I am running an OLS regression and it turned out that the effect of my main predictor X on the outcome Y depends on one of my control variables (age). The central focus of the paper is describing the multiple mediators that link X to Y, and splitting up the sample reduces my power drastically. What are my options?

I have probed the interaction and calculated the regions of significance, splitting my sample based on these regions of significance eliminates most of my significant effects. Sigh..

Is it OK to simply include the interaction in all analyses: i.e, Does including the interaction adequately control for the subgroup differences?


It would be helpful (at least to me) if you can characterise the context and what you mean by 'describing the multiple mediators that link X to Y' a bit more clearly. Given that you are looking at regression, I am assuming that X and Y are both are continuous.

In response to your question, I see no immediate issue with the idea of introducing an interaction term (Y ~ X + Age + X:Age + e) into the model. Have you tried this and if so, what do the diagnostics show? Furthermore, is the interaction term significant?

| cite | improve this answer | |
  • $\begingroup$ Hi Mark, thanks for your help. You are correct, both X and Y are continuous. I am using Andrew Hayes' macros for multiple mediation (afhayes.com/…). I was testing whether the focal predictor interacted with the covariates, and it turns out it interacted with age. Now I am wondering in what ways it affects my model in which X is clearly related to Y through 3 separate mediators. The age*X interaction is significant. The diagnostics seem fine- Are you concerned about anything in particular there? Thanks! $\endgroup$ – user27193 Jun 23 '13 at 13:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.