Skip to content
Prev 4018 / 5636 Next

[R-meta] Moderator/predictor in multi-level meta-analysis

Dear Paul,

such discrepancies are to be expected when making group comparisons (not 
only in the meta-analytic context). Subgroup analyses based on separate 
data sets will sometimes lead to different estimates than the combined 
data set, because the latter contains more information. Therefore, group 
comparisons within one model are often recommended.

In the multi-level context there is a phenomenomen called shrinkage, 
which reduces extreme estimates. It is often regarded as a good thing, 
because extreme estimates might be unreliable. The extent of the 
shrinkage depends on the data and random effects that were included 
which could explain different estimates across analyses.


Best,
Lukasz
Message-ID: <ea12502f-dff5-dd23-9ef3-91258e833d1b@uni-osnabrueck.de>
In-Reply-To: <mailman.4018.15.1649325610.58803.r-sig-meta-analysis@r-project.org>