# 'moderator' analyses in a meta-analysis

Sorry if I use the wrong terms here, I am still learning how to do this.

I am conducting an meta-analysis comparing two treatment types, using Cohen's d. I have completed the primary analyses of the two groups. So I know the overall effects, I know which ones need random-models instead of fixed. I know the q-, z-, and c-statistics as well for each group.

Now however, I want to look at how a particular component of treatment influenced the overall mean effect size for each group. In particular, I want to see if treatments that used exposure techniques influenced the overall mean effect size for each group.

How do I do this?

Do I conduct another 'meta-analysis' on the subgroups and compare the overall mean effect size and z-score to the other sub-group (i.e., treatment group 1 with exposure versus treatment group 1 without exposure), do I compare the sub-group to the overall treatment group (i.e., treatment group 1 with exposure versus all of treatment group 1), or do I use some other statistical test?

EDIT Or do I compute an over all q statistic comparing each group? And if so what do I do when one group according to its q-stat is random-model and another is a fixed-effects?

Thank you.

The issues have been discussed at some length here http://www.metafor-project.org/doku.php/tips:comp_two_independent_estimates. You may need to come back with some other issues, in particular using a fixed effect model for one group and random for the other would not in general be a good idea since they are different models with different goals. Many authors suggest you choose on the basis of observed heterogeneity but you should really pick the one which corresponds to you scientific question. More detail about that issue is available in this answer Meta-analysis of standard deviation using the metafor package in R: can we distinguish between the different types of variability?