Can I create effect sizes from published reports without access to the data? Thinking about standardizing unstandardized coefficients?

Question

I'm reviewing and synthesizing 48 published articles on the same topic (association between x and y). Instead of describing findings based on statistical significance and power, I'd like to be able to compare effect sizes between papers. However, few authors report effect sizes or standardized regression coefficients. Is there a way for me to conduct effect sizes based on their publications, without access to the data?

In general, they report unstandardized regression coefficients and basic descriptives. I'm wondering if I can standardize those coefficients? If a standardized coefficient = the unstandardized coef *(SDx/SDy), can I draw the standard deviations from the descriptives table and multiply them by the unstandardized coefficient provided? Any issue with this method? I'm concerned, for instance, that some unstandardized coefficients will be main effects and others will account for covariates, which differ between studies?

Any alternative option, like a Cohen's d?

Answer 1

Yes. You can convert unstandardized regression coefficients (from linear regression models) into effect sizes (e.g., Cohen's d). However, these coefficients will be subjected to different degrees of residual confounding, since they are most likely obtained with different models.

For details, check:

Cohen's d from a linear regression model