Dear Statalist Member,

I have several Effect Sizes (ES) with accompanied 95% Confidence Intervals (ES_lci = lower confidence interval of estimate and ES_uci = upper confidence interval of estimate) of different studies and want to get a summary estimate using a random-effect model.

I now used the following commands:
Code:
metan ES ES_lci ES_uci, random

*** Getting a model with log-transformed ES
gen ln_ES = ln(ES)
gen ln_lci = ln(ES_lci)
gen ln_uci = ln(ES_uci)

metan ln_ES ln_lci ln_uci, random eform
First, I was surprised that although the forrest plots are quite similar, the weights and thus the summary estimate is different between the "log-transformed-effect-size" model and the "normal effect-size" model.

Does anyone know what is the correct way to do or if there is an advantage when to use the log-transformed model (at the first glance, the weights seem to be closer to each other, thus studies with a smaller sample size are weighted higher and studies with a large sample size smaller)?

Thank you in advance for your opinion.
Martin