Egen and Gen with Pcorr

[Nicholas Otis]

Hi Statalisters,

I am attempting to create a variable equal to the partial correlation between an IV and a DV, stratified on a categorical variable.

For models with a single IV, this has proven quite simple:
sysuse auto
egen PMcorr=corr(mpg price), by(foreign)
ta PMcorr


However, it has proven problematic to replicate this procedure with additional IVs.
The model I wish to create is functionally similar to:

sysuse auto
ta rep78, gen(rep_)
by foreign: pcorr mpg price rep_*

in that it is stratified by a categorical variable, has multiple IVs, and contains both continuous and categorical IVs. I wish to generate from this a variable equal to the partial correlation between price and mpg. Is this possible? The limited documentation on pcorr did seem promising.

Thanks in advance,

Nick

[Nick Cox]

corr() as a egen function will, I believe, work only if you download egenmore from SSC, as you are asked to explain.

I don't know of an equivalent function for partial correlation. It wouldn't seem a good idea to me to try to write a general egen function. I don't know what use a variable containing partial correlations might be, but using statsby to create a reduced dataset with partial correlations and then combining with the original dataset via a merge would seem one way forward.