[R-meta] Meta-regression or simple GLMM

Viechtbauer Wolfgang (SP) wolfgang.viechtbauer at maastrichtuniversity.nl
Sat Mar 24 14:56:15 CET 2018

rma.glmm() is for binomial and Poisson data, not mean proportions, so it cannot be used for this type of data.

But as far as I understand your data, you have estimates (mean proportions) and corresponding standard errors (you wrote 'standard deviations', so if you have the SD of the individual proportions within a study, then just divide the SD by the sample size and you have the SE of the mean proportion). You can just analyze these data with rma(), that is,

rma(meanprop, sei=se, data=dat)

where meanprop is the mean proportion and se the corresponding standard error.


-----Original Message-----
From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces at r-project.org] On Behalf Of Denis Lafage
Sent: Thursday, 08 March, 2018 10:35
To: r-sig-meta-analysis at r-project.org
Subject: [R-meta] Meta-regression or simple GLMM

Hello all,

I am conducting my first meta-analysis on the diet of arthropod predators (expressed as a proportion of aquatic prey) in riparian ecosystems. Given the heterogeneity of technics to estimate this proportion, I recomputed all the estimates from raw data extracted from figures and tables. Now I have a dataset with a proportion of aquatic prey and a standard deviation (given by the Bayesian mixing model used to calculate the proportion) for each species, on each sampling site of each study.

I want to see if local and landscape factors (data I produced from satellite data) explain the proportion of aquatic prey in the diet. For each random variable, I have one value per sampling site.

As it is almost a primary dataset, I started with a simple GLMM on mean proportion of aquatic prey with study and site as fixed factors and local and landscape variables as random factors.

But maybe I'd rather go for a meta-regression with rma.glmm?

- In this case, how do you deal with a proportion (which is not the result of failure/success) as effect size? Should I use measure="PLO" for logit transformed proportions (i.e., log odds) ?

- Should I use the standard error of the proportion of aquatic preys for each site or calculate the standard error from the standard deviations I obtained from the Bayesian mixing model?

Thank you for your help

Best regards


More information about the R-sig-meta-analysis mailing list