of Pe~nality and Social Psychology Copyright 1986 by the American Psyc - PDF document

of Pe~nality and Social Psychology Copyright 1986 by the American Psychological Association, Inc. 1986, Vol. 51, No. 6, 1173-1182 0022-3514/86/$00.75 Moderator-Mediator Variable Distinction in Social Psychological Research: Conceptual, Strategic, and Statistical Considerations M. Baron and David A. Kenny University of Connecticut In this article, we attempt to distinguish between the properties of moderator and mediator variables at a number of levels. First, we seek to make theorists and researchers aware of the importance of not using the terms and mediator by carefully elaborating, both concep- tually and strategically, the many ways in which moderators and mediators differ. We The purpose of this analysis is to distinguish between the properties of moderator and mediator variables in such a way as to clarify the different ways in which conceptual variables may This research was supported in part by National Science Foundation Grant BNS-8210137 and National Institute of Mental Health Grant R01 MH-40295-01 to the second author. cal function of this article is to clarify for experimental re- searchers the importance of respecting these distinctions. This is not, 1173 REUBEN M. BARON at the same time (e.g., sex, the zero-order two other Stem, McCants, the correlation had reduced the direction an interaction conditions for 1972). A would be Toward Establishing Testing Moderator Effects only when the Moderator model. the interaction The moderator and the interaction term. the same Choosing an Appropriate Testing Moderation we consider cedures for appropriately a conceptual a moderator from a strong as opposed finding a crossover interaction. although crossover interactions are stronger as they are effect shifts a 2  2 the simple deficiencies. First, has equal variance at the moderator. For be the for the the moderator, tween the correlations are influenced in the variances in the the two in the variable across then need for the Three different ways moderator changes the independent variable dependent variable: linear (top), qua- dratic (middle), step (bottom). fear-arousing attitude- for low-IQ in which gradual, steady change in the in the the rational in Figure the rational at the where the to occur are usually moderator and gression equasion, Cohen and Kessler (1982). independent variable dependent variable Y is regressed significant effect while X are con- simple effects independent variable different levels West (1986). (Measurement error in independent variable Case 2.) can be Figure 2, independent variable moderator. Alternatively, hierarchical regression Y is regressed X, Z, is given of XZ and the variable relation a step one can step takes place. Case 2. variable is a regression independent variable variable (Y) varies linearly (Z), the Case 3 should squared must be interpreting these regres- Case 4 greatly complicates analysis. Busemeyer multiplicative interactions interactive effects. variables, resulting in proper interactive effects. However, these variable is Although the systematic search variables is rel- atively recent, psychologists have long tween stimulus response, is generic for- Theorists as diverse Hull, Tolman, vene between general, a given variable relation between criterion. Mediators explain physical events internal psychological significance. variables specify effects will effects occur. example, choice this effect is a dissonance To clarify the meaning mediation, we depicting a causal in mediation assumes a three-variable independent variable is also a path from A variable ing conditions: (a) variations able significantly a), (b) variations in the in the variable (i.e., (c) when controlled, a previously significant relation between variables is longer significant, tion occurring c is zero. envisage a When Path c is zero, we have a single, c is zero, this indicates the operation factors. Because including social, have multiple causes, a realistic goal dependent variables altogether. retical perspective, a significant is indeed a necessary a sufficient condition for an by Judd models should for hierarchical the mediational third, the in the in multicollinearity when the in the the co- As in the depen- Vb2sa 2 be used for the two error in internal, psychological in the la). Additionally, in the in the enhanced to the extent approach to quires two are the directly tested feedback are the second feedback in presumed to only the are estimated in James be used in the demonstrate mediation REUBEN M. BARON that transform or input in some regard the constructs the groupthink, and causal modeling. in the mechanism than in the (e.g., a holds in variable as the other Moderator to mediator. certain instructional for the instructional in anxiety be postulated mediate the has been race would over social one should the self-monitoring Mediator to moderator. in the be needed at times There are the hy- which the an independent reasons, as to methodological as in issue we to increase the data. and moderation. C  S P X S C x P C X P X S and that 2 X 2 control may These two a 2  2 claim that the relation, indicated that This equation been added. to which in CS they represent moderated mediation and linking REUBEN M. BARON This practice only the other the mediational trol for only when stronger evidence the crucial link and Metcalf is, a in the control-mediation model the claim and Ajzen's one can and Ajzen's behavior only and Ajzen's such an and treated at the level. Specifically, best used the hypothesized one would although Fishbein's intentions are effect, a better for in regard Global Dispositions to Behavior: Attitudes and areas in perhaps the the prediction (Scheier, 1980), variables. Moreover, and Sherman and conceivable processes the kind providing some the same MODERATOR-MEDIATOR DISTINCTION 1 role played 1984). Specifically, merely involves For example, consistent with their us to then went gard to distinction both terms of & West, analysis models containing interactions power polynomials. published manuscript, Arizona State University, Tempe. Ajzen, I., & Fishbein, Understanding attitudes social behavior. N J: Prentice-Hall. Baron, R. M., & Rodin, J. (1978). Personal control as a mediator Valins (Eds.), Advances in envi- (pp. 145-190). Hillsdale, N J: Erlbaum. M. (1982). A structural equations program. Unpublished manuscript, University M., & Speckart, (1979). Models attitude-behavior re- Psychological Review, 86, Brehm, J. & Cohen, R. (1962). cognitive disso- beyond the H. Gruber, Terrell, & M. Wertheimer (Eds.), Contemporary approaches (pp. 258-290). Cambridge, MA: Harvard University J. R., & Jones, L. (1983). Analysis multiplicative com- rules when the causal variables are measured with error. chological Bulletin, Campbell, D. & Fiske, D. (1959). Convergent Psychological Bul- letin, 56, D., & Social Behavior, 23, Cohen, J., & Cohen, ed.). Hillsdale, N J: Erlbaum. & Hoyt, M. G. (1972). Personal responsibility for conse- the forced compliance liter- D. (1975). structural equation models. (1983). Aggregation basic issues Personality, 51, Findley, M. J., & H. M. (1983). Locus achievement: A literature Finney, J. Mitchell, R. Cronkite, R., & R. H. (1984). amples from the coping/social support stress field. Paper presented the meeting the Psychonomic Fishbein, M., & Ajzen, I. (1975). attitude, intention Kenny, D. A., & (1982). Structural models for the mediation salience effects Freedman, J. San Francisco: Free- Glass, D., & J. (1972). Urban stress: on noise social stressors. New York: G., Latan6, & Williams, K. (1980). Social loafing: Allo- cating effort R., & Brett, J. M. (1984). Mediators, moderators, for mediation. Psychology, 69, R., & Singh, K. (1978). logic, assump- basic analytic procedures least squares. chological Bulletin, K. J., & S6rbom, D. (1984). structural equations likelihood methods (3rd ed.). Mooresville, IN: Scientific M., & Kenny, D. New York: Cambridge University M., & Kenny, D. A. (1981 Process analysis: Estimating me- evaluation research. Evaluation Research, Kenny, D. & Judd, M. (1984). Estimating the Psychological Bulletin, J., & Saegert, (1977). Crowding cognitive control. McNemar, Q. (1969). Psychological statistics (4th ed.). Defensible space. New York: Rodin, J., Solomon, & Metcalf, J. (1978). Role E (1980). private self-consciousness the public expressions personal beliefs. J., & Fazio, R. H. (1983). Parallels between attitudes traits as predictors Personality, 51, R. (1982). Beliefs, attributions, evaluations: Nonhierar- chical models social cognition. situations: Implica- tions for understanding the links between personality social be- Personality, 51, REUBEN M. BARON & Ickes, (1985). Personality social behavior. Lindzey & Aronson (Eds.), social psychology ed., pp. 883-948). Reading, Sobel, M. (1982). Asymptotic confidence intervals structural equations models. Leinhart (Ed.), methodology 1982 (pp. 290-312). San Francisco: Sommer, R. (1969). Personal space. Englewood Cliffs, (1976). Situational factors affecting predictor components Fishbein model. tal Social Stern, G. R., & Pettine, P. W. relative contri- uncontrollable life events Stokols, D. (1976). J. (1971). Statistical principles in experimental design York: McGraw-Hill. Woodworth, R. C. Murchison (Ed.), MA: Clark University J. (1981). cohesion source on process loss in group performance. Unpublished doctoral dissertation, University Connecticut, Storrs. typed on in the prohibits an submitting the manuscripts submitted editorial review for at years after the date For further in the citation does be required standards in the the APA Ethical Principles the APA 1200 17th when submitting the manuscript. typed on should be in quadruplicate, should be Dittoed and Section editors