I am conducting a three-level meta-analysis using metaSEM in R. I want to find out the mean effect size for four subgroups of my dataset and later I also want to use these groups as a moderator.

My question is: Why do the estimates differ when I conduct an analysis with a categorial moderator in the meta3 function vs. when I conduct an analysis with a subset of the studies? The four groups of my categorial variable outcat are mot, ent, ges and lei.

This is the code when I use outcat as moderator:

summary(Model2 <- meta3(y=d, v=vd, intercept.constraints = 0, x=cbind(mot, ent, ges, lei), cluster = sid,data=myData, model.name = "Model 2"))

This is the code for the subset (mot as an example):

submot <- subset(myData,outcat==1)
summary(Model1 <- meta3(y=d, v=vd, cluster=sid, data=submot, model.name="Model 1"))

Means for the four groups are (outcat as moderator vs subset for each group)

  1. 0.40 vs 0.23
  2. 0.74 vs 0.81
  3. 0.16 vs 0.24
  4. 0.30 vs 0.18

I have found this thread Why am I getting different means when conducting multilevel meta-analysis with factorial moderator vs. as subgroups? but somehow the answers did not help me that much and also I am using metaSEM so I don't know if that might make a difference.

As the results are so different, I don't know which one I should report or if I have made a mistake and they should actually be the same. In my report, I want to include one table with the main effect for each group and one table for the moderator analysis (is one group significantly higher than the others?). I am wondering if these two tables should be the same (because I can use the estimates from the moderator analysis as main effects for the four groups) or if I should make two tables (one for a subgroup analysis and one for a moderator analysis).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.