I am currently performing a meta-analysis on bank relationship and firm performance with 27 different studies; almost all of them report different cases.
I calculated the partial effect size (I have different dependent and independent variable), its variance and confidence interval using the formula provided by Aloe & Thompson, "The synthesis of partial effect size".
Due to the studies-specifics I have now to face two issues: - Studies have different sample size and different specification, so heteroskedasticity is likely to arise; - Studies present several cases, so some of the observation in my meta-analysis are not independent;
I thought to face the second issue by using what in econometrics would be called an "study-fixed effect regression" that, if I got correctly, in the metafor package should be:
res <- rma(Yi, Vi, mods = ~ I(study)) res
That is, a Mixed-Effects Model. I would also like to compare these results with the one of a multilevel\hierarchical model (Gelman & Hill, 2006).
ram.mv package good for that or I have to manually write the function following Gelman & Hill? If it is good, how can I can calculate my Variance-Covariance Matrix considering the fact that I have different independent variable? Last, how can I account for heteroskedasticity?
Thank you for your time