Hello,
I'm using Stata IC 12 for Windows, and I have created two PSM models using psmatch2 (from SSC), one each for two subgroups of the sample (sample contains 1478 obs, model 1 uses 803, model 2 uses 675). The commands used for each model are:
psmatch2 vouch0 if momsch==1, outcome(prsch_c) pscore(ps3) kernel k(normal) logit
psmatch2 vouch0 if momsch==1, outcome(prsch_c) pscore(ps4) kernel k(normal) logit
The models both produce an output, including the estimated difference in ATT between the treated and controls. I would like to test for a significant difference between the two differences, but I have looked at the help for psmatch2 and it does not include any postestimation commands to do so. When I store the estimates and run suest, it returns
"score variables for model pslow contain missing
r(322);"
Is there a command to do what I'm looking for, and if not do I just do it manually, in which case, is it just as simple as calculating confidence intervals with the standard errors provided for each estimate of the difference between treated and controls, and seeing if they overlap?
Thanks in advance for any help,
Alex Feuchtwanger
I'm using Stata IC 12 for Windows, and I have created two PSM models using psmatch2 (from SSC), one each for two subgroups of the sample (sample contains 1478 obs, model 1 uses 803, model 2 uses 675). The commands used for each model are:
psmatch2 vouch0 if momsch==1, outcome(prsch_c) pscore(ps3) kernel k(normal) logit
psmatch2 vouch0 if momsch==1, outcome(prsch_c) pscore(ps4) kernel k(normal) logit
The models both produce an output, including the estimated difference in ATT between the treated and controls. I would like to test for a significant difference between the two differences, but I have looked at the help for psmatch2 and it does not include any postestimation commands to do so. When I store the estimates and run suest, it returns
"score variables for model pslow contain missing
r(322);"
Is there a command to do what I'm looking for, and if not do I just do it manually, in which case, is it just as simple as calculating confidence intervals with the standard errors provided for each estimate of the difference between treated and controls, and seeing if they overlap?
Thanks in advance for any help,
Alex Feuchtwanger
