You are on page 1of 43

Yeim Kaya Marmara University Research Metodology Course 15 June 2010

Introduction Model of Interaction Effect Differences Between Mediator and Moderator Effect Categorical Moderator Variable Continous Moderator Variable Importance and Scope of Interaction Moderating Effect Types of Moderating Effect

Different Moderation Patterns Testing the Significance of the Moderator Effect. Statistical Tests of Moderating Effects Examples of Interaction Moderating Effect from literature Advantages of including an interaction that is relevant to a model Conclusion References

Many interesting findings in the social sciences involve interaction or moderator effects . Two independent variables have an interaction effect on a dependent variable if the relationship of an independent variable with the dependent variable changes across values of the other independent variable (Hargens, 2006). For example, in an early study of radio listenership, Lazersfeld found that age was positively related to listening to classical music programs among the highly educated, but negatively related to it among the less educated (Zeisel 1968, pp.123-25).

A moderator variable modifies the form or stregth of the relation between an independent and dependent variable. Moderator effects are also called interactions to signify that the third variable interacts with the relation between two other variable.

X
Z

The interest in moderated relationships has increased dramatically over the past 20 years,social sciences have noted the existence of moderator variables for almost half a century . (Aguinis, 2004) The label moderator variable seems to have been used first by Saunders in 1955,but according to Zedeck (1971),the concept had been discussed previously. For example,Court (1930) used joint causation. Other tems that were used Population control variable (Gaylord & Carroll, 1948); Subgrouping variable (Frederiksen & Melville, 1954); Predictability variable (Ghiselli, 1956); Referent variable (Toops, 1959); Modifier variable (Grooms & Endler, 1960); Homologizer variable (Johnson, 1966).

The interaction effect model is shown below; Y=i1+c1X+c2Z+c3XZ+e1 where Y is the dependent variable,X is the independent varible,Z is the moderator variable and XZ is the interaction of the moderator. The independent variable which is called product variable; e1 is a residual and c1 represent the relation between the dependent variable and the independent varible,c2 represent the relation between moderator variable and the dependent variable,c3 represent the relation between moderator by independent variable interaction,respectively. The interaction variable XZ is formed by the product of X and Z. Often X and Z are centered.

Independent variable (X) (Predictor variable)

Dependent variable (Y) (Outcome variable)

Moderator variable (Z)

Sometimes it is difficult to distinguish mediators and moderators when forming hypotheses about variables . The definitional difference, that a mediator is predicted by the independent variable and a moderator is a separate independent variable, is important but not always obvious.

Mediating Effect

A mediator is the mechanism through which a predictor influences an outcome variable X causes M and M then causes Y Mediating effects explain how or why such effects occur (Baron and Kenney 1986) One typically looks for mediators if there already is a strong relation between a predictor and an outcome and one wishes to explore the mechanisms behind that relation.

Predictor Variable (X)


Moderating Predictor Variable (X) (Y)

Mediator Variable (M)

Outcome Variable (Y)

Effect;
Outcome Variable

Moderator Variable (Z)

Moderators are often are introduced when they are unexpectedly weak or inconsistent relations between a predictor (independent) and an outcome (dependent) across studies. Decisions about potential moderators and mediators should be based on previous research and theory and are best made a priori in the design stage rather than post hoc. Also,a given variable may function as either a moderator or a mediator, depending on the theory being tested.

One also can examine mediators and moderators within the same model. Moderated mediation refers to instances in which the mediated relation varies across levels of a moderator. Mediated moderation refers to instances in which a mediator variable explains the relation between an interaction term in a moderator model and an outcome. These more complex models have been described in more detail elsewhere (e.g., Baron & Kenny, 1986;Hoyle & Robinson, in press; James & Brett, 1984; Wegener & Fabrigar, 2000).

Moderators can be:


Continuous (e.g., age,rumination) or Categorical (e.g., gender, country, etc.)

Well consider the categorical MV case first. If either the predictor or moderator variable is categorical, we need to dummy code the moderator and the predictor variable. Jose (2008) tested that the relationship between anxiety and depression is the same or different between males and females. The gender as a moderator must be coded as males=0, females=1.

Moderation Model

Anxiety

Depression

Gender

Hierarchical regression with three steps: 1. Anxiety; 2. Gender (0 = males; 1 = females); and 3. Anxiety X Gender (product term: just multiply these two variables in SPSS, but note that the IV must be centered)

(Note that gender is dummy coded (not 1 = males; 2 = females), and NOT centered.)

What are we looking for? 1. Does anxiety predict depression? 2. Does gender predict depression, i.e., is there a gender difference in depression? 3. Does the product term predict depression? If so, then one has obtained a significant moderation effect.

Coefficientsa Unstandardized Coefficients B Std. Error 9.530 .279 .259 .024 9.055 .492 .254 .024 .708 .605 8.700 .508 .141 .049 1.005 .612 .148 .056 Standardized Coefficients Beta .430 .421 .047 .233 .066 .212

Model 1 2

(Constant) anxiety t1 centered (Constant) anxiety t1 centered Are you male or female? (Constant) anxiety t1 centered Are you male or female? anxiety X gender

t 34.149 10.958 18.397 10.536 1.170 17.136 2.854 1.641 2.632

Sig. .000 .000 .000 .000 .242 .000 .004 .101 .009

a. Dependent Variable: dept1

Two significant results: one main effect and the interaction. Can interpret the main effect from its beta, but not the interaction.

We found that anxiety significantly and positively predicted depression, b = .43, R2 = .19, p < .001. No main effect for gender, b = .05, R2 = .00, p = .24. Depression was not higher among female than male adolescents (after anxiety was entered). The most important finding is the third term, which was a significant predictor, b = .21, R2 = .01, p < .01. It is important to find that this is a sig. predictor above and beyond the two main effects. Cannot enter this in isolation. What the significant interaction term tells us is that the association between anxiety and depression is significantly different between the two groups.

There are many situations in which the moderator variable is continuous. Aiken and West stated that one should center ones main effects before computing the interaction term because of multicollinearity (high correlations among the predictor variables). Centering reduces problems associated with multicollinearity (i.e., high correlations) among the variables in the regression equation (for further explanation, see Cohen et al., 2003; Cronbach, 1987; Jaccard et al., 1990; West et al., 1996).

Lets consider the case of a continuous moderator. In the same dataset, Jose (2008) investigated how rumination moderates the stress to depression relationship. (Note that rumination have examined as both a moderator and a mediator.). Obtain the means for your two IVs: stress and rumination. Remove the means from the variables to create new centered variables. Multiply Stressc X Ruminc to obtain your new interaction term. Enter these variables in the hierarchical regression. Obtain the results on the following page.

Coefficientsa Unstandardized Coefficients B Std. Error 5.409 .251 1.454 .213 5.407 .240 .932 .236 .069 .016 5.067 .264 .679 .248 .072 .016 .033 .011 Standardized Coefficients Beta .449 .288 .317 .210 .331 .190

Model 1 2

(Constant) stressc (Constant) stressc ruminc (Constant) stressc ruminc strxrum

t 21.552 6.824 22.569 3.950 4.348 19.220 2.735 4.606 2.845

Sig. .000 .000 .000 .000 .000 .000 .007 .000 .005

a. Dependent Variable: depression total

Stress and rumination both worsen depressive symptoms by themselves, and we also obtained a significant interaction. Enter the following numbers into ModGraph under continuous moderator

Simple Slope T-Value High Medium Low .10 .68 1.25 2.73 2.18 0.16

Significance Level (p) .007 .03 .87

Notice that the Low group is practically flat (nonsignificant).

Going beyond main effects We typically say it depends More complex models There are many instances in which researchers are interested in whether relations between predictor and outcome variables are stronger for some people than for others Identification of important moderators of relations between predictors and out comes is at the heart of theory in social science. (Cohen,2003).

If we want to know how well we are doing in the biological, psychological, and social sciences, an index that will serve us well is how far we have advanced in our understanding of the moderator variables of our field (Hall & Rosenthal, 1991, p. 447) There are also a lot of application areas such as economy,marketing,business,medicine and education. Additional research areas: training, turnover, performance appraisal, return on investment, mentoring, self-efficacy, job satisfaction, organizational commitment, and career development, among others

According to Sharma,Durand and Gur-Arie (1981) there are three types of moderator effects: homologizer moderator quasi moderator pure moderator.

Homologizer Moderator The true relation between the independent variable and the dependent variable does not change across levels of the moderator,but the error variance does change across the levels of moderator. If an effect is examined by subgroups,the strength of the standardized relation varies because the error variance varies across subgroups. The error variance may vary across subgroups because of different measurement proporties such as response reliabilities across the subgroups. Influences the strength of the link between the independent variable and dependent variable because of differences in error variance across groups.

The second and third forms of a moderator variable are consistent with most discussions of moderators. Quasi-moderator If the moderator variable is also significant predictor of the dependent. Pure moderator If the moderator variable is not a significant predictor of the dependent variable is also called a psychometric moderator because the form of the relation between the independent variable and the dependent variable changes as a function of the moderator. (p.275)

Cohen et al. (2003, pp. 285286) described three patterns of interactions among two continuous variables: enhancing interactions (in which both the predictor and moderator affect the outcome variable in the same direction and together have a stronger than additive effect), buffering interactions (in which the moderator variable weakens effect of the predictor variable on the outcome) antagonistic interactions (in which the predictor and moderator have the same effect on the outcome but the interaction is in the opposite direction.). (as cited in Frazier , Tix , Barron, 2004).

To test H0: 22- 12 = 0,and F statistic (distributed with k2-k1 and N-k2-1 degrees of freedom) is computed using the following formula: = a + b1 X + b2 Z ; R12 = a + b1 X + b2 Z + b3 XZ; R22

F= (R22-R12)/(k2-k1) (1-R22)/(N-k2-1) If the test is significant you can conclude that there is an interaction. (Aguinis ,2004)

Testing significance of a moderator effect,if the predictor or moderator variables are categorical or continous,researcher can use multiple regression. Moderated Multiple Regression is the preferred statistical method for identifying moderator effects (interaction effects) when the predictor and the moderator are continuous variables or when the predictor is continuous and the moderator is categorical. ANOVA can also be used for identifying interactions, but is more appropriately used for the analysis of planned experiments than for observational and survey data. (Jennifer R. Villa*, Jon P. Howell, Peter W. Dorfman, David L. Daniel, 2002).

Michon,Chebat and Turley (2005) investigate the moderating effects of ambient odors on shoppers emotions, perceptions of the retail environment, and perceptions of product quality under various levels of retail density. The results shows that ambient odors positively influence shoppers perceptions only under the medium retail density condition. Shin and Park investigates the moderating effect of group cohesiveness both at the individual and at the group level. In this research collective competency influenced positively to team performance at high level of group cohesiveness, but it influenced negatively at low level of group cohesiveness.

Baeza,Lao,Menese and Roma (2009) appraised the role of leader charisma in fostering positive affective team climate and preventing negative affective climate. The analysis of a longitudinal database of 137 bank branches by means of hierarchical moderated regression shows that leader charisma has a stronger effect on team optimism than on team tension. In addition, the leaders influence and the frequency of leader-team interaction moderate the relationship between charisma and affective climate. Results show that to better understand the impact of leaders charisma on team affective climate it is necessary to differentiate between positive and negative affect.

Ismail,Rahma,Wan Ismail (2007) examines the moderating role of procedural justice in the relationship between participation in pay systems and personal outcomes. The results of tests moderating model using hierarchical regression analyses showed that the inclusion of procedural justice into analysis had increased the effect of participation in pay systems on job commitment, but procedural justice had not increased the effect of participation in pay systems on job satisfaction. This result shows that procedural justice does act as a partial moderator in the pay system models of the hotel industry sector.

if an interaction does in fact exist and is not included in the estimation, this introduces a specification error in the form of omitted variable bias. Estimation of a model that fails to account for the interaction will not provide an accurate estimation of the true relationship between the dependent and independent variables. A model that includes the interaction term provides a better description of the relationship between the independent and dependent variables.

the inclusion of the product term will offer a more accurate estimation of the relationship and explain more of the variation in the dependent variable. including a product term according to Friedrich (1982) is a "low-risk strategy" in that if the product term is significant then keep it in the model otherwise one can drop the product term out of the model.

In summary, moderators are variables that affect the association between an independent variable and an outcome variable. If a researcher fails to consider the possibility of a mediator or moderator effect in the data, a more exact explanation for an outcome may be missed. we defined the concept of moderator variables ,different types of moderator variable. Then we described difference between mediator and moderator variables and we discussed the importance and advantages of moderators. We gave some examples from literature. The goal of this paper is that using these tools will allow researchers to make more informed decisions regarding the operation of moderating effects.

Aguinis H.(2004). Regression Analysis for Categorical Moderators. Retrieved from internet on April 24,2010-04-29 http://books.google.com/books?hl=tr&lr=&id=6sdRuhBTOLQC&oi=fnd&p g=PA1&dq=Regression+Analysis+for+Categorical+Moderators&ots=1bhc kcefI2&sig=BG--TiyVhIe_7Px99_qiIuKYCts#v=onepage&q&f=false

Baeza* A. H., Lao A. C., Meneses and V. G., Rom* I. G. (2009). Leader charisma and affective team climate: The moderating role of the leaders influence and interaction. Universitat Valncia * e Idocal. Bennet J.A.(2000). Focus on Research Methods: Mediator and Moderator Variables in Nursing Research: Conceptual and Statistical Differences. School of Nursing, San Diego State University. Frazier A., Tix P., Barron E.(2004). Testing Moderator and Mediator Effects in Counseling Psychology Research. Journal of Counseling Psychology.

Jose P. (2008). Statistical Moderation. Victoria University of Wellington. SASP Conference Hargens L. (2006). Interpreting Product-Variable Models of Interaction Effects. Center for Statistics and the Social Sciences, University of Washington. Michon R., Chebat J.C., Turley L. W.(2005). Mall atmospherics: the interaction effects of the mall environment on shopping behavior. Journal of Business Research. Ismail A., Rahma H.A., Wan Ismail W. K. (2007). Moderating effect of procedural justice in the relationship between participation in pay systems and personal outcomes. Jurnal Kemanusiaan bil.9, Jun 2007.

MacKinnon D.P.( 2008). Introduction to Statistical Mediation Analysis. Retrieved from internet on April 25,2010 http://www.google.com/books?hl=tr&lr=&id=qg0Eiz1ZmagC&oi=fnd&pg =PR7&dq=interaction+moderating+effect+book+chapter+pdf&ots=hQPj KB_kQY&sig=5ZX8nXux7Uj8-Lyk_bxiwyE9ACc#v=onepage&q&f=false

Shin S. Y., Park W. Moderating Effects of Group Cohesiveness in Competency-Performance Relationships: A Multi-Level Study. Journal of Behavioral Studies in Business
Villa J. R., Howell J. P., Dorfman P. W., Daniel D. L. ( 2003). Problems with detecting moderators in leadership research using moderated multiple regression. The Leadership Quarterly 14 (2003) 323. Zeisel, Hans. (1968) Say it with Figures (Fifth Edition). New York: Harper and Row.

You might also like