Calculating Partial Correlation from Published Work

vltompkins

New Member
I am working on a meta-analysis, and I read in a published meta-analysis that the researchers calculated partial correlations controlling for two variables from published studies that did not actually report partial correlations--I'm guessing just the zero-order correlations matrix? I cannot find from the web or this forum or that article, however, exactly how to do that and what values you would need to calculate this on your own.

So, I would be looking at published studies that have means, SD, and basic correlations among the variables of interest, and I want to calculate the partial correlation between two variables controlling for the other two when those partial rs are not reported in the publication.

spunky

Can't make spagetti
I am working on a meta-analysis, and I read in a published meta-analysis that the researchers calculated partial correlations controlling for two variables from published studies that did not actually report partial correlations--I'm guessing just the zero-order correlations matrix? I cannot find from the web or this forum or that article, however, exactly how to do that and what values you would need to calculate this on your own.

So, I would be looking at published studies that have means, SD, and basic correlations among the variables of interest, and I want to calculate the partial correlation between two variables controlling for the other two when those partial rs are not reported in the publication.
I am working on a meta-analysis, and I read in a published meta-analysis that the researchers calculated partial correlations controlling for two variables from published studies that did not actually report partial correlations--I'm guessing just the zero-order correlations matrix? I cannot find from the web or this forum or that article, however, exactly how to do that and what values you would need to calculate this on your own.

So, I would be looking at published studies that have means, SD, and basic correlations among the variables of interest, and I want to calculate the partial correlation between two variables controlling for the other two when those partial rs are not reported in the publication.
have you by any chance looked at the wiki page for the partial correlation? there you'll find the formula (for the case where you wanna partial out the effect of the variable z from the correlation between x and y):

$$r_{xy.z} = \frac{r_{xy}-r_{xz}r_{yz}}{\sqrt{1-r_{xz}^{2}}\sqrt{1-r_{yz}^{2}}}$$

if you have more than one variable that you'd like to partial out (so more than one 'z' in this case) then you just have to re-do the whole process again with the previous correlation that has already been partialled. that's why it's called a "recursive" formula in this case. this is the kind of stuff for which software is very useful in case you can use a computer to help you out.