Blog

By Victoria Briones, Ph.D.
Statistics Consultant, HelpWithStatistics.com | DissertationAdvisors.com | DissertationWriting.com

Novice researchers often think the terms moderator and mediator are interchangeable. But these are two different concepts that require different statistical procedures (Baron & Kenny, 1986). The best way to remember the major distinction between these two terms is that a moderator tells you WHEN a relationship will take place while a mediator tells you WHY or HOW a relationship occurs.

In most preliminary research studies, one seeks to determine whether there is a relationship between and independent (the cause or predictor) and the dependent (the effect or the criterion) variables. It is only when many researchers have demonstrated repeatedly that there is indeed a relationship between the independent and dependent variables that other researchers begin to test WHEN and WHY or HOW this relationship occurs.

Moderators

Romer, Rispens, Giebels, and Euwema (2012) noted that interpersonal conflict between colleagues (i.e., the independent variable) affects employee stress (i.e., the dependent variable). They wanted to see if leader’s conflict management behavior would moderate this relationship. Specifically, they wanted to see if leader’s avoiding behavior AND leader’s problem-solving behaviors would moderate this relationship. Romer, et al (2002) demonstrated that WHEN a leader was avoidant, the relationship between interpersonal conflict between colleagues and employee stress was stronger than WHEN a leader was not avoidant. They also found out that WHEN a leader exhibited constructive problem-solving behaviors, the relationship between interpersonal conflict between colleagues and employee stress was weaker than WHEN a leader did not exhibit such behaviors. Note here that the moderators tell us when the relationship is stronger, weaker, or non-existent.

Traditionally, two-way analysis of variance (ANOVA) procedures were used to test whether a variable moderated a particular relationship. The independent and moderator variables always had to be categorical (i.e., nominal or ordinal). Thus, if one had an independent or moderator variable that was measured continuously on an interval or ratio scale, one would have to recode the continuously-measured variable into categories. In the last few years, however, multiple linear regression procedures have been used to test for moderation. When using such a procedure, the independent and moderator variables can be categorical or continuous. Note that researchers, when reporting the statistical results testing moderation, often use the term “interaction” or “product term.” This is because when testing for moderation, the product of the independent and moderator variables is used to test for moderation.

Mediators

Nima, Rosenberg, Archer, and Garcia (2013) noted that lack of self-esteem (i.e., the independent variable) can lead to depression (i.e., the dependent variable). They wanted to see if anxiety would mediate this relationship. Nima, et al. (2013) demonstrated that lack of self-esteem did lead to changes in anxiety levels; in turn, these changes in anxiety levels led to changes in depression levels. Note here that the mediator tells us WHY lack of self-esteem leads to depression: it leads to depression because of its effect on anxiety. It is important to demonstrate that the occurrence of the mediator takes place after the independent variable but before the occurrence of the dependent variable. Many times, however, researchers are unable to show this.

Mediation is assessed statistically in one of two ways: via several linear regression procedures (as noted by Baron & Kenny, 1986) and via path analysis or structural equation modeling procedures. Path analysis or structural equation modeling is the more parsimonious procedure because it allows for the simultaneous test of all equations. Mediating variables are generally continuously-measured (i.e., interval or ratio variables).

REFERENCES:

Baron, R.M. & Kenny, D.A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51(6), 1173-1182.

Nima, A.A., Rosenberg, P., Archer, T. & Garcia, D. (2013, September). Anxiety, affect, self-esteem, and stress: Mediation and moderation effects on depression. PLoS One, 8(9). Retrieved from http://www.ncbi.nlm.nih.gov/

Romer, M., Rispens, S., Giebels, E. & Euwema, M.C. (2012). A helping hand? The moderating role of leaders’ conflict management behavior on the conflict-stress relationship of employees. Negotiation Journal, 28(3), 253-277.

 

ABOUT THE AUTHOR

DR. VICTORIA BRIONES graduated with a PhD in Organizational Psychology from Columbia University. While completing her graduate studies, she taught applied regression analysis to graduate students in education and psychology. Students enjoyed her regression course because she was able to translate complex statistical concepts into a language that the “stats phobic” students could easily understand. Victoria was also an assistant lecturer in research methods (and received the highest mean evaluation for teaching performance). After graduating, she was a research vellow at Harvard University’s Kennedy School of Government. As a fellow, she conducted statistical analyses and wrote articles on negotiation behavior and conflict resolution with her former dissertation adviser.

In the last two years, Victoria has worked as a statistical consultant, helping graduate students in psychology, education, nursing, biology, and business hone their study hypotheses, arrive at better operational definitions of their study variables, and improve procedures to increase the internal and/or external validity of their study. She also performed general statistical procedures such as reliability analyses, non-parametric tests (e.g., Mann-Whitney, Kruskal-Wallis, and chi-square tests), t-tests, analysis of variance (ANOVA), analysis of covariance (ANCOVA), exploratory factor analysis (EFA), and linear regression. Further, she conducted multivariate tests such as multivariate analysis of variance (MANOVA), logistic regression, and structural equation modeling (SEM; using AMOS, LISREL, and EQS). Victoria also created summary tables and graphs of statistical findings and helped students interpret their study results. More importantly, she enjoyed explaining basic statistical procedures and findings to clients who had a limited understanding of such concepts.