Heckman model and logistic regression to address selection Bias

[Shawna Bayerman]

Hi there,

I am working through a selection bias in my dissertation. For context, my independent and dependent variables are dichotomous. I examine the use of a dietary tool between Mexican Americans and Whites. The use of the tool is contingent upon awareness of the tool (hence the selection bias). It seems that the heckman model only runs if using a regression but my analysis requires a logistic regression. Here is the code that I was playing with if it is helpful. Is there a way to run a heckman model with a logistic regression or is a different method of addressing selection issues more appropriate? I also welcome article suggestions.

*****Estimating Heckman Selection Models****; 

* In my case MyPlate use is only observed for ppl who are aware of MyPlate; 

sum myplateaware myplateuse if inanalysis; 

***probability of entering the first stage of the sample (being aware of MyPlate) based on certain characteristics like income and education; 
probit myplateaware male insurance incomecat education if inanalysis; 

***to generate the probability of entering the second stage of the sample (Using MyPlate) you have to generate an inverse mills ratio; 

predict yhat, xb; 
gen imr = normalden(yhat)/normal(yhat);
label variable imr "Inverse Mills Ratio";

regress myplateuse male insurance foodsecurity incomecat education imr if inanalysis==1; 

****Now I will do the same but with the heckman model*; 

heckman i.mex male insurance foodsecurity incomecat education, select (male insurance foodsecurity wic snap incomecat education) twostep; 

*exclusion restriction - variable that impacts liklihood of entering stage 1 but not stage 2 (i.e. - snap and WIC);

[Andrew Musau]

See

help heckprobit