I am using eprobit in stata 15.1 to run an endogenous treatment model with binary treatment and binary outcome variables. I am trying to replicate the calculation of the ATET from estat teffects, atet using margins. However, the results only match when the treatment is specified as exogenous and not endogenous. Is there a reason it does not match with the endogenous treatment? Which approach is more likely correct using estat teffects or margins?
This is code I run for the exogenous treatment and the ATET calculated from margins
and using estat teffects I calculate the same atet
however when I specify an endogenous treatment and calculate the ATET with margins
it does not match what I get when using the estat teffects command.
Any insight as to why the ATET from margins and from estat teffects would vary when the treatment is endogenous would be greatly appreciated.
This is code I run for the exogenous treatment and the ATET calculated from margins
Code:
eprobit acutehc_all_1yr i.acs##i.cohort_present_opioid $adjust_charlson, extreat(acs, nointeract) vce(robust)
...
margins r(0 1).acs if acs, contrast(effects nowald)
Contrasts of predictive margins
Model VCE : Robust
Expression : Pr(acutehc_all_1yr==1), predict()
------------------------------------------------------------------------------
| Delta-method
| Contrast Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
acs |
(1 vs 0) | -.0208889 .032102 -0.65 0.515 -.0838076 .0420297
------------------------------------------------------------------------------
Code:
estat teffects, atet
Predictive margins Number of obs = 1,024
Subpop. no. obs = 448
------------------------------------------------------------------------------
| Unconditional
| Margin Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
ATET |
acs |
(1 vs 0) | -.0208889 .0321058 -0.65 0.515 -.0838151 .0420372
------------------------------------------------------------------------------
Code:
eprobit acutehc_all_1yr i.acs##i.cohort_present_opioid $adjust_charlson, entreat(acs=discharge_weekend $adjust_suds, nointeract) vce(robust)
...
margins r(0 1).acs if (acs==1), contrast(effects nowald)
Contrasts of predictive margins
Model VCE : Robust
Expression : Pr(acutehc_all_1yr==1), predict()
------------------------------------------------------------------------------
| Delta-method
| Contrast Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
acs |
(1 vs 0) | -.031926 .2499081 -0.13 0.898 -.521737 .4578849
------------------------------------------------------------------------------
Code:
. estat teffects, atet
Predictive margins Number of obs = 1,024
Subpop. no. obs = 448
------------------------------------------------------------------------------
| Unconditional
| Margin Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
ATET |
acs |
(1 vs 0) | .3896779 .2054155 1.90 0.058 -.012929 .7922848
------------------------------------------------------------------------------
Comment