thus, with some exception, the two proximal factors can
can be seen in Table 2, all of the separate indirect effects
be said to largely mediate the association of the two distal
models provided a satisfactory fit to the data; for attitude
factors on prejudice.
toward homosexuals, χ2(17) = 15.93, ns, GFI = .98, stan-
dardized RMR = .12; for attitude toward other races,
χ2(17) = 28.28, ns, GFI = .97, standardized RMR = .16; GENERAL DISCUSSION
and for attitude toward elderly persons, χ2(17) = 17.95, We began this article by asking a deceptively simple
ns, GFI = .98, standardized RMR = .11. Of principal inter- question: “What causes prejudice?” Results from the cur-
est, however, is the comparison of each of these indirect rent investigation lead us to answer “both proximal and
effect models with their respective direct effect counter- distal factors, themselves causally linked.” Such a
parts. In no case did the corresponding direct effects response echoes the sentiment of Allport (1954) and
Agnew et al. / PROXIMAL AND DISTAL INFLUENCES ON PREJUDICE 413

others who have suggested that prejudice involves a mul- rally distal factors such as family education. Viewing tra-
tilevel, multifactor set of causes, ranging from those ditional prejudice in this proximal-distal manner per-
causes that are remote to those that are immediate. mits us to align prejudice research with central tenets in
Unfortunately, these past theoretical pronouncements attitude theory and, ultimately, to explore in finer detail
have not been paralleled by systematic empirical valida- the role that social-cultural, family, and individual vari-
tion. Data from two independent samples provide con- ables play in the formation of current attitudes.
verging evidence of the viability of a general model of Such a model permits one to examine some of the
prejudice, one that incorporates both proximal and dis- complex contributions of distal factors. For example,
tal causes of traditional prejudice. Family Status was not manifested directly in Conserva-
Specifically, the results from structural equation tive Beliefs, as was hypothesized and as might have been
analyses suggest that an indirect effects proximal-distal expected given previous research (e.g., Hyman &
model, in which the impact of distal factors (i.e., Family Wright, 1979). Rather, Family Status exerted its effects
Status and Contextual Exposures) on prejudice is medi- on Conservative Beliefs via Contextual Exposures (Paths
ated by proximal factors (i.e., Conservative Beliefs and 2 and 4). It is possible that the locus of socioeconomic
Tolerant Personality traits), provides a significant and status in the current investigation may be partially
parsimonious fit to the data. The proximal-distal model responsible for the lack of direct effect. Previous find-
explains substantial variance in individuals’ prejudicial ings are based on the examination of a given individual’s
attitudes in absolute terms, and with the exception of the own socioeconomic circumstances and his or her own
hypothesized direct path from Family Status to Conser- conservative beliefs. In contrast, the socioeconomic
vative Beliefs, every path and correlation hypothesized status of our college student participants may be quite
in the model proved to be statistically significant. More- different from that of their parent(s) (note the educa-
over, the addition of direct paths from Family Status and tional attainment means of parents in the appendix).
Contextual Exposures to prejudice failed to improve the The direct association between Family Status and Con-
fit of the original model significantly, and both of these servative Beliefs is diminished as Family Status deviates
direct paths failed to reach statistical significance. increasingly from a participant’s own current socioeco-
Additional mediation analyses suggest that the two nomic circumstances. It should be stressed, however,
distal factors influence general prejudice via slightly dif- that this explanation is speculative.
ferent proximal routes, with Family Status exerting its Can a generalizable model of prejudice (one that
effect through both Personality and Beliefs and Contex- includes multiple outgroup targets) be supported by a
tual Exposures exerting its effect mainly through Beliefs. single causal framework? The results suggest an affirma-
Descriptive in nature, these finding are probably best tive response to this question. In both samples, we found
interpreted at face value. That is, these subtle differ- a significant amount of shared variance among the sin-
ences are best not overemphasized or interpreted as gle outgroup measures of prejudice (see Table 1).
meaning that either Family Status or Contextual Expo- Moreover, the latent prejudice factor was found to be sig-
sures are more or less important in understanding tradi- nificantly associated with both proximal and distal fac-
tional prejudice. Rather, the dual mediating roles of tors, in both bivariate (or bifactorial) correlations and
Conservative Beliefs and Tolerant Personality under- within the structural models (see Table 2). These find-
score the subtle manner in which both Family Status and ing are consistent with past research that showed interre-
Social and Environmental Contextual Exposures are lations among measures of prejudice toward different
manifested in Negative Attitudes Toward Outgroups. We target outgroups (e.g., Adorno et al., 1950; Crandall,
believe these results demonstrate that there is a multi- 1994; Weigel & Howes, 1985).
variate causal nexus underlying prejudice. Whereas vari- Although we found much commonality in overall pat-
ables that are more proximal to the outgroup attitude terns of association, analyses of specific outgroups
should predict better than variables that are further revealed some differences in overall variance accounted
removed (Agnew et al., 1993), these proximal variables for and in the magnitude or significance of individual
are themselves largely derived from (or influenced by) path loadings (see Table 2). The most prominent of
more distal antecedent variables. In this sense, the find- these differences emerged for attitude toward the eld-
ings are consistent with those found for other proximal- erly. The overall amount of variance in this attitude
distal social psychological models, such as the reasoned accounted for by model factors was markedly lower than
action model (Ajzen & Fishbein, 1980). Just as one’s that for the other outgroups. Furthermore, the path
behavior is best predicted by factors that are most proxi- between Conservative Beliefs and negative attitude
mal and specific to the behavior (e.g., behavioral inten- toward the elderly was found to be essentially 0 (see Path
tions), one’s attitude toward outgroups is best predicted 7 in Table 2 under “the elderly”), whereas for the other
by current beliefs and personality traits, not by tempo- outgroups this path was found to be strongly and posi-
414 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

tively significant. This lack of association is consistent cal testing in the past. However, it is also true that the cur-
with recent theorizing concerning the unique nature of rent climate of political correctness on college campuses
the elderly as an outgroup. As Lambert and Chasteen nationwide might be expected to reduce the level of
(1997) note, although it is tempting to view the elderly as expressed negativity toward outgroups (Jones, 1997).
“ ‘just another outgroup,’ this group differs from other Therefore, one might reasonably expect stronger effects
social categories in that for most people, what was once from a noncollege sample than those reported here.
an out-group inevitably becomes an in-group” (p. 470). In addition, it would be preferable to test the model
In their article, Lambert and Chasteen offer a concep- with a more racially and ethnically diverse sample. Such
tual framework that differentiates conservatism from lib- testing might not only demonstrate the generalizability
eralism. Conservatism is held to be associated “with of the overall model but also suggest any important dif-
respect for conventional, old fashioned values,” whereas ferences in the predictive value of specific measures for
liberalism is viewed as associated “with the desire to assist specific groups of respondents. Although it would
disadvantaged groups” (p. 469). Because the elderly are clearly be instructive to have data from additional seg-
seen as both highly disadvantaged and highly conven- ments of the general population, the present study
tional, the authors hypothesized and obtained evidence remains a useful demonstration of our major point: a
that conservatism was negatively correlated with atti- multivariate, proximal-distal framework is capable of
tudes toward Blacks but positively correlated with atti- integrating a number of conceptually distinct social psy-
tudes toward the elderly. Although the proximal-distal chological approaches to prejudice.
model highlights important similarities in predictors of Despite noted limitations, this study has a number of
negativity toward outgroups, the observed differences strengths. Although the bivariate findings within the
for the elderly on this dimension underscore the impor- model are largely consistent with past research, the
tance of considering perceived differences in an out- empirical combination of distal and proximal variable
group’s conventionality versus disadvantaged status. sets tested within a single model employing a latent out-
Because of the cross-sectional nature of the present group factor is unique. In addition, the data collection
data, it is useful to note the logical temporal order of the procedure employed in the CAPS project (Sample 1)
factors contained in the theorized proximal-distal provided desirable temporal separation of the instru-
model. The model begins with distal social factors exis- ments used to measure the model components (e.g., the
tent early in life and proceeds to the current beliefs and measure of ageism was collected during the 3rd week of
personality traits of the individual. It is not plausible, the study, whereas the racism measure was administered
theoretically, for the causal sequence to go from the during the 10th week). This measurement separation
proximal to the distal. An individual who currently pos- reduced the possibility that participants’ data were
sesses conservative beliefs or has a tolerant personality affected by transitory factors (e.g., temperature or
would not be capable of affecting his or her family’s mood).
socioeconomic status when the individual was growing This study represents one step toward the establish-
up. The logic of this temporal ordering strengthens the ment of a broad framework that consolidates existing
validity of the model (see Heise, 1975). It should, of knowledge about traditional prejudice. The proximal-
course, be noted that only longitudinal data can estab- distal framework allows for the simultaneous test of the
lish causality. Furthermore, because we do not directly effects of numerous predictor variables that have
measure distal factors but rather contemporaneous received past support. In so doing, the framework dem-
memories, recollections, and reports of these factors, it onstrates that it facilitates not only the organization of
is possible that such reports may be distorted by current predictive factors but also their meaningful consolida-
beliefs and personality factors. tion. Past research suggests that the factors in the
Limitations regarding the samples used in this study proximal-distal model correlate with traditional preju-
also should be noted. First, it might be argued that col- dice in a bivariate sense. However, knowledge regarding
lege student samples are not representative of the gen- the interconnectivity of these factors adds appreciably to
eral public and may not be an appropriate population our understanding of the mechanism through which
with which to test the model (Sears, 1986). Indeed, vari- particular variable sets exert their effects.
ance on such critical distal variables as education and Traditional prejudice remains a real-world phenome-
age is severely restricted in a college sample. Consider- non with profound implications for victims and perpe-
ing the multiplicity of measures administered, it would trators alike. Although the atrocities of World War II
not be easy to collect the requisite data from a more rep- served as the impetus for contemporary social psycho-
resentative general population survey. This difficulty logical inquiry on prejudice (see Snyder & Ickes, 1985),
might account for why multivariate, integrative models current events such as the rise of neo-Nazi hate groups in
of prejudice have not been subject to widespread empiri- the United States and Europe, the recent ethnic cleans-
Agnew et al. / PROXIMAL AND DISTAL INFLUENCES ON PREJUDICE 415

ing in the former Yugoslavia, and the increasing outcry (Jones, 1997). A genuine understanding of the nature of
against immigrants in the more developed countries of such prejudice requires a thorough exploration of these
the world suggest that the ultimate causes of traditional causal precursors and of the chain of causality underly-
prejudice are just as firmly rooted today as ever before ing outgroup rejection.

APPENDIX
Description of Measured Model Variables, Samples 1 and 2

Variablea Scale Range Mean (SD) Description

Family status
1a. Father’s education 1 (8th grade) to 7 (Ph.D.) 4.77 (1.38)
“What is the highest grade in school completed . . . by your
father or man who raised you?”
1b. Parents’ education 1 (8th grade) to 7 (Ph.D.) 4.38 (1.07) Two items similar to the question above, one for each parent;
responses averaged
2a. Father’s occupational 0 to 100 (e.g., 20 = waiter, 73.58 (17.10) “If you had to rank the occupation that your father had most
status 60 = bookkeeper, 98 = doctor) of the time that you were growing up, where would you put
it on the above scale?”
2b. Parents’ occupational 0 to 99 (e.g., 88 = garbage 64.24 (15.52) Two items similar to the question above, one for each parent;
status collector, 02 = physician) responses averaged and reverse coded.
3a. Social class of family 1 (lower) to 5 (upper) 3.47 (.66) “Generally speaking, what would you say was the social class
of your family while you were growing up?”
3b. Social class of family 3.14 (.86)

Contextual exposures
4a. Religious exposure 1 (at least once a week) to 4 2.13 (1.14) “While in high school, how often would you say you attended
outside the home (almost never) religious services?”
4b. Early outgroup contact 1 (absolutely no contact) to 5 1.80 (.64) “How much contact would you say you had with Black
(very frequent contact) people [old people, foreigners, homosexuals] in the
neighborhood where you grew up?” (separate responses
averaged)
5a. Moral/religious 29 to 71: lower = more moral/ 45.47 (10.66) 9-item standardized scale. Measures emphasize ethical and
emphases in home religious emphases at home religious issues and values in the family (from the Family
Environment Scale) (Moos & Moos, 1981).
5b. Early neighborhood 1 (all White), 2 (mostly White), 3 1.56 (.58) “What was the racial mixture (the ratio of Whites and non-
racial mix (about half White, half non-White), Whites) in the neighborhood where you grew up?”
4 (mostly non-White),
5 (all non-White)
6a. Development of 1 (rural farm), 2 (rural 3.27 (.95) “In which of the following was your home (during high
community nonfarm), 3 (suburb), or school) located?”
4 (city/town)
6b. Development of 1 (rural farm) to 5 (large 3.19 (1.24) “In which of the following was your home located while
community city or town) growing up?”
7a. Size of community 1 (less than 5,000 people) to 3 1.99 (.70) “What would you say was the size of the community in which
(greater than 100,000 people) you lived during high school?”
7b. Size of community 1 (less than 5,000 people) to 5 2.56 (1.13) “What would you say was the size of the community in which
(greater than 1,000,000 people) you lived while you were growing up?”

Conservative beliefs
8a. Orthodox religious beliefs 12 to 108: higher = more 79.60 (26.72) 12-item scale measures belief in traditional religious
religious orthodox beliefs doctrines (e.g., American Protestantism) (Batson & Ventis,
1982)
8b. Orthodox religious beliefs 80.04 (26.84)
9a. General conservatism 13 to 78: higher = more 44.21 (7.12) 13-item scale assesses general conservatism as a separate
conservative factor from liberalism (Kerlinger, 1984)
9b. General conservatism 42.20 (8.78)
10a. Traditionality in gender- 0 to 75: lower = more 16.73 (10.01) 25-item scale measures beliefs about the rights and roles of
role beliefs traditional role beliefs women in major activities that could be the same for men
(reversed in analyses) and women (Spence, Helmreich, & Stapp, 1973)
10b. Traditionality in 25.80 (11.34)
gender-role beliefs

(continued)
416 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

APPENDIX Continued

Variablea Scale Range Mean (SD) Description

Tolerant personality
11a. Openness 0 to 192: higher = more 117.73 (17.39) 48-item subscale from the NEO Personality Inventory
openness (Costa & McCrae, 1985)
11b. Openness 122.37(18.52)
12a. Flexibility 1 to 28: higher = more 14.40 (4.00) 28-item subscale from the California Personality Inventory
flexibility (Gough, 1956)
12b. Flexibility 16.36 (4.87)
13a. Tolerance 1 to 32: higher = more 19.89 (4.50) 32-item subscale from the California Personality Inventory
tolerance (Gough, 1956)
13b. Tolerance 20.75 (4.31)

Negative attitudes toward outgroups


14a. Attitude toward 5-point Likert scales, ranging 43.68 (13.25) 14-item composite measure (α = .95). For example, “Homo-
homosexuals from strongly agree to strongly sexuality is something unhealthy and abnormal that should
disagree. Scores ranged from be cured.”
14 to 70 (higher = more
negative attitude).
14b. Attitude toward 5-point Likert scales, ranging 62.06 (18.48) 21-item Homosexual Attitude Scale (α = .92) (Kite & Deaux,
homosexuals from strongly agree to strongly 1986). For example, “Homosexuals should be forced to have
disagree. Scores ranged from psychological treatment.”
21 to 105 (higher = more
negative attitude).
15a. Attitude toward other 5-point Likert scales, ranging 22.80 (7.13) 9-item composite measure (α = .87). For example, “The
races from strongly agree to strongly races should be considered equal in all matters public and
disagree. Scores ranged from private.”
9 to 45 (higher = more
negative attitude).
15b. Attitude toward Blacks 5-point Likert scales, ranging 15.03 (5.00) 7-item composite measure of “old-fashioned racism” (α =
from strongly agree to strongly .82) (McConahay, 1986). For example, “Black people are
disagree. Scores ranged from generally not as smart as Whites.”
7 to 35 (higher = more
negative attitude).
16a. Attitude toward 4-point Likert scales, ranging 16.22 (4.56) 10-item composite measure (α = .83). For example, “If old
elderly persons from strongly agree to strongly people want to be liked they should get rid of their irritating
disagree. Scores ranged from faults” (adapted from Kogan, 1961).
10 to 40 (higher = more
negative attitude).
16b. Attitude toward 5-point Likert scales, ranging 70.40 (13.70) 29-item Fraboni Scale of Ageism (α = .83) (Fraboni,
elderly people from strongly agree to strongly Saltstone, & Hughes, 1990). For example, “Old people
disagree. Scores ranged from complain more than other people do.”
29 to 145 (higher = more
negative attitude).
17b. Attitude toward foreigners 5-point Likert scales, ranging 16.40 (4.89) 7-item composite measure adopted from McConahay’s
from strongly agree to strongly (1986) “old fashioned racism” items (a = .83). For example,
disagree. Scores ranged from “It is a bad idea for Americans and foreigners to marry one
7 to 35 (higher = more another.”
negative attitude).

a. Sample 1 variables are marked “a” (e.g., 1a, 2a, etc.) and Sample 2 variables are marked “b” (e.g., 1b, 2b, etc.).

NOTES als. In addition, no attempt was made in the earlier article to examine
the interrelationships between variables’ classes or to examine the
1. In Agnew, Thompson, Smith, Gramzow, and Currey (1993), per- manner in which variables’ classes might mediate the effect of other
sonality variables were classified as “distal individual factors,” whereas classes on prejudice. In examining such interrelationship and media-
in the current article, they are considered to be proximal. The differ- tional issues, we see personality variables to be more proximal in the
ence in classification is necessitated by the different goals of these arti- chain of causality hypothesized to underlie prejudice toward out-
cles. The Agnew et al. (1993) article examined how classes of individual groups.
and family factors independently predict prejudice toward a specific 2. Note that we are not suggesting that these are the only sets that
outgroup (i.e., homosexuals). Accordingly, personality constructs could be conceptualized to underlie generalized prejudice; rather,
were considered a more distal individual factor than either personal these factors represent clear examples of past single-factor distal and
beliefs or relevant attitudes and experiences with respect to homosexu- proximal approaches to prejudice.
Agnew et al. / PROXIMAL AND DISTAL INFLUENCES ON PREJUDICE 417

3. Of course, a number of conditions must be met for the “contact Faul, F., & Erdfelder, E. (1992). GPOWER: A-priori, post-hoc, and compro-
hypothesis” to be supported. See Duckitt (1992b) for a review of these mise power analysis for MS-DOS [Computer program]. Bonn, Ger-
conditions. many: Bonn University, Department of Psychology.
4. Data were merged only after preliminary analyses indicated no Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior:
significant differences on study measures between participants from An introduction to theory and research. Reading, MA: Addison-Wesley.
the three Computer Assisted Panel Study (CAPS) cohorts. Fraboni, M., Saltstone, R., & Hughes, S. (1990). The Fraboni Scale of
5. Although not of central interest, we also tested three equivalent Ageism (FSA): An attempt at a more precise measure of ageism.
alternatives to Model 1 (i.e., alternative models in which the degrees of Canadian Journal on Aging, 9, 56-66.
freedom were held constant while certain paths and/or correlations Gaines, S. O., Jr., & Reed, E. S. (1995). Prejudice: From Allport to
were altered) (see MacCallum, Wegener, Uchino, & Fabrigar, 1993), DuBois. American Psychologist, 50, 96-112.
each of which focused on the link between Conservative Beliefs and Gough, H. G. (1956). The California Personality Inventory. Palo Alto, CA:
Tolerant Personality. Model 4 tested the possibility that Conservative Consulting Psychologists Press, Inc.
Beliefs cause Tolerant Personality traits, Model 5 tested the hypothesis Hamilton, D. L., & Trolier, T. K. (1986). Stereotypes and stereotyping:
that Tolerant Personality traits cause Conservative Beliefs, and Model 6 An overview of the cognitive approach. In J. F. Dovidio & S. L.
tested the supposition that Conservative Beliefs and Tolerant Personal- Gaertner (Eds.), Prejudice, discrimination, and racism (pp. 127-158).
ity traits are reciprocally causal. All three of these equivalent models Orlando, FL: Academic Press.
Harris, D. B., Gough, H. G., & Martin, W. (1950). Children’s ethnic atti-
provided a fit to the data that was equivalent to that of Model 1, χ (17) =
2
tudes: Relationship to parental beliefs concerning child training.
24.03, ns, GFI = .98. Therefore, the hypothesized correlation between Child Development, 21, 169-181.
Conservative Beliefs and Tolerant Personality appears to be repre- Heise, D. R. (1975). Causal analysis. New York: John Wiley.
sented just as well by a causal path from Conservative Beliefs to Toler- Hyman, H. H., & Wright, C. R. (1979). Education’s lasting influence on
ant Personality, or vice versa. values. Chicago: University of Chicago Press.
Insko, C. A., & Schopler, J. (1967). Triadic consistency: A statement of
affective-cognitive-conative consistency. Psychological Review, 74,
REFERENCES 361-376.
James, L. R., Muliak, S. A., & Brett, J. M. (1982). Causal analysis: Assump-
Abrahamson, M., & Carter, V. J. (1986). Tolerance, urbanism and tions, models, and data. Beverly Hills, CA: Sage.
region. American Sociological Review, 51, 287-294. Jones, J. M. (1997). Prejudice and racism (2nd ed.). New York: McGraw-
Adorno, T. W., Frenkel-Brunswik, E., Levinson, D. J., & Sanford, R. N. Hill.
(1950). The authoritarian personality. New York: Harper. Jöreskog, K. G., & Sörbom, D. (1989). LISREL 7: A guide to the program
Agnew, C. R., Thompson, V. D., Smith, V. A., Gramzow, R. H., & Currey, and applications. Chicago: SPSS, Inc.
D. P. (1993). Proximal and distal predictors of homophobia: Fram- Jöreskog, K. G., & Sörbom, D. (1993). LISREL 8: User’s reference guide.
ing the multivariate roots of outgroup rejection. Journal of Applied Chicago: Scientific Software International, Inc.
Social Psychology, 23, 2013-2042. Kerlinger, F. N. (1984). Liberalism and conservatism: The nature and struc-
Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting ture of social attitudes. Hillsdale, NJ: Lawrence Erlbaum.
social behavior. Englewood Cliffs, NJ: Prentice Hall. Kite, M. E., & Deaux, K. (1986). Attitude toward homosexuality: Assess-
Allport, G. W. (1954). The nature of prejudice. Reading, MA: Addison- ment and behavioral consequences. Basic and Applied Social Psychol-
Wesley. ogy, 7, 137-162.
Anderson, J. C., & Gerbing, D. W. (1988). Structural equation model- Kogan, N. (1961). Attitudes toward old people: The development of a
ing in practice: A review and recommended two step approach. Psy- scale and an examination of correlates. Journal of Abnormal and
chological Bulletin, 103, 411-423. Social Psychology, 62, 44-54.
Ashmore, R. D. (1970). The problem of intergroup prejudice. In B. E. Lambert, A. J., & Chasteen, A. L. (1997). Perceptions of disadvantage
Collins (Ed.), Social psychology (pp. 245-296). Reading, MA: versus conventiality: Political values and attitudes toward the eld-
Addison-Wesley. erly versus Blacks. Personality and Social Psychology Bulletin, 23,
Batson, C. D., & Ventis, W. L. (1982). The religious experience: A social psy- 469-481.
chological perspective. New York: Oxford University Press. Latané, B. (1989). Social psychology and how to revitalize it. In M. R.
Bernstein, I. H., Garbin, C. P., & Teng, G. K. (1988). Applied multivariate Leary (Ed.), The state of social psychology: Issues, themes, and controver-
analysis. New York: Springer-Verlag. sies. Newbury Park, CA: Sage.
Bierly, M. M. (1985). Prejudice toward contemporary outgroups as a Lipsey, M. W. (1990). Design sensitivity: Statistical power for experimental
generalized attitude. Journal of Applied Social Psychology, 15, 189-199. research. Newbury Park, CA: Sage.
Braithwaite, V. A., & Scott, W. A. (1991). Values. In J. P. Robinson, P. R. Loehlin, J. C. (1992). Latent variable models: An introduction to factor, path,
Shaver, & L. S. Wrightsman (Eds.), Measures of personality and social and structural equations (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum.
psychological attitudes (pp. 661-753). San Diego: Academic Press. MacCallum, R. C., Wegener, D. T., Uchino, B. N., & Fabrigar, L. R.
Butler, R. N. (1969). Ageism: Another form of bigotry. The Gerontologist, (1993). The problem of equivalent models in applications of
9, 243-246. covariance structure models. Psychological Bulletin, 114, 185-199.
Costa, P. T., Jr., & McCrae, R. R. (1985). The NEO Personality Inventory McConahay, J. B. (1986). Modern racism, ambivalence, and the mod-
manual. Odessa, FL: Psychological Assessment Resources, Inc. ern racism scale. In J. F. Dovidio & S. L. Gaertner (Eds.), Prejudice,
Crandall, C. S. (1994). Prejudice against fat people: Ideology and self- discrimination and racism (pp. 91-126). San Diego, CA: Academic
interest. Journal of Personality and Social Psychology, 66, 882-894. Press.
Dale, A., Arber, S., & Procter, M. (1988). Doing secondary analysis. Lon- McGuire, W. J. (1985). Attitudes and attitude change. In G. Lindzey &
don: Unwin Hyman. E. Aronson (Eds.), The handbook of social psychology (3rd ed., Vol. 2,
Dovidio, J. F., & Gaertner, S. L. (1991). Changes in the expression and pp. 233-346). New York: Random House.
assessment of racial prejudice. In H. J. Knopke, R. J. Norrell, & R. W. Moos, R., & Moos, B. (1981). Family Environment Scale manual. Palo
Rogers (Eds.), Opening doors: Perspectives on race relations in contempo- Alto, CA: Consulting Psychologists Press.
rary America (pp. 119-148). Tuscaloosa: University of Alabama Nunnally, J. C. (1967). Psychometric theory. New York: McGraw-Hill.
Press. Pettigrew, T. F. (1958). Personality and sociocultural factors and inter-
Duckitt, J. (1992a). Psychology and prejudice: A historical analysis and group attitudes: A cross-national comparison. Journal of Conflict
integrative framework. American Psychologist, 47, 1182-1193. Resolution, 2, 29-42.
Duckitt, J. (1992b). The social psychology of prejudice. New York: Praeger. Ray, J. J., & Lovejoy, F. H. (1986). The generality of racial prejudice.
Erdfelder, E., Faul, F., & Buchner, A. (1996). GPOWER: A general Journal of Social Psychology, 126, 563-564.
power analysis program. Behavior Research Methods, Instruments, & Schumacker, R. E., & Lomax, R. G. (1996). A beginner’s guide to structural
Computers, 28, 1-11. equation modeling. Mahweh, NJ: Lawrence Erlbaum.
418 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN

Sears, D. O. (1986). College sophomores in the laboratory: Influences Stroebe (Eds.), Stereotyping and prejudice: Changing conceptions (pp. 3-
of a narrow database on social psychology’s view of human nature. 34). New York: Springer Verlag.
Journal of Personality and Social Psychology, 51, 515-530. Stryker, S., & Statham, A. (1985). Symbolic interaction and role theory.
Simonton, D. K. (1990). Personality and politics. In L. A. Pervin (Ed.), In G. Lindzey & E. Aronson (Eds.), The handbook of social psychology
The handbook of personality theory and research (pp. 670-692). New (3rd ed., Vol. 1, pp. 311-378). New York: Random House.
York: Guilford. Tajfel, H., & Turner, J. C. (1985). The social identity theory of inter-
Simpson, G. E., & Yinger, J. M. (1985). Racial and cultural minorities: group behavior. In S. Worchel & W. G. Austin (Eds.), Psychology of
An analysis of prejudice and discrimination (5th ed.). New York: intergroup relations (pp. 7-24). Chicago: Nelson-Hall.
Plenum. Weigel, R. H., & Howes, P. W. (1985). Conceptions of racial prejudice:
Snyder, M., & Ickes, W. (1985). Personality and social behavior. In G. Symbolic racism reconsidered. Journal of Social Issues, 41, 117-138.
Lindzey & E. Aronson (Eds.), The handbook of social psychology (3rd Wilson, T. C. (1985). Urbanism and tolerance: A test of some hypothe-
ed., Vol. 2, pp. 883-947). New York: Random House. ses drawn from Wirth and Stouffer. American Sociological Review, 50,
Spence, J. T., Helmreich, R., & Stapp, J. (1973). A short version of the 117-123.
Attitudes Toward Women Scale (AWS). Bulletin of the Psychonomic Wirth, L. (1938). Urbanism as a way of life. American Journal of Sociology,
Society, 2, 219-220. 44, 1-24.
Stangor, C., Sullivan, L. A., & Ford, T. E. (1991). Affective and cognitive
determinants of prejudice. Social Cognition, 9, 359-380. Received December 23, 1997
Stroebe, W., & Insko, C. A. (1989). Stereotyping, prejudice, and dis- Revision accepted December 18, 1998
crimination. In D. Bar-Tal, C. F. Grauman, A. W. Kruglanski, & W.

You might also like