I try to estimate 3 consecutive models using the "diff" command in Stata:
1. Diff model:
Code:
diff `var', period(treatperiod) treated(treatment) robust
Code:
diff `var', period(treatperiod) treated(treatment) cov(`indepvars') robust
Code:
diff `var', period(treatperiod) treated(treatment) cov(`indepvars') robust kernel id(idnum) support
My outcomes in the first model (which does not include other covariates) produce significant results. When I estimate it with the second model (including the covariates), its coefficient significantly reduces. Yet, as the treatment and control observations have certain differences, I use matching in the third model using kernel PSM, including the covariates. Yet, as far as I see, there are only slight (very minor) changes in the coefficients from model 1 to model 3, despite the significant and relatively huge changes from model 1 to model 3. It is also curious that the R-Square results significantly reduce as well.
To give a concrete example:
Model 1: B is 0.65, R2 is 0.04.
Model 2: B is 0.24, R2 is 0.54
Model 3: B is 0.66, R2 is 0.06.
Is it something that I miss in the Stata code, or am I missing something about these estimations?
Following is the key variables, if it is relevant:
Code:
* Example generated by -dataex-. To install: ssc install dataex clear input float idnum byte panelwave float treatperiod byte treatment float did 10009 1 0 . . 10009 2 1 0 0 10009 3 1 0 0 10010 1 0 0 0 10010 2 1 0 0 10010 3 1 0 0 10011 1 0 0 0 10011 2 1 1 1 10011 3 1 1 1 10012 1 0 1 0 10012 2 1 0 0 10012 3 1 0 0 10013 1 0 0 0 10013 2 1 0 0 10013 3 1 0 0 10014 1 0 0 0 10014 2 1 0 0 10014 3 1 0 0 10015 1 0 0 0 10015 2 1 0 0 10015 3 1 0 0 10018 1 0 0 0 10018 2 1 0 0 10018 3 1 0 0 10019 1 0 0 0 10019 2 1 0 0 10019 3 1 0 0 10038 1 0 0 0 10038 2 1 1 1 10038 3 1 1 1 10051 1 0 1 0 10051 2 1 0 0 10051 3 1 0 0 10052 1 0 0 0 10052 2 1 0 0 10052 3 1 0 0 10054 1 0 0 0 10054 2 1 0 0 10054 3 1 0 0 10055 1 0 0 0 10055 2 1 0 0 10055 3 1 0 0 10056 1 0 0 0 10056 2 1 0 0 10056 3 1 0 0 10057 1 0 0 0 10057 2 1 0 0 10057 3 1 0 0 10058 1 0 0 0 10058 2 1 0 0 10058 3 1 0 0 10061 1 0 0 0 10061 2 1 0 0 10061 3 1 0 0 10062 1 0 0 0 10062 2 1 0 0 10062 3 1 0 0 10065 1 0 0 0 10065 2 1 0 0 10065 3 1 0 0 10081 1 0 0 0 10081 2 1 0 0 10081 3 1 0 0 10083 1 0 0 0 10083 2 1 0 0 10083 3 1 0 0 10105 1 0 0 0 10105 2 1 0 0 10105 3 1 0 0 10106 1 0 0 0 10106 2 1 0 0 10106 3 1 0 0 10107 1 0 0 0 10107 2 1 0 0 10107 3 1 0 0 10108 1 0 0 0 10108 2 1 0 0 10108 3 1 0 0 10110 1 0 0 0 10110 2 1 0 0 10110 3 1 0 0 10111 1 0 0 0 10111 2 1 0 0 10111 3 1 0 0 10112 1 0 0 0 10112 2 1 0 0 10112 3 1 0 0 10114 1 0 0 0 10114 2 1 0 0 10114 3 1 0 0 10115 1 0 0 0 10115 2 1 0 0 10115 3 1 0 0 10118 1 0 0 0 10118 2 1 1 1 10118 3 1 1 1 10141 1 0 1 0 10141 2 1 0 0 10141 3 1 0 0 10142 1 0 0 0 end
All my best.
0 Response to Diff-in-Diff with Kernel PSM: R-Square and Coefficients
Post a Comment