How to perform a Regression after Propensity Score Matching?

[Shanna OReilly]

Hi everyone,

I performed a propensity score matching with the command psmatch2 to reduce my sample bias (this worked successfully). The treatment variable is statistically significantly different between the treatment and control group. Since I am doing this as part of my final thesis, I also want to include a series of regressions. However, I am not sure how to do a (logistic) regression comparing the matched pairs after propensity score matching. Because if I have applied the psmatch2 command and also tested with the pstest command whether the treatment and control group are balanced with regard to the variables (yes!), and then run a logistic regression (logit dependent variable independent variable control variables...), then this is again based on the unmatched data. So my question is, how do I apply the regression to the matched data?

Thank you very much in advance!

[Wei Hai Deng]

Hi Shanna,

It has been a while since I last used psmatch2 but from what I remember and looking through the help file, psmatch2, creates matching variables _id, _nk, or _nn, as well as _treated/_support depending on your matching criteria.

[Andrew Musau]

You should not really extract the matched sample and run a regression as the propensity scores are themselves estimates. But, the linked thread shows you how to do it under -Regression on the 'Matched Sample'-

https://www.ssc.wisc.edu/sscc/pubs/stata_psmatch.htm

[David Radwin]

You should not really extract the matched sample and run a regression as the propensity scores are themselves estimates.

Others disagree. See the references in https://www.statalist.org/forums/for...81#post1422481.

[Andrew Musau]

Others disagree.

I see it done all the time, but I am on the other side of the fence on this!