Meta-analysis: different information about relationships between two variables

Hello everyone

I need help with my meta-analysis. I have collected all the studies that I wish to include and would like to take Pearson's correlation as effect size. However, many of the studies do not report correlations, but rather path analysis or SEM coefficients or regression coefficients. Is there a way to include those studies anyway (like, say, transform those into correlations???)

I am sorry, if it is a stupid question. I have managed to confuse myself quite completely in the past few weeks...


Less is more. Stay pure. Stay poor.
Your correlations have just two variables and the other approaches have marginal effect (controlling for other variables)?
Wow, you're fast.

Well, it is a bit more complicated than that. I have two independent variables, which are used to predict one dependent variable. I was hoping I could just compare the two correlations (one which derives from the correlations between iv 1 and dv, and the other from the correlations between iv 2 and dv) in the meta-analysis.
However, sometimes the two iv are used to predict the dv in SEM, path analysis or regression...and I feel bad removing all those studies, just because they did not report any correlations.
I know I could write to the authors, but then you never know whether they are going to answer (especially if the papers are rather old). So I was hoping for a statistical solution to this issue...


Less is more. Stay pure. Stay poor.
I have not done SEM, but I know in the graphs they some times use correlations. Is this not the case for your project? I also know that in linear regression you can rework results to get correlations, but if other variables are in the model this may not be entirely accurate for your project, since those values would have derived in a different fashion (controlling for the other variables).
Do you know any good literature or source where I might find out about how to rework correlations from linear regression? I was not successful in finding this before...

Unfortunately, many of the SEM included additional variables (e.g., demographics), which means that the numbers used in the graphs are not correlations. But thank you very much for your input, I feel better now, because I was starting to think that I was just missing the point...


Less is more. Stay pure. Stay poor.
I may have misspoken a little. After a quick search it appears you can convert beta coefficients into correlation coefficients in simple linear regression with the following:

r = b (Sy/Sx)

r = correlation coefficient

b = beta coefficient

S = standard deviations

I would double check this formula, and I will point out that you may or may not be able to get standard deviations or back-calculate them. Lastly, not sure if this holds for multiple regression scenarios.


Less is more. Stay pure. Stay poor.
Good luck and we will see if others post as well. Also, keep us updated on what you figure out or how you proceed.