[R-meta] Random vs. Fixed Effects model in metafor with multiple effect sizes per study

[Viechtbauer, Wolfgang (SP)]

Hi Mia,

1) I cannot answer that question in general, except to say: It depends. Without more information, I would recommend to use an appropriate random-effects model for your data.

2) While conceptually one could argue that, in the absence of heterogeneity, it doesn't make sense to run a moderator analysis, I would still do so (at least as long as the moderator analysis is pre-planned). The test for heterogeneity may not be significant due to low power, but when testing a specific moderator, one might still find that it is relevant. Consider this example:

set.seed(1236)

k <- 15
vi <- runif(k, .001, .50)
xi <- sample(0:5, k, replace=TRUE)
yi <- rnorm(k, 0.1 * xi, sqrt(vi)) # xi is a true moderator

rma(yi, vi) # test for heterogeneity is not significant (p = 0.25)
rma(yi, vi, mods = ~ xi) # test of moderator is significant (p = .005)

Best,
Wolfgang

-----Original Message-----
From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces using r-project.org] On Behalf Of Mia Daucourt
Sent: Tuesday, 25 June, 2019 18:32
To: r-sig-meta-analysis using r-project.org
Subject: [R-meta] Random vs. Fixed Effects model in metafor with multiple effect sizes per study

Good afternoon,

I’d like to run a fisher’s z meta-analysis with multiple effect sizes per study. As such, I have it set up in a multilevel framework. I have two questions:

1.Do I have to run it as a random effects model or can I run it as a fixed effects multilevel model?

2. If not, then I should run random effects model. However, when I run it as a random effects multilevel model I get non-significant heterogeneity. Is it still proper to run a moderator analysis when you do not find significant heterogeneity?

Please let me know if you have any insight on the best course of action to take!

My best,

Mia Daucourt

Florida State University
Developmental Psychology PhD student
IDCD Hart lab
daucourt using psy.fsu.edu