Fractional logit and inverse mills ratio

[Donovan Pollack]

Hello,

I am testing the impact of X on Y, where Y is a percentage variable. Many observations have Y=0. I am using a fractional logit to test X on Y. However, my fractional logit does not account for sample selection, I believe I also need to first run the Heckit in Stata to obtain an inverse mills ratio to include in the estimation of X on Y. I am reluctant because I have seen many papers using fractional logit not include the heckman, which leads me to believe something in my logic is wrong.

Thoughts appreciated,

thank you,
Donovan

[FernandoRios]

Hi Donovan,
Without further information its difficult to assess if you have or not sample selection in your model. Assuming your intuition is correct, and you would like to correct sample selection using a heckman type of correction (ie using a sort of IMR) i would suggest you to look in to the heckprobit model.
In other words, if you can program the Maximum likelihood function for a heckprobit, you could easily modify it to use a fractional probit (rather than logit) in your outcome.
I would imagine that you can also use the command CMP specifying the main outcome to be fractional, but have not use that command for that type of model.
HTH
Fernando