Mediation Analysis: Conceptualization, Interpretation, and Reporting Mediation using SmartPLS

Mediation analysis centers on a causal chain: an independent variable (IV) influences a mediator (MV), and that mediator then drives the dependent variable (DV). In practical terms, the IV may not affect the DV directly; instead, its impact is transmitted through the mediator. This matters because it distinguishes “how” an effect happens (the mechanism) from simply whether an effect exists.

A mediator variable is defined as the variable that explains the relationship between IV and DV—specifically, it carries the influence of IV to DV. When the IV’s effect on the DV shrinks after the mediator is added to the model, that pattern signals mediation. The transcript lays out the conditions typically used to justify mediation: (1) IV significantly affects the mediator, (2) IV significantly affects the DV when the mediator is absent, (3) the mediator has a significant unique effect on the DV, and (4) the direct effect of IV on DV becomes smaller once the mediator is included.

The framework uses a path model with three core quantities. The total effect (labeled C) is the IV → DV relationship without the mediator. Once the mediator is included, the direct effect becomes C′ (IV → DV controlling for the mediator), and the indirect effect is the product of paths A and B (IV → MV times MV → DV). If the indirect effect (A×B) is significant, mediation is present. If both indirect and direct effects are significant, the result is partial mediation; if the direct effect is insignificant but the indirect effect is significant, the result is complete mediation; if the indirect effect is insignificant, there is no mediation.

A concrete example is provided: servant leadership (SL) is treated as the IV, career satisfaction (CS) as the mediator, and life satisfaction (LS) as the DV. The reported path results show that SL significantly predicts CS (path A is significant) and CS significantly predicts LS (path B is significant). However, when CS is included, the direct effect of SL on LS becomes insignificant (C′ is insignificant), while the indirect effect through CS is significant. The transcript interprets this combination—significant total effect, significant indirect effect, and insignificant direct effect—as complete mediation, meaning the influence of servant leadership on life satisfaction is fully transmitted through career satisfaction.

For reporting, the transcript emphasizes using a structured template: state the mediating role, report the total effect (C), the direct effect (C′), and the indirect effect (A×B), along with statistical indicators such as beta coefficients, T values, p values, and bias-corrected confidence intervals. In the example, the total effect of SL on LS is reported as significant (beta 0.268; T 6.650; p < .1). With the mediator included, the direct effect becomes insignificant (beta 0.034; T 0.958; p 0.338), while the indirect effect via CS is significant (beta 0.234; T 7.336; p 0). The bias-corrected confidence interval for the indirect effect does not include zero, reinforcing that the mediation effect is statistically significant. The takeaway is that mediation reporting requires separating total, direct, and indirect effects and then classifying the mediation type based on which paths remain significant.