[R-meta] Modeling dependent effect sizes (MASEM/TSSEM)

[Viechtbauer, Wolfgang (SP)]

Hi Marius,

Even though we may want to use cluster-robust methods in the end, I think one should try to have a 'working model' that is at least a decent approximation. This will increase the efficiency of the parameter estimates and can also prevent that studies with a lot of effects are receiving 'undue' weight in the analysis. So, I would still use:

Step1 <- rma.mv(yi=r, V=N_INV,
                data=Robust1,
                random=list(~1|K_ID,~1|ESID),
                method="ML", mods=~factor(Zelle)-1)

as my working model and then:

step2 <- robust(Step1, cluster = Robust1$K_ID, adjust = TRUE)
summary(step2)

for the cluster-robust inferences.

Best,
Wolfgang

-----Original Message-----
From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces using r-project.org] On Behalf Of Marius Wuketich
Sent: Saturday, 17 November, 2018 15:33
To: r-sig-meta-analysis using r-project.org
Subject: [R-meta] Modeling dependent effect sizes (MASEM/TSSEM)

Dear colleagues,

I have a question about different approaches to modeling dependent effect
sizes. I'm doing a TSSEM study (based on correlation coefficients)
according to Mike Cheung in metaSEM. My path model is a model with one
independent and one dependent variable and three mediators.

A small part of my work deals with different methods to consider dependent
effect strengths in a MASEM analysis. I would like to compare the results
of the following three methods:

1. simple averaging the effect sizes

2. the approach of Wilson/Polanin/Lipsey 2016 (
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4905597/) - a combination of a
three-level hierarchical model and TSSEM

3. the robust variance estimation according to Tipton/Hedges/Johnson 2010 (
https://onlinelibrary.wiley.com/doi/abs/10.1002/jrsm.5) for the first part
of the TSSEM

With the third approach, I now have a basic understanding problem - how do
I have to specify my model in metafor for the robust approach?

K_ID = Study ID
ESID = Effect sizes ID

Second approach (three-level-model):

Step1 <-rma.mv(yi=r, V=N_INV,
               data=dataset,
               random=list(~1|K_ID,~1|ESID),
               method="ML", mods=~factor(Zelle)-1)
summary (Step1)

Third approach (robust variance approach):

Step1 <-rma.mv(yi=r, V=N_INV,
               data=Robust1,
               random= ~1|K_ID,
               method="ML", mods=~factor(Zelle)-1)
summary (Step1)

step2 <- robust(Step1, cluster = Robust1$K_ID, adjust = TRUE)
summary(step2)

######################################################

Is that specified correctly yet? It just seems too simple to me.

Best,
Marius