Jeffrey B. Vancouver on Tests of Mediation

[We’re pleased to welcome Jeffrey B. Vancouver of Ohio University. Dr. Vancouver collaborated with Bruce W. Carlson, also of Ohio University, on their article “All Things in Moderation, Including Tests of Mediation (at Least Some of the Time)” from the January issue of Organizational Research Methods.]

One of the fundamental empirical activities of scholars in psychology and management is the testing of mediational hypotheses. Nearly every study testing for mediation uses some variation of the statistical analysis approach popularized by Baron and Kenny (1986) and more recently refined by Preacher and Hayes (2004). However, because of the designs used to obtain data for the statistical analysis approach, the strong causal inferences cannot be made. Yet, mediational hypotheses are inherently causal hypotheses. To address this problem, researchers have been reminded that the Baron and Kenny (1986) approach was introduced as an alternative to the causal chain approach when such an approach is impractical. Apparently, the constraint of practicality is overwhelming because promoters of this approach found no examples of its use.

However, there is a third approach to testing mediational hypotheses. Ironically, this third approach involves identify and measuring or manipulating a moderator. Indeed, it is called the moderation-of-process approach, though other terms for the design include the blocking design and the enhancing design exist for it.

The central idea for the moderation-of-process approach to testing mediation is that some variable, either manipulated or measured, would be expected to affect the operation of a mediator by blocking, dampening, or enhancing the mediating mechanism (or links to or from it). If this variable was measured or manipulated, and moderation between the cause and the effect was found, it would imply the mediator was involved. For example, if one thought that performance led to job satisfaction via the rewards performance causes, if one manipulated (or measured) the performance-reward contingency, one would presumably see the performance satisfaction relationship would be affected (i.e., depending on the nature of the performance-reward contingency). Indeed, this study was done and the hypothesis was supported. Several other examples of how the moderator-of-process approach might be employed are provided in the paper.

One of the principle advantages of the approach is that it might also test an implication of mediation. That is, the field’s desire for an explanation for a cause-effect relationship is because such knowledge might lead to interventions that either enhance the relationship (if the effect is desired) or dampen it (if the effect is undesired). For example, if an organization wanted to make sure their most productive workers were their happiest; the moderator-of-process study described above would also indicate that a merit pay reward system would be effective.

By reminding researchers about this often neglected method for assessing mediation, we hope to increase the internal validity of conclusions drawn about mediators via a larger repertoire of approaches. Indeed, the paper ends with a strategy for developing a program of research for triangulating on confirming (or refuting) a mediational hypothesis where multiple approaches are used and combined.

You can read “All Things in Moderation, Including Tests of Mediation (at Least Some of the Time)” from Organizational Research Methods for free by clicking here! Want to know about all the latest from Organizational Research Methods? Click here to sign up for e-alerts!

Jeffrey B. Vancouver, PhD, is a professor of psychology at Ohio University. Much of his work involves using complex protocols that generate time series data to assess dynamic computational models of a person interacting with the environment (usually the experimental protocol) based on a self-regulation perspective.

Bruce W. Carlson, PhD, is an associate professor of psychology at Ohio University. His research interests include probabilistic and statistical reasoning. He serves as an associate editor for the Journal of Statistics Education.