[Limdep Nlogit List] Sample Selection probit / mnl

[Diane Dancer]

Hi Group,

I am trying to estimate a sample selection model with the probit as the first stage and a multinomial logit as the second stage.

However, I get an error message saying there is not enough variation in the dependent variable of the MNL model.

I am wondering what I am doing wrong, as I can estimate the probit model by itself, and the MNL by itself. I can also estimate the sample selection model with the probit as the first stage and an ordered probit as the second stage, using the same data.

The y2 variable has 0, 1, 2, 3 and 4, but only 0, 1 and 2 should be present after selection. Note: the ordered probit is only using 0, 1, and 2, as we expected.

The code I am using is:
Probit; LHS = y1; RHS = one, x1,x2, x3, x4; hold results $
MLOGIT: LHS = y2; RHS = one, x1,x2; select $

Does anyone have any ideas what is going on?

JT Toman