I am currently trying to apply different propensity score matching estimators to my data. I have played around with nearest-neighbor-matching, kernel matching, radius and caliper matching with psmatch2. In order to assess the matching quality of each estimator, I have looked at the t-tests and Pseudo-R^2s before and after matching. For the matching quality assessment, it was fairly convenient to simply run "pstest $covariates, sum" after psmatch2.
Now I am working on stratification matching with psmatch2. I am trying to run kernel matching on the stratified sample.
Code:
gen attw = . egen gw = group(blockwf) levels gw, local(gr) quietly foreach j of local gr { psmatch2 treatment if gw==`j', outcome(outcome1) pscore(pscore1) kernel common logit kerneltype(epan) bw(0.06) gen att`j' = r(att) if gw==`j' gen sd`j' = r(seatt) if gw==`j' replace attw = r(att) if gw==`j' gen treated`j' = _treated if gw==`j' gen support`j' = _support if gw==`j' gen weight`j' = _weight if gw==`j' gen outcome1`j' = _outcome1 if gw==`j' } sum attw
Unfortunately, the output does not include t-statistics for the ATTs that I would need. I am also struggling to get the Pseudo-R^2 after matching so that I can compare the matching quality of stratification matching to the other matching algorithms. I manage to get Pseudo-R^2s for each block but no overall Pseudo-R^2. Any ideas how I could construct this or does it even make sense to do this? I would be very grateful for any thoughts on that.
Best,
Ricka
0 Response to Pseudo-R^2 and t-statistics after stratification matching
Post a Comment