Hi everyone,
I have run across this question several times throughout the forum, and have also wondered it myself. Basically, people have questioned if there is a way to implement propensity score matching with exact matches on a variable using -teffects psmatch- as with -teffects nnmatch-. I recently came across Ariel Linden's -mmws- file and wondered whether or not this could work out. My question is therefore conceptual.
Assume we would like to exact-match firms within the same industry.
NNMatch code:
PSMatch with exact matching:
My two questions:
1) Would my second block of code be exact matching on industry and then using propensity scores to estimate the ATE for all firms?
2) If the above code is correct, would I have to specify industry indicators in my logit model, or is that not advisable? As in:
References:
Linden, A. (2014). Combining propensity score‐based stratification and weighting to improve causal inference in the evaluation of health care interventions. Journal of evaluation in clinical practice, 20(6), 1065-1071.
Linden, A. (2014). mmws: Stata module for implementing mean marginal weighting through stratification. http://ideas.repec.org/c/boc/bocode/s457886.html
I have run across this question several times throughout the forum, and have also wondered it myself. Basically, people have questioned if there is a way to implement propensity score matching with exact matches on a variable using -teffects psmatch- as with -teffects nnmatch-. I recently came across Ariel Linden's -mmws- file and wondered whether or not this could work out. My question is therefore conceptual.
Assume we would like to exact-match firms within the same industry.
NNMatch code:
Code:
teffects nnmatch (outcome x1 x2) (treat), ematch(industry)
Code:
logit treat x1 x2 predict pscore, pr mmws treat, pscore(pscore) strata(industry) regress outcome treat [pw = _mmws], robst
1) Would my second block of code be exact matching on industry and then using propensity scores to estimate the ATE for all firms?
2) If the above code is correct, would I have to specify industry indicators in my logit model, or is that not advisable? As in:
Code:
logit treat x1 x2 i.industry
References:
Linden, A. (2014). Combining propensity score‐based stratification and weighting to improve causal inference in the evaluation of health care interventions. Journal of evaluation in clinical practice, 20(6), 1065-1071.
Linden, A. (2014). mmws: Stata module for implementing mean marginal weighting through stratification. http://ideas.repec.org/c/boc/bocode/s457886.html
