[Limdep Nlogit List] Negative standard errors in selection model?

[Andrea Mannberg]

Dear all,

 

I am estimating a selection model with a bivariate probit in the first step
and an OLS in the second step. The model runs with exit status=0, but at the
bottom of the output  I get the message: “Selectivity corrected std. error
for the b on ONE      is negative.

The set of OLS standard errors is given.”  

 

The message is not printed as an “error” or “warning”, but clearly
something is wrong in the model. However I cannot figure out exactly what is
causing this, or whether or not I should worry sick about it.  I wonder if
it may have something to do with the fact that the bivariate probit is in
itself a selection model in terms of y2 only being observed if y1=1? The
code is given below:

 

BIVARIATE ; Lhs=y2,y1; Rh1=z2; Rh2=z1;selection;hold$

SELECT; Lhs=y3; Rhs=x$

 

 I’m not sure that the bivariate probit selection model supports the case
when there is selection in the first step. Could this be the problem?

 

Has anyone encountered a similar problem? 

 

Thanks,

 

Andréa