We get many questions from clients who use the terms mediator and moderator interchangeably.
They are easy to confuse, yet mediation and moderation are two distinct terms that require distinct statistical approaches.
The key difference between the concepts can be compared to a case where a moderator lets you know when an association will occur while a mediator will inform you how or why it occurs.
Moderation
To explain the concept of moderation, I will offer this question: Is the relationship between interpersonal differences among employees (that is, the independent variable) and employee stress (that is, the dependent variable) different when the conflict resolution behavior of a supervisor differs?
Let’s assume that when a supervisor is considered avoidant, the relationship between interpersonal conflict and employee stress is stronger than when a supervisor is not avoidant.
If that were true, then the supervisor behavior (avoidant or not) would be the moderator of the relationship between the IV (employees’ interpersonal differences) and the DV (employee stress).
In other words, the moderator actually changes the strength or direction of the relationship between the IV and the DV.
How to test moderation
To actually test moderation, simply include an interaction term between the moderator and the independent variable in your model.
Which type of model you use will depend on the measurement level of the variables in the model, the design, and any data issues. Luckily, interaction terms are easy to include in most types of models.
Mediators
Now let’s see a twist in the previous scenario and ask: why would interpersonal conflict among employees (independent variable) relate to employee stress (dependent variable)?
One possibile explanation is anxiety. To explain this relationship we will include anxiety in this scenario.
This allows us to see (1) whether increases in interpersonal differences lead to increases in anxiety levels; then (2) whether increases in anxiety in turn relate to increases in employee stress.
The mediator (anxiety) is the reason why more interpersonal conflict results in more employee stress–because of its impact on anxiety.
How to test mediation
Statistical evaluation of mediation is achieved through path evaluation or structural equation modeling processes or through MLR (multiple linear regressions).
Structural equation modeling or path analysis remains the most appropriate methodology as it permits the simultaneous evaluations of all equations and to directly test the indirect effect of the IV on the DV through the mediator.
Therefore….To Moderate or to Mediate?
When or Why? That is the question!
See the full Series on Easy-to-Confuse Statistical Concepts
Leave a Reply