In short
What a mediation report hangs on is the indirect effect (the a × b path) with its bootstrap confidence interval, not a p-value: ab = 0.14, SE = 0.04, 95% bootstrap CI [0.06, 0.23], 5,000 resamples. The effect is significant when that interval does not include zero. Alongside it you report the a and b paths, the direct effect c' and the total effect c, and you state the software and the model you used (PROCESS model number, lavaan, Mplus) plus the number of resamples and the type of interval. The Sobel test is no longer the standard because it assumes the indirect effect is normally distributed, which it usually is not.
Mediation is one of the analyses people get wrong most when reporting, partly because many still use the Sobel test when bootstrapping has been the standard for years. Here's how to report it properly in APA 7, with the indirect effect and its confidence interval, which is what really matters. The general guide is in how to report results in APA 7.
When you have several paths to report (the indirect effect, a, b, c' and c, plus each one's bootstrap CI), the APA 7 mediation table generator lines them up into a table ready to paste, no manual formatting in Word required.
What a mediation report must include
The heart of the report is the indirect effect (the a×b path) with its confidence interval obtained through bootstrapping (ideally 5,000 or 10,000 resamples, percentile or bias-corrected interval). The indirect effect is significant if its CI does not include zero. You also report the individual paths (a, from X to M; b, from M to Y; c', the direct effect), the total effect (c) and, increasingly, a mediation effect size. Important: bootstrapping doesn't give a p-value for the indirect effect; the decision is made with the interval.
Example in the text
"The indirect effect of self-esteem on well-being through social support was significant, ab = 0.14, SE = 0.04, 95% bootstrap CI [0.06, 0.23] (5,000 resamples, percentile method). The direct effect remained significant, c' = 0.31, p < .001, suggesting partial mediation."
Avoid two things: reporting only the Sobel test (discouraged because it assumes normality of the indirect effect) and talking about "full vs. partial mediation" as a rigid dichotomy; describe the magnitudes.
Moderated mediation and multiple mediators
If you use PROCESS (Hayes's models) or lavaan, state the model or the syntax. In moderated mediation you report the index of moderated mediation with its bootstrap CI. With several parallel mediators, report each specific indirect effect as well as the total. And if the PROCESS output is in front of you and you cannot spot which of all those numbers is your indirect effect, paste the whole block here and it points out what to report.
Common mistakes
The usual ones: using Sobel instead of bootstrapping, not stating the number of resamples or the type of interval, giving a p-value for the indirect effect (not appropriate with bootstrapping), and confusing the total effect with the indirect one.
Before you submit: a bootstrapped indirect effect that never says how many resamples you ran, or a p-value sitting where the confidence interval belongs, is the kind of detail a reviewer flags on the first read, along with bigger issues you can no longer see because you have lived inside the manuscript for months. If yours is already written, run it through the Q1 Reviewer: a free Reviewer 2 style pre-review that points out what an editor would turn into a desk-rejection, before the manuscript reaches their desk.
Frequently asked questions
How do you report a mediation analysis in APA 7?
Give the indirect effect with its bootstrap confidence interval, the a and b paths, the direct effect c' and the total effect c, plus the software and the model: "The indirect effect of self-esteem on well-being through social support was significant, ab = 0.14, SE = 0.04, 95% bootstrap CI [0.06, 0.23] (5,000 resamples, percentile method; PROCESS model 4)." Everything else, including the mediation diagram, is secondary to that line.
Do I report a p-value for the indirect effect?
Not when you use bootstrapping. The bootstrap interval is the test: if it excludes zero, the indirect effect is significant at that level. Adding a p to an interval obtained by resampling mixes two logics and is one of the mistakes reviewers who know mediation flag most often. If your journal insists on a p, the honest route is a Monte Carlo interval or a formal test, stated as such.
How many bootstrap resamples should I use, and which interval?
5,000 as a minimum and 10,000 if the model is small enough to allow it; both figures are conventions that have become the norm, not thresholds with a derivation behind them. On the type: the percentile interval is the safe default recommendation today. Bias-corrected intervals are somewhat more powerful but can inflate the Type I error rate, so if you use one, say so explicitly rather than leaving the reader to guess from the software's defaults.
Do I need a significant total effect before testing mediation?
No. The old Baron and Kenny requirement that the total effect c be significant has been superseded: an indirect effect can exist when the total effect is not significant, because the direct and indirect paths may work in opposite directions and cancel each other out. Requiring a significant c mainly costs you power. What you do report is the total effect, so the reader can see the full picture.
Is full versus partial mediation still used?
It is discouraged, and with reason. Whether c' comes out significant depends heavily on your sample size, so with a small sample almost everything looks like full mediation and with a large one almost nothing does. A non-significant direct effect is not evidence that there is no direct effect. Describe the magnitudes: how large the indirect effect is, how large the direct one, and what that means for your model.
What effect size do I report for a mediation?
The most portable option is the completely standardized indirect effect, which puts ab on a standard-deviation scale and can be compared across studies. The proportion mediated (ab/c) reads well but becomes unstable when the total effect is small or the paths point in opposite directions, so if you report it, do so with that caveat. In any case, the confidence interval of the indirect effect already carries most of the information about magnitude and precision.
Mediation with a 5,000-resample bootstrap, and the simple slopes of moderation with their figure, are worked through on a single study in chapter 7 of the book "Report Your Results in APA 7", with the dataset and the R scripts to reproduce every result.
Mediation, moderation or a conditional model?
I'm a PhD in psychology. I build your mediation with bootstrapping (PROCESS, lavaan or Mplus), including moderated and multiple-mediator models, and hand you the APA 7 report ready to defend.
Get a free consultation →That's your mediation reported the way any Q1 journal expects. If the model gets complex, we look at it in my statistical consulting for theses and papers.