Hi,
I have a set of studies that looked at the association of sex w.r.t to multiple variables. The majority of the studies reported regression variables such as beta, b values, t-stats, and standard errors. Is it possible to run a meta-analysis using any of the above-mentioned variables? If so, which software would be more meaningful to perform a meta-analysis? I did a wee bit of research and found out that Metafor in R would be the better choice to perform these kinds of meta-analyses.
Any help would be highly appreciated!
Thanks!
Hi,
you may meta-analyze the bivariate correlation coefficients that are depicted in most studies. If not, write an email to the author.
You may convert the relationship between an IV and the DV from the regression analysis in a semi-partial or partial correlation but the problem is that these don't follow a defined sampling distribution. The reason is that the context (i.e., the set of other predictors and covariates) affect the regression coefficient.
Recenty, a paper by Aloe proposed to solve this in a meta-regression where the sets of control variables can be represented as dummies in a meta-regression. I have not yet tested this idea but think that studies will differ to such a large degree (wrt the controls) that you'll end up with one dummy per study....Perhaps an extension could be to create a dummy for each used covariate and control for these......
Aloe, A. M. (2015). Inaccuracy of regression results in replacing bivariate correlations. Research Synthesis Methods, 6(1), 21-27.
Best,
Holger
- Badges
- Science topic