[R-meta] [R meta] How to print z score for subgroup analysis
Viechtbauer, Wolfgang (SP)
wo||g@ng@v|echtb@uer @end|ng |rom m@@@tr|chtun|ver@|ty@n|
Wed May 22 10:32:31 CEST 2019
With "mixed-effects model for subgroup analysis", I assume you mean a meta-regression model with a categorical predictor and allowing for residual heterogeneity within subgroups. And I assume you are contrasting this with fitting random-effects models within each subgroup (so we get an estimate of mu_j for subgroups j = 1, ..., m) and then testing whether there are differences between subgroups (i.e., H_0: mu_1 = mu_2 = ... = mu_m). These two approaches are conceptually nearly identical, except that the mixed-effects meta-regression approach (usually) assumes that the amount of residual heterogeneity within subgroups (tau^2) is the same for all subgroups while the second approach allows the amount of heterogeneity within subgroups to differ across subgroups (so we get an estimate of tau^2_j for subgroup j). However, one can also fit a mixed-effects meta-regression model that allows tau^2 to differ across subgroups and then the two approaches are exactly identical. See:
For a further discussion/comparison of these two approaches (i.e., assuming a single tau^2 vs. allowing different tau^2 values per subgroup), see the following article:
From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces using r-project.org] On Behalf Of Cath Kids
Sent: Tuesday, 21 May, 2019 0:56
To: Guido Schwarzer
Cc: r-sig-meta-analysis using r-project.org
Subject: Re: [R-meta] [R meta] How to print z score for subgroup analysis
Dear Michael and Guido,
Thank you very much for your reply!
Yes, what I mean is z-scores for the individual subgroup effects. Thank you
for the R command from Guido!
I did report the between subgroup differences but I also see from the
meta-analyses (at least those in my field) report each subgroup effect (z
scores), so I thought I should report that too.
I've got one more question, can anyone please enlighten me whether is it
generally more preferable to use random-effect model for subgroup analysis
than mixed effect model? (as suggested by Cochrane handbook
Or is there any circumstance where it is better to use the mixed effect
model? I read a few meta-analyses which used mixed effect model for
subgroup analysis so I'm wondering why.
Thank you very much!
On Mon, May 20, 2019 at 10:21 AM Guido Schwarzer <sc using imbi.uni-freiburg.de>
> The z-scores for the individual subgroup effects (if this is what you
> are looking for) are not shown in the output, however, they are part of
> the meta-analysis object.
> You can use the following command to extract the z-scores and p-values
> for the fixed effect and random effects model:
> data.frame(bylevs, zval.fixed.w, pval.fixed.w,
> zval.random.w, pval.random.w))
> However, you should abstain from selectively reporting significant
> subgroup results. Instead, the Cochrane Handbook
> (https://handbook-5-1.cochrane.org/) gives the following advice (among
> other things) on subgroup analyses: (1) conduct a test for subgroup
> differences (section 18.104.22.168) which Michael already mentioned and (2) do
> not compare the statistical significance of the results within separate
> subgroup analyses (section 9.6.6).
> Best wishes, Guido
> Dr. Guido Schwarzer
> Institute of Medical Biometry and Statistics,
> Faculty of Medicine and Medical Center - University of Freiburg
> Postal address: Stefan-Meier-Str. 26, D-79104 Freiburg
> Phone: +49/761/203-6668
> Mail: sc using imbi.uni-freiburg.de
> Homepage: http://www.imbi.uni-freiburg.de
More information about the R-sig-meta-analysis