# Publication bias and proportional data in metafor

I have been recommended to use rma.glmm for an analysis of proportions and I am in need of publication bias tests. Are any possible in metafor given proportional data, please? The analysis is using the "PLO" effect size which is a mixed-effects logistic regression (according to the manual). How does the rma.glmm approach differ from the rma function in this instance? Both appear to be mixed-effects regression.

Can a glmm analysis be translated into rma by first calculating the logit and SE and then the publication bias estimates run?

• If this is about recent trials (e.g. for a relatively new drug), then you could also consider checking clinicaltrials.gov for registered trials with missing results. Commented Jun 14, 2017 at 15:19

You can easily do a 'regression test' for funnel plot asymmetry manually (and inspect the funnel plot). Here is an example:

library(metafor)

dat <- get(data(dat.pritz1997))

### random-effects model with logit transformed proportions
dat <- escalc(measure="PLO", xi=xi, ni=ni, data=dat)
res <- rma(yi, vi, data=dat)

### check for funnel plot asymmetry
funnel(res)
regtest(res)

### regression test is just the same as using sqrt(vi) as moderator
res <- rma(yi, vi, mods = ~ sqrt(vi), data=dat)
res

### random-effects logistic regression model
res <- rma.glmm(measure="PLO", xi=xi, ni=ni, data=dat)

### check for funnel plot asymmetry
funnel(res)
res <- rma.glmm(measure="PLO", xi=xi, ni=ni, mods = ~ sqrt(vi), data=dat)
res

• Wolfgang, thanks very much but the syntax for the glmm funnel plot test returns an error (using the Pritz data set). On my computer it says "Error in sqrt(vi) : non-numeric argument to mathematical function". Commented May 9, 2017 at 20:46
• Runs fine here. Make sure you have the most current version of metafor installed (version 1.9-9). Commented May 10, 2017 at 8:01