When using meta-regression with factor moderators, result differ a bit from using seperate estimation based on subgroup, even if the same model (mixed effects) is used for both. I understand the difference mostly stems from estimation of $\tau^2$. I can't say from data, but it seem the subgroup-based method is more usually used for forest plots.

I'm using R and metafor, and it's easy to add subgroup polygons, but not with results from meta-regression. Am I wrong in assuming this is for a reason? Wouldn't it be better to display results from meta-regresson, unlessof course differences in amount of heterogeneity are of interest?


1 Answer 1


The difference between subgrouping versus using a meta-regression model is explained in detail here:


Indeed, it comes down to estimating a single $\tau^2$ value in the meta-regression model versus different $\tau^2$ values when using subgrouping. But one can also fit the meta-regresssion model with different $\tau^2$ values for each level of the factor, in which case the two approaches are identical.

I don't know why you find it difficult to add polygons based on predicted effects to a forest plot. An example of that is provided under help(addpoly.default) (yes, the examples given there is based on a continuous moderator, but the same principle is easily adapted when the moderator is a factor).

  • $\begingroup$ Thanks @Wolfgang, both fir the answer and the wonderful package. $\endgroup$
    – Ozeuss
    May 2, 2016 at 5:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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