Defining Level 2 clusters in three-level meta-analysis

Posted on Sat, 04/27/2024 - 04:00

jennies001 Joined: 10/19/2023

Forums

Hello community,

I've recently been exploring the fascinating method of three-level meta-analysis for handling dependent effect sizes and incorporating correlations among them. In my research, I came across an insightful guide by Cheung (2019) that highlights the benefits of this approach.

I understand that in three-level meta-analysis, it's crucial to define the groups or categories within which the effect sizes are correlated. These groups, often referred to as 'clusters,' can be at the study level or based on other factors like countries. However, I'm unsure about how to determine if a particular grouping is appropriate and well-defined.

Could you provide some guidance on evaluating the suitability of a grouping or cluster definition in three-level meta-analysis? What criteria or considerations should I keep in mind while deciding how to organize the effect sizes into these groups?

Thank you in advance for your help!

Reference:
Cheung, M.W.L. A Guide to Conducting a Meta-Analysis with Non-Independent Effect Sizes. Neuropsychol Rev 29, 387–396 (2019). https://doi.org/10.1007/s11065-019-09415-6

Replied on Mon, 04/29/2024 - 09:54

Mike Cheung Joined: 10/08/2009

One starting point is the

One starting point is the multilevel literature. From there, you may learn about hierarchical structures. Then, you may refer to the applications of three-level meta-analysis to learn more about how different levels relate.

Log in or register to post comments

Replied on Fri, 05/10/2024 - 09:18

jennies001 Joined: 10/19/2023

Thank you!

Dear Professor Cheung,

Thank you for your valuable feedback. I find your suggestions extremely helpful. Recently, I have been exploring the distribution of variance across different levels in three-level meta-analysis. I am curious if there are any rules of thumb to determine the appropriate amount of heterogeneity at each level. Specifically, given the assumed dependence among effect sizes, I wonder whether it is preferable for the heterogeneity at the cluster level to be small. However, if this heterogeneity is reduced to zero, it essentially eliminates the distinction between a three-level meta-analysis and a traditional two-level approach. This seems quite tricky to balance.

Thank you very much for your insights.