Implementing Common Support in Teffects smatch

[Gary Chapman]

Hi all,

I am attempting to generate treatment effect estimates of a government program. I originally was using psmatch2 command however, I have been unable to find a work-around for estimating the standard errors without bias when matching on the propensity score. If anyone has any ideas for estimating these it would be incredibly helpful?

code: psmatch2 treatment, outcome(variable) pscore(p) caliper(0.05) common

Accordingly, I have been using teffects psmatch (code below), but I have been unable to find a way to implement common support; it implements the calliper restriction however, not common support. Would anyone know if there is a command similar to "common" in psmatch2?

Code:
set seed 12345
gen u=uniform()
sort u
teffects psmatch (outcomevar) (treatment covariates), first generate(nn_) caliper(0.05) atet

Also, teffects psmatch estimates the propensity score and outcome in one, however, this means that each time you drop or exclude observations outside the calliper restriction, it re-estimates the propensity score on the new sub-set. Whereas, psmatch2 estimates the pscore separately and then excludes those not on common support without re-estimating the propensity score. This is meaning I am getting slightly different results for both commands when observations need to be dropped due to them falling outside the 0.05 calliper restriction I'm using. Would anyone know if the propensity score can be estimated separately using the pscore command and then entered into teffects psmatch? or another way of going about this.

Thanks for your help

Gary

[Albert Zhao]

Hi Gary, same question here. Do you have any solution to this problem?

Bert