Using three-level meta-analysis in OSMASEM

Posted on
No user picture. Dabiriyan. Joined: 06/17/2020
Forums
Dear all,
I am currently working on MASEM to assess the direct effect of LSC on ASB as well as the indirect effect of LSC on ASB via SB.
I have a question about how to use the three-level meta-analysis in OSMASEM for studies that did not provide the correlation coefficient of an aggregated measure between LSC, SB, and ASB. In this respect, some studies provide multiple effect sizes for a single relationship.
As far as I know, there are two ways to solve this problem: first, aggregate effect sizes and perform OSMASEM, or second, perform analysis on bivariate relationships by fitting the three-level meta-analysis and using the correlation matrix as input.
I aimed to focus on the second approach in my research to estimate average effect sizes. To do this, I wrote the R code on bivariate relationships by fitting the three-level meta-analysis, which is attached. However, I'm not certain that what I did was correct to estimate average effect sizes for studies that provide multiple effect sizes for a single relationship.

I'd be extremely appreciative if you could point me in the right direction.

Replied on Mon, 09/06/2021 - 22:36
Picture of user. Mike Cheung Joined: 10/08/2009

It does not seem right. You have put study1 to study32 as predictors and Cluster (study1 to study32) as a high-level cluster.

If you don't have any moderators, one suggestion is to conduct a three-level or robust multivariate meta-analysis on the 3 correlation coefficients. You may then use the wls() to fit the correlation structures on the average correlation matrix.

The three-level OSMASEM is not available yet. However, I am working on it.