The eprobit example 4b in STATA EXTENDED REGRESSION MODELS REFERENCE MANUAL RELEASE 19, p.250 estimates a probit regression with endogenous treatment and sample selection.
In the example the main binary outcome is whether a second heart “attack” is had by a sample of 625 men who have had a prior heart attack. Sample selection is present due to participants not being observed for the “full” observation period. An additional endogenous treatment variable in the main equation is whether the participant joined a gym - “exercise”. However, in the example it states “We did, however, observe all cases where a second heart attack was fatal”. Hence, the example actually has observations for all variables for the original 625 sample whereas typically if there is sample selection we wouldn’t observe “attack” and “exercise”?
I want to estimate a model equivalent to the example but where :
Can the above be estimated?
My actual example is using a sample of single people. When observed 5-years later do those who have partnered (selection) own a property (outcome). The endogenous treatment variable is whether they are married.
Thanks
In the example the main binary outcome is whether a second heart “attack” is had by a sample of 625 men who have had a prior heart attack. Sample selection is present due to participants not being observed for the “full” observation period. An additional endogenous treatment variable in the main equation is whether the participant joined a gym - “exercise”. However, in the example it states “We did, however, observe all cases where a second heart attack was fatal”. Hence, the example actually has observations for all variables for the original 625 sample whereas typically if there is sample selection we wouldn’t observe “attack” and “exercise”?
I want to estimate a model equivalent to the example but where :
- “attack” is not actually observed for those with “full”==0 (i.e., there is not the convenient case of the example that they still know “attack” for those who aren’t observed at the end or “full” ==0)
- The endogenous treatment “exercise” is also only observed for those who experience the second heart attack (“attack”==1).
- replace attack=. if full==0
- replace exercise=. if full==0
Can the above be estimated?
My actual example is using a sample of single people. When observed 5-years later do those who have partnered (selection) own a property (outcome). The endogenous treatment variable is whether they are married.
Thanks
