method notes

method note

Mediation analysis: from Baron–Kenny to bootstrapping

How is a mediation analysis run and reported?

Mediation is tested through the confidence interval of the indirect effect (a×b); Baron and Kenny's four steps and the Sobel test are no longer the standard. If a bias-corrected bootstrap confidence interval excludes zero, the indirect effect is significant. The 'full versus partial mediation' distinction is no longer recommended: a non-significant direct effect does not establish full mediation.

Why bootstrap

The indirect effect is a product of two coefficients, and the sampling distribution of that product is not normal. The Sobel test assumes normality and is therefore underpowered. Bootstrapping builds the distribution by resampling from the data; 5,000 resamples is a common choice.

Mediation in cross-sectional data

Mediation is a claim about process: X affects M, then M affects Y. Cross-sectional data collected at a single moment cannot show that ordering. A statistically significant indirect effect is not evidence of a causal process, and the limitation belongs in the write-up.

The work behind this note

Other method notes

Written for researchers and graduate students. Conventions described here are conventions, not rules — check them against your own design before you report.