I'm conducting a meta-analysis where I have a given set of studies, each with multiple effects (correlations in my case) I'm interested in. Each correlation coefficient answers a different research question, so I'm actually conducting many meta-analyses, one per effect... For each effect I've performed the following (I FDR-correct everything at the end): I've first used a random effects model (metafor package) to estimate the overall effect. Then, using the rma function, I re-ran the model this time using moderators. So I end up with estimations of the overall effect and its significance, as well as heterogenity estimations, from the basic model. Then I also obtain indication of moderators that are significantly associated with the effect. My question is about interpreting these two together.

What's not clear to me, from the meta-regression results, is what can I learn about the overall effect? Is it possible that one moderator will come out significant, but looking at the forest plot - none of the correlations were actually significant? (i.e. all had huge confidence intervals crossing the zero-effect line)?

Secondly, is this a legit approach? First identifying significant effects, then examining significant moderators even if the effect by itself did not come out significant?

  • $\begingroup$ If you have a moderator there is no overall effect. $\endgroup$ – mdewey Feb 16 at 16:57
  • $\begingroup$ Interesting question... you may have an overall effect, but no significant moderators, but also the opposite... of course you may have both... the overall effect is a weighted average, whereas a moderator tests the impact of a covariate on the association between treatment and effect (weighting typically for study precision)... the bottom line in your case is that only in specific patient settings the treatment could be beneficial (eg only the elderly)... however, a new dedicated randomized trial is still needed to confirm this $\endgroup$ – Joe_74 Feb 17 at 10:37
  • $\begingroup$ Thank you mdewey and Joe_47 $\endgroup$ – EfratM Feb 25 at 8:01

Your Answer

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

Browse other questions tagged or ask your own question.