I am conducting a meta-analysis to understand the impact of an intervention X on people’s responses as measured by three categories of measures. I have a couple of questions regarding the methodology:
Some studies looked at the effect of X on the dependent variable directly, as well as on a mediating variable. I was wondering if it is ok to treat the mediating variables as another DV to include the effect size in my meta-analysis? The reason for this is because X is a latent variable, hence, including both effect sizes for the DV and mediating variable could give more information on how participants respond to X.
As I mentioned, I have data on three types of DV. Given that the same set of participants is used in each study for these DVs, there is a correlation between the sampling errors of the effect sizes within each study. To take this into account, I have considered using a multivariate meta-analysis, which would yield an overall effect size for each category of DV. However, several studies in my sample have multiple effect sizes for each DV category, and I’m not certain how to aggregate those. An alternative option would be to use a multi-level meta-analysis, where the individual effect sizes would be grouped by category of DV, which would in turn be grouped by study. However, it is my understanding that this model assumes that the participants are not the same across different experiments, so there is no correlation between sampling errors, which is not the case for me. Given the above, I was wondering which methodology would be most appropriate?
Apologies for the long post, any replies would be greatly appreciated!