Can I combine the bivariate correlations given in the correlation matrices of several studies regardless which estimation method they used? (E.g. multiple regression, logistic regression, multinomial logit regression...)
It is not possible to combine estimates from binary regressions (probit, logit...) with estimates from continuous variable studies (OLS). However, if you have enough studies, a separate MA of the logistic regression results using the log-odds ratio can be undertaken. Another strategy is to change the focus of the MA (see Garcia-Quevedo, Kyklos, 2004 for example).
Thank you for your reply! However, I am still not sure about the question. This is because I know of several studies that have done a meta analysis with several studies using different analysis techniques.
For example Crook et al. (2013). They meta-analysed all transaction cost studies. I have looked at the articles they have analysed and these articles use techniques such as logistic regression, multiple regression, multinomial logistic regression etc.
It seems that they used simple correlation coefficient as effect size. I haven't checked the primary studies but I assume that they transformed some statistics (chi-square test, etc.) into zero-order correlation. However, it does not make sense to me to transform logit regression coefficients to simple correlation coefficients.
In Meta Analysis we combine the same effect size in different studies, you need to check out what effect size they really combine.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis