[Kolo Yeo]

Good morning all,
I estimate a probit with sample selection using the heckprobit command.
how to correct selection bias with heckprobit?

[Nishant Kathuria]

Hi Jeff Wooldridge ,

It would be very helpful if you could comment on the following queries.

In certain context, I want to examine if firms with more diversity (continuous var) are more likely to select a particular type of training (a binary variable training_type1). Now, around 37% of the firms in my sample (total # firm-year obs=12000 for 4 years) do not give (any) training at all (binary var. training_done=1/0). Also, the firms are treated for a given effect such that the treated firms are more likely to provide training (correlation b/w training_done and treatment =0.41). I hypothesize that the greater the diversity, the lesser the likelihood of choosing training_type1. I am confused if heckprobit model is an appropriate model to test the hypothesis. If not, please suggest me one. If yes,

1) Is it appropriate to choose training_done (=1 if training provided) as the selection variable? After reading Natalie G's query, I think it won't make sense, but please correct me if I am wrong.

2) Although I am interested in determining the effect of treatment on the choice of training_type1, the #firms that offer training_type1 are very few. # treated firms that offer training_type1=93 and # non-treated control firms that offer training_type1=11 . If I restrict the sample that that offers training (training_done==1), does it still make sense that I use the select variable as the treatment effect (treated=1/0)?

3) Can I use the treated variable (1/0) in both main equation and selection equation? Would that be forbidden regression?

The following are the models that I formed:
Model 1 xi:heckprobit training_type1 diversity treated var1 var2 i.fyear i.nic, select(training_done= diversitytreated var1 var2 excl_var) vce(robust) first

Model 2 xi:heckprobit training_type1 diversity var1 var2 i.fyear i.nic if training_done==1, select(treated= diversity var1 var2 excl_var) vce(robust) first

where var1 and var2 are control vars, and excl_var is the exclusion variable.

Thank you so much for your time.