Partial correlation coefficients on multiply imputed dataset

[Mohamad El Moheb]

Hi all,
I am trying to derive the partial correlation coefficients R2 from a linear regression performed on a multiply imputed dataset. I tried running the command "mi estimate: pcorr depvar indepvar1 indepvar2" but pcorr is not compatible with imputations. Any thoughts on how to do it?

Thanks!

Edit: Is it mathematically sound to calculate the partial R2's for each imputed dataset and then average them as per Rubin's rule?

[Federico Tedeschi]

Here: https://www.statalist.org/forums/for...ese-equivalent a similar issue is discussed, i.e. how to estimate pairwise correlation in a multiply-imputed dataset. By adapting the solution proposed there, you can use

mibeta depvar indepvar1 indepvar2

Of course the caveats there related to the user-written command "mibeta" (I installed it through "search mibeta") apply. In terms of point estimates, results should, however, be very similar to what you get with

pcorr depvar indepvar1 indepvar2

(that ignores the multiply-imputed nature of the data).