Hello!
I have a few questions about MASEM I hope you can help me with. We conducted a random-effects three-level meta-analysis looking at the effects of five predictors on an outcome. We coded the correlations between the predictors and the outcome and then conducted a moderation analysis to compare these predictors in their strength of association with that outcome using Pearson's r correlation effect sizes.
A reviewer has asked about the possibility of testing partial correlations or a path model to account for multiple predictors and their covariance in order to provide better insight into the relative importance of predictors. However, in my reading of MASEM I understand that we would need to extract the bivariate correlations between predictors in order to model a path analysis. So my questions are:
-
Is it possible to conduct a path analysis without the correlations between predictors? Are there any other statistics we could use to run a path analysis or to get an indication of the relationships between predictors (e.g., can we export statistics from the level 1 analysis)?
-
How well does MASEM deal with "patchy"/inconsistent data? Our biggest complication with our dataset is that some studies report few predictors, while others report many predictors, so the correlations would be based on different samples (which we expect would significantly limit the accuracy of the data).
-
Building from Q2, what is the benefit of MASEM over a three-level analysis?
Many thanks in advance!
Crystal
Printer-friendly version
Hi Crystal,
1) MASEM is built on top of SEM. Thus, you need to have the correlation matrices among the variables.
2) MASEM works well with studies with different numbers of variables. For example, one study reports a correlation matrix between Y and X1, and another study has a correlation matrix among Y, X1, X2, and X3.
3) They are addressing different research questions. A three-level meta-analysis is a standard meta-analysis extended to handle non-independent effect sizes. It tries to estimate the average effect, heterogeneity variances at levels 1 and 2, and moderating effect. MASEM attempts to combine correlation matrices to fit some structural equation models.
The following two papers may be helpful to you:
MASEM: https://psyarxiv.com/epsqt/
SEM-based meta-analysis (including a three-level meta-analysis): https://psyarxiv.com/93nfr/
Mike
Hi Mike,
Thank you very much for your helpful (and quick!) response.
Kind regards,
Crystal