What is Mediation Analysis: Theory, Issues, and Suggestions

Mediation analysis matters because many social-science relationships are not purely direct: an independent variable (X) can influence an outcome (Y) by first changing an intermediate mechanism (M). Relying only on the “direct effect” between X and Y can mislead interpretation—especially when X has little or no direct impact but still exerts a meaningful influence through an indirect pathway. That’s why contemporary research increasingly treats mediation models as essential rather than optional, particularly for publication where journals expect more than simple bivariate testing.

At the core is a causal sequence: X affects M, and M affects Y. In a concrete example, servant leadership may improve life satisfaction, but the mechanism runs through career satisfaction. In that setup, career satisfaction is the mediator because it clarifies how leadership focused on followers translates into better life satisfaction. The practical takeaway is that mediation analysis is not just a statistical add-on; it is a theory-driven explanation of mechanism. Before running any models, researchers need to justify why M belongs in the framework—answering not only “what mediates?” but “why this mediator explains the X→Y relationship,” and what theoretical contribution it adds.

The transcript also lays out the main quantities used in mediation. The total effect (C) captures X’s impact on Y without the mediator. Once M enters the model, the direct effect becomes C′ (the effect of X on Y controlling for M), while the indirect effect is computed as the product of paths a and b (a×b), representing X→M and M→Y. Depending on whether the indirect effect and direct effect are significant—and on the signs of these effects—different mediation patterns emerge.

A major shift is away from the classic Baron and Kenny causal-step approach, which required X to significantly predict Y before testing mediation steps. That “precondition” is now treated as overly restrictive because mediation can exist even when the total effect C is insignificant, particularly under competitive mediation where the indirect and direct effects can oppose each other. Modern guidance emphasizes testing the indirect effect directly rather than treating the direct relationship as a gatekeeper.

For PLS-based mediation, the recommended workflow starts with the significance of the indirect effect (a×b). Bootstrapping is used to assess significance because normality assumptions for indirect effects are not reliable in this context; thousands of resamples (e.g., 5,000) are typically used. If the indirect effect is significant, researchers then evaluate the direct effect C′ to classify mediation as full (direct effect insignificant) or partial (direct effect significant). If the direct effect is significant, the sign of a×b relative to C′ helps distinguish complementary mediation (same direction) from competitive mediation (opposite direction).

The transcript further warns against overclaiming “full/complete mediation,” arguing that it implies all other mediators have been measured perfectly—an unrealistic standard that can discourage future theory development. Instead, model fit checks (e.g., SRMR thresholds such as < 0.8 in PLS software like SmartPLS) and strength measures like VAF (variance accounted for) can help interpret how much of Y’s variance the mediation pathway explains. Finally, multiple-mediator models can produce inconsistent mediation patterns where specific indirect effects are significant but the combined indirect effect is not, underscoring the need to interpret pathways, not just totals.