Mediation and Moderation are often confused topics. Here, I’ll talk about the differences between the two and what they’re statistically testing. Put simply:
- Moderation attempts to show that a variable changes the relationship between X and Y.
- Mediation attempts to show that a variable is the channel through which X influences Y.
Moderation
Moderate is just a fancy word for “changed” or “influenced.” Thus, moderation is about whether the relationship between X and Y is influenced by a third variable. Suppose you have a theory that the environment of a classroom changes students’ interest. You create the environment of a classroom to be either stereotypically geeky (with star trek posters and techy gadgets), or a neutral environment (with nature posters and some potted plants). You are primarily concerned with the relationship between the environment (X) and course interest (Y). However, you find the effect is moderated (changed) by gender (M). This is done by conducting a regression analysis with X, M and the X*M interaction. Visually, it’s represented like this:
You then decompose the interaction by looking at each gender separately. You find that for me, geeky or neutral environment (X) doesn’t correlate with interest in the class (Y). Maybe because geeky environments are more typical for me anyway. However, for women, class environment matters! Women are more interested in the neutral environment classroom, but are turned off by the geeky stereotypes. Again, this is moderation because gender changes the relationship between X and Y. With men, the relationship isn’t present. With women, the relationship is. Visually, it looks like this:
For Men: Classroom environment (Geeky vs. Masculine) does not influence interest.
Mediation
Mediation is looking for a channel through which X influences Y. For example, supposed that for women, class environment (X) predicts course interest (Y) through perceptions of belonging (M). Statistically, mediation attempts to show two things.
- First, that the pathway connection X –> M–> Y is altogether significant.
- Second, that when this pathway from X –> M –> Y is controlled for, the correlation between X –> Y vanishes. In this case, that would mean that even when women are in the neutral classroom (rather than the geeky one), if they don’t feel like they belong, there is no course interest.
