regression after propensity score matching with exact match?

[Maurice McCourt]

I estimate propensity score weights as follows, where T is the treatment indicator variable, Y is the outcome variable, the covariates X1, X2 and X3 are continuous, while X4 and X5 are discrete (eg year) and must be exact matches

kmatch ps T X1 X2 X3 (Y), ematch(X4 X5) att wgenerate(W)

Note: kmatch is SSC contributed



I then want to reproduce the ATT coefficient using a regression. Should I include the exact match covariates or not: ie

regress Y T X1 X2 X3 [pweight=W]

or

regress Y T X1 X2 X3 X4 X5 [pweight=W]


thanks

[George Ford]

All of them gets you closer.

clear
webuse cattaneo2, clear

kmatch ps mbsmoke mage fage prenatal1  (bweight), ematch(mmarried fbaby) att wgenerate(W) replace
reg bweight mbsmoke mage fage prenatal1 mmarried fbaby [pweight=W]
reg bweight mbsmoke mage fage prenatal1 [pweight=W]

[George Ford]

This is the model that produces the ATT from ksmatch

reg bweight mbsmoke [pweight=W]

just Y and treatment.

[Maurice McCourt]

Great that makes sense, thanks!