Hello,
I'm currently working on a meta-analysis using pre-calculated effect sizes. Unfortunately, the effect sizes are not consistent between papers e.g., some present regression coefficients, while others have odds ratios. Therefore, I'm trying to convert the effect sizes into the same format before conducting the meta-analysis.
To ensure the effect sizes are the same, I am trying to convert a regression coefficient to hedge's g. The parameters I have for the linear regression studies are the regression coefficient and standard error - am I able to convert this to hedge's g using these parameters?
If you could point me in the right direction or offer some insight, that would be much appreciated.
Thanks, and best wishes.
I don't think that this is possible. What you would need is the (residual) standard deviation. This can not be directly calculated from the standard error of a regression coefficient.
Hi if it is possible to convert regression effects, Cohen's and OR to Pearson's correlation measures and vice versa with this online calculator to specify the use of a single effect size for a possible meta-analysis (e.g. pearson's r).
https://www.psychometrica.de/effect_size.html
Article The use of Bayes factor in clinical cardiology research
Thank you for your responses. Jochen Wilhelm, in cases where you have the regression coefficient and the SE, would you recommend converting to Fisher's and just comparing results with the same metric?
I am not in meta-analyses, so my opinion here likely won't reflect the field's standards. So I would ask myself what the relevant metric is.Then I would thy to find studies from which these relevant metric can be assessed. If studies do not provide it, I would try to find out of there is any possible bias related to which metrics are reported or can be assessed, but I would not switch to "some" metric which is just possible to assess from all (or most) studies. Yes, it might be that there are simply not enough useful studies left, but that is, to my opinion, the most relevant result: that relevant information is often not published.
I suggest using two meta-analysis. One for OR and one for registration coefficient and then compare.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression