[Limdep Nlogit List] Sample selection models with random effects

[William Greene]

Dear Allan:  The random parameters model in LIMDEP 8.0 allows precisely this 
model - it's the bivariate probit model with sample selection (your number 2), 
and allows random parameters and a panel structure.  For your random effects
model, allow only random constant terms in the two equations.  The precise model
you are looking for is on pages E17-24 to E17-27 of the version 8.0 manual, with
an application of the bivariate probit model with random effects on page E17-25.
(It's the BIVARIATE command, and for  your case, you would add ;SELECTION to
the command.)
Regards,
Bill Greene

************************************************
Professor William Greene
Department of Economics
Stern School of Business
New York University
44 West 4th St., Rm. 7-78
New York, NY   10012
Ph. 212.998.0876
Fax. 212.995.4218
URL. http://www.stern.nyu.edu/~wgreene
Email. wgreene at stern.nyu.edu
************************************************

----- Original Message -----
From: "Little, Allan" <a.little at lancaster.ac.uk>
Date: Tuesday, May 23, 2006 8:53 am
Subject: [Limdep Nlogit List] Sample selection models with random effects

> 
> 
> 
> 
> I have a query concerning the estimation of sample selection 
> models in LIMDEP. I'm looking to set up a model with the following 
> features:
> 1.	Binary dependent variables, in both models (i.e. the selector 
> z=0/1 and in the second-step y=0/1, conditional on z=1)
> 2.	Random effects components, in both models 
> 
> 
> 
> I have identified two possible options in LIMDEP.
> 
> 
> 
> 1.	Run 'PROBIT' (in step 1) and HOLD the results, then run a 
> 'SELECT' model (in step 2) 
> 
> ·        This appears to allow for random effects components in 
> both stages
> 
> ·        BUT... SELECT is a linear model (?), and doesn't seem to 
> allow for logit/probit estimation.
> 
> 
> 
> 2.	BIVARIATE PROBIT using the 'Selection' option 
> 
> ·        This does allow for probit estimation in both steps
> 
> ·        BUT... doesn't seem to allow for random effects 
> estimation (?)  
> 
> 
> 
> So my question is, how can I run a Heckman-style probit model, for 
> panel data? Perhaps I'm misinterpreting the limitations of the 
> above options, or maybe there's another way?
> 
> 
> 
> ...Just to give a little more detail on the nature of my problem. 
> The model is a random effects binary logit/probit, based on a 
> panel data sample of unemployed workers. y=0 if individual remains 
> in unemployment, and y=1 if the individual moves back into 
> employment. The selection problem arises because employed workers 
> are precluded from the sample. So I intend to set up the selector 
> model, where z=1 if unemployed, z=0 if employed. Again, I have a 
> panel of data for the estimation of the selector equation.
> 
> 
> 
> Any suggestions would be greatly appreciated.
> 
> 
> 
> Thanks,
> 
> 
> 
> Allan Little
> 
> Lancaster University
> 
> a.little at lancaster.ac.uk
> 
> _______________________________________________
> Limdep site list
> Limdep at limdep.itls.usyd.edu.au
> http://limdep.itls.usyd.edu.au
>