Mediation Analysis for a
Health Behavior Intervention

RESEARCH QUESTION

What the Study Set Out to Test

This mediation analysis example follows a health psychology research team that randomized adults to tailored or standard walking reminders. Alongside the main comparison of activity at week 12, the team wants to know how any effect arises: whether engagement with the reminders carries part of it.

The question is about a mechanism, so it is framed as an indirect effect (reminder type to engagement to activity) alongside the direct effect that remains. Reminders were randomized but engagement was not, so the indirect effect rests on stronger assumptions than the total effect, and the plan says so before any outcome is examined.

STUDY DESIGN AND DATA

A Randomized Trial With a Measured Mediator

A two-arm randomized intervention with around 300 participants. The dataset holds the randomized arm, baseline activity, age and device type, weekly engagement recorded in weeks 1 to 4, and an activity measure at week 12. Measuring engagement before the outcome fixes the temporal order the mediation model requires.

METHOD AND ASSUMPTIONS

Why Estimate the Indirect Effect Directly

Many plans start from the causal-steps sequence, which requires a significant total effect before mediation is examined. This plan sets it aside before any outcome is examined and estimates the indirect effect directly (the product of the reminder-to-engagement and engagement-to-activity paths) with a percentile bootstrap confidence interval, with no precondition on the total effect. In SPSS this is run with the PROCESS macro, model 4, with baseline activity, age and device type as covariates.

Randomization supports the reminder-to-engagement path, but the engagement-to-activity path can be confounded by anything that drives both, such as motivation. The analysis therefore adjusts for the measured baseline variables and states that the indirect effect is interpretable as causal only if no important confounder of engagement and activity remains. The bootstrap is used because the product of two coefficients is not normally distributed, and the number of resamples and the seed are fixed and reported.

OUTPUT AND INTERPRETATION

What the Results Showed

The output has the indirect, direct and total effects, each with a 95% confidence interval (a percentile bootstrap interval for the indirect effect), together with the two path coefficients and a path diagram. The interval for the indirect effect carries the message: if it excludes zero, the data are consistent with engagement carrying part of the reminder effect, a causal reading that holds only under the no-confounding assumption stated in the plan, and the share of the total effect it represents is described with its own uncertainty, not as a precise percentage.

A sensitivity analysis asks how strong unmeasured confounding of engagement and activity would have to be to remove the indirect effect. This is a separate calculation outside the PROCESS macro and is reported as a statement about robustness, not as proof. The write-up states what the estimates support for a design in which only the reminder was randomized.

REPORTING

What the Researcher Received

A written report formatted to APA 7 for the target journal, with statistical wording for the methods, matched to the output, for the team to check and incorporate, and the mediation model described in terms a reviewer can check. The research question, substantive interpretation and conclusions remain with the research team.