Mediation and moderation are two of the most requested analyses in social science, management and health research — and two of the most frequently confused. The difference is simple once you see it.

Mediation: how or why an effect happens

A mediator (M) sits between the independent variable (X) and the dependent variable (Y). X influences M, which in turn influences Y.

Example: Training (X) improves employee confidence (M), which increases job performance (Y). Confidence explains how training affects performance.

Moderation: when or for whom an effect happens

A moderator (W) changes the strength or direction of the relationship between X and Y. It is tested with an interaction term (X × W).

Example: The effect of training (X) on performance (Y) is stronger for employees with more experience (W). Experience changes when training works best.

How to test mediation today

The older "causal steps" approach is no longer recommended on its own. Current practice tests the indirect effect directly using bootstrapping:

  • If the bootstrapped confidence interval for the indirect effect does not include zero, mediation is supported.
  • Report whether mediation is full (no significant direct effect remains) or partial (direct effect remains significant).

Common tools: the PROCESS macro for SPSS/R (Model 4 for simple mediation), SmartPLS bootstrapping of specific indirect effects, or user-defined estimands in AMOS.

How to test moderation

  • Create an interaction term (or use the two-stage approach in SmartPLS for latent variables)
  • Check the significance of the interaction effect
  • Plot a simple slopes graph to show how the relationship changes at low, medium and high levels of the moderator

Mean-centring continuous predictors helps with interpretation of the main effects.

Moderated mediation

When a moderator changes the strength of an indirect effect, you have moderated mediation (for example, PROCESS Models 7 or 14). Report the index of moderated mediation and conditional indirect effects.

What to report

  • Path coefficients with standard errors or t-values
  • The indirect effect with its bootstrapped 95% confidence interval
  • Number of bootstrap samples (often 5,000 or more)
  • For moderation: the interaction effect, ΔR² and a simple slopes plot

The bottom line

Ask yourself whether your theory explains how an effect works (mediation) or when it changes (moderation). Let the theory drive the model — then test it with bootstrapping and report it in full.