A mediation analysis tests whether a predictor (X) affects an outcome (Y) through an intermediate variable, the mediator (M). The modern approach tests the indirect effect (the product of path a, from X to M, and path b, from M to Y) with a bootstrapped confidence interval, rather than the older causal-steps method. The write-up reports every path and puts the indirect effect at the centre.
What you need before you write a single word
- Path a: X → M.
- Path b: M → Y, controlling for X.
- Total effect c: X → Y without the mediator.
- Direct effect c′: X → Y controlling for M.
- The indirect effect ab with its bootstrapped 95% CI and the number of bootstrap samples.
The APA 7 format template
[X] was associated with [M] (path a), b = X.XX, SE = X.XX, p = .XXX, and [M] with [Y] controlling for [X] (path b), b = X.XX, SE = X.XX, p = .XXX. The indirect effect of [X] on [Y] through [M] was significant, ab = X.XX, 95% CI [X.XX, X.XX] (5,000 bootstrap samples). The total effect was c = X.XX, p = .XXX, and the direct effect c′ = X.XX, p = .XXX.
All 53 templates, ready to paste with the italics set: the Rescue Vault, $27 →
Formatting rules reviewers actually check:
- The indirect effect is judged by its confidence interval: significant if it excludes zero. There is no p value for a percentile bootstrap.
- Say how many bootstrap samples you used and which tool (for example PROCESS Model 4).
- Report all four paths; a path diagram with the coefficients helps.
A worked example
For 180 adults, you tested whether rumination mediates the link between mindfulness and depressive symptoms (5,000 bootstrap samples).
- Path a (mindfulness → rumination): b = −0.45, SE = 0.06, p < .001
- Path b (rumination → depression, controlling for mindfulness): b = 0.49, SE = 0.06, p < .001
- Total effect c: b = −0.30, SE = 0.06, p < .001
- Direct effect c′: b = −0.09, SE = 0.06, p = .166
- Indirect effect ab = −0.22, 95% CI [−0.32, −0.14]
Written up in APA 7:
Mindfulness was associated with lower rumination (path a), b = −0.45, SE = 0.06, p < .001, and rumination with more depressive symptoms when controlling for mindfulness (path b), b = 0.49, SE = 0.06, p < .001. The indirect effect of mindfulness on depressive symptoms through rumination was significant, ab = −0.22, 95% CI [−0.32, −0.14], based on 5,000 bootstrap samples. Mindfulness had a significant total effect on depressive symptoms, c = −0.30, SE = 0.06, p < .001, but its direct effect was not significant once rumination was included, c′ = −0.09, SE = 0.06, p = .166, consistent with the effect operating largely through rumination.
Mistakes reviewers catch
- Using the Baron and Kenny steps or a Sobel test alone. Reviewers now expect a bootstrapped indirect effect.
- "Full mediation" claims. A non-significant c′ can reflect low power; describe the pattern instead of labelling it.
- Causal language from cross-sectional data. Mediation implies a causal order the design may not support; say so in the limitations.
- No CI for the indirect effect. It is the main result.
Let KyroStat do the write-up for you
KyroStat estimates every path, bootstraps the indirect effect, and writes this paragraph and a labelled path diagram from your data.
Frequently asked questions
Does the total effect need to be significant? No. Modern practice tests the indirect effect directly, even when c is not significant.
How is mediation different from moderation? Mediation asks how X affects Y; moderation asks when or for whom.
Should I report standardized coefficients? You can add them, or a completely standardized indirect effect, as an effect size. Say which you report.
How do I report the regressions behind each path? See how to report a multiple regression.