How to run Mediation analysis in SPSS (Traditional method of Barron and Kenny & Process Macro)

Mediation analysis in SPSS is used to test whether an independent variable affects a dependent outcome through an intermediate mechanism—rather than only through a direct link. In this walkthrough, transformational leadership is treated as the independent variable, psychological well-being as the dependent variable, and job satisfaction as the mediator. The core takeaway is that job satisfaction carries part of the effect of transformational leadership onto psychological well-being, meaning the relationship is best described as partial mediation.

The session starts by laying out the mediation logic associated with Baron and Kenny: the independent variable (X) must predict the dependent variable (Y), X must also predict the mediator (M), and M must predict Y when X is included. A final check distinguishes full from partial mediation. In the “ideal” full-mediation case, once the mediator is controlled, the X → Y path becomes non-significant. If the X → Y path remains significant but shrinks, the mediator is functioning as partial mediation. The transcript also frames mediation as a way to probe underlying mechanisms—how X influences Y through M—rather than just reporting association.

To implement the traditional method in SPSS, the workflow begins with creating scale averages: transformational leadership (TL) is computed from its items, job satisfaction (JS) is averaged across its items, and psychological well-being (PWB) is averaged across eight items. With these composite variables ready, the analysis proceeds via linear regression in three stages. First, PWB is regressed on TL to confirm that TL significantly predicts PWB (meeting the first condition). Second, JS is regressed on TL to confirm that TL significantly predicts the mediator (meeting the second condition). Third, PWB is regressed on JS to confirm that JS significantly predicts PWB (meeting the third condition).

The final traditional step tests mediation by regressing PWB on both JS and TL, entering JS first and TL second to control for the mediator’s influence. TL remains significant after controlling for JS, which rules out full mediation. The transcript highlights the change in the TL coefficient: it drops from about 0.628 to about 0.541 after including the mediator, indicating partial mediation.

The walkthrough then repeats the mediation test using PROCESS Macro in SPSS, emphasizing a model-based approach. After installing PROCESS, the analysis selects model 4 for mediation, sets TL as X, PWB as Y, and JS as the mediator. Bootstrap resampling is used (5,000 resamples), and significance is assessed using confidence intervals: mediation is supported when the indirect effect’s confidence interval does not include zero. The output reports coefficients for the X → M path (TL to JS), the M → Y path (JS to PWB), and the direct and indirect effects. The direct effect decreases after accounting for the mediator (reported around 0.548), while the indirect effect is significant because its confidence interval excludes zero. The conclusion aligns with the traditional method: job satisfaction partially mediates the relationship between transformational leadership and psychological well-being.