Hello,

I am reaching out with a question about the appropriateness of a multivariate vs. multilevel meta-analytic model for some data that I am collecting.

I am collecting daily diary studies (i.e., where people are measured several times per day, per week, etc.) and hoping to meta-analyze within-person relationships. It's easy enough to collect within-person correlations (vs. between-person correlations), but a very frequent occurrence is when there is more than one effect size for the same X-Y measures due to there being different time points. For example:

* Correlation 1: Morning Mood (X) and Morning Job Satisfaction (Y)

* Correlation 2: Afternoon Mood (X) and Afternoon Job Satisfaction (Y)

Again, it's easy enough to form composites here to create a single effect size, but I am also interested in moderator variables that might have different values for each effect size (i.e., and would become meaningless when forming a composite for the study as a whole). For example, if one moderator variable is "socializing with others," there might be a 0 value for the morning measures and a 1 value for the afternoon measures.

So, my question, is whether this data situation would be more appropriate for multivariate vs. multilevel meta-analysis? My thinking would be the latter would be more appropriate, although I am uncertain as to the level specification (i.e., Level 1 = effect size, Level 2 = ... measurement occasion?, Level 3 = study).

Any assistance would be greatly appreciated!

John

Hi,

The data structure suggests that a multivariate meta-analysis is more appropriate if you can estimate the sampling covariances.

Miike