[Limdep Nlogit List] A selection qeustion--a quick follow-up

Dr. Anat Tchetchik manes at agri.huji.ac.il
Thu Oct 18 20:10:18 EST 2007


Thank you very much indeed for your detailed and helpful answer. 

-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au
[mailto:limdep-bounces at limdep.itls.usyd.edu.au] On Behalf Of Fred Feinberg
Sent: Thursday, October 18, 2007 8:46 AM
To: Limdep and Nlogit Mailing List
Subject: Re: [Limdep Nlogit List] A selection qeustion--a quick follow-up

This exact question seems to come up a lot!

Two-step estimators via the Mills Ratio are "very 20 years ago". There are
well-known problems with the entire approach (see the paper by Puhani on the
issue). Quite a bit better is full-information maximum likelihood, which
unless I'm wrong is like what Limdep uses for the ordinary Heckman model and
the Heckman Probit model. But better still is using Bayesian methods. This
is
all covered in a perpetually-unpublished paper by me and two co-authors:


http://webuser.bus.umich.edu/feinf/research/Working_Papers/Feinberg,_Brown,_
Ying_(2004)_-_Choice_Endogeneity.pdf

In that paper, you will find the Puhani cite, and also one to Lee (1983),
who
actually solved this problem classically long ago, as well as a discussion
of
the problem in general. There used to be Limdep code on the web to do this,
and Google may be able to find it for you. We used Bayesian estimation
(actually, a hierarchical Bayesian formulation, which allows for parameter
heterogeneity), and found it to work quite well, if slowly.

Fred

=====

Fred Feinberg
Hallman Fellow and Professor of Management
Stephen M. Ross School of Business
University of Michigan
feinf at umich.edu



> I wish he'd had the sense to say something more scientific, such as that
one
> should not assume equal geographic distribution of genes for any
particular
> human abilities.
>
He didn't.

> Why he needed to say what he did is beyond me, unless he LIKES starting
> trouble (which he seems to).
>
 Yes, and I'm sure he's not the only one.

Christer.Thrane at hil.no wrote:

> Hi,
>
> If the the second "choice" involves three outcomes (and not two) - that
> is, if you have a multinomial regression model with sample selection (and
> only indeps describing the choosers) - how do I estimate this?
>
> My own suggestion thus far is:
>
> 1. Estimate selection equation with probit or logit
>
> 2. Obtain Inverse Mills' Ratio, IMR, from 1.
>
> 3. Add IMR to the list of indpendent variables in the multinomial
> regressinon using robust SE's
>
> Is this reasonable?
>
> BR
>
> Christer
>
> ***********************************************************
> Professor Christer Thrane <christer.thrane at hil.no>
> Lillehammer University College
> Postboks 952
> 2604 Lillehammer
> Norway
> +47 61 28 81 70 (fax)  | +47 61 28 82 47 (phone, work)
> +47 92 29 54 39 (cell) | +47 61 25 53 04 (phone, private)
> ***********************************************************
>
>
>
> William Greene <wgreene at stern.nyu.edu>
> Sent by: limdep-bounces at limdep.itls.usyd.edu.au
> 18.10.2007 02:17
> Please respond to
> Limdep and Nlogit Mailing List <limdep at limdep.itls.usyd.edu.au>
>
> To
> Limdep and Nlogit Mailing List <limdep at limdep.itls.usyd.edu.au>
> cc
>
> Subject
> Re: [Limdep Nlogit List] A selection qeustion
>
> Dear Professor Tchetchik:  The "Bivariate Probit with Selection" model
> that
> you describe below has been available in LIMDEP for many years.  The
> program
> contains a full information maximum likelihood estimator.  You can find
> discussion of the model in my textbook, Econometric Analysis, in the
> chapter on discrete choice models.
> Sincerely,
> William Greene, developer.
>
> --
> Professor William Greene
> Department of Economics
> Stern School of Business
> New York University
> 44 West 4th St., Rm. 7-78
> New York, NY   10012
> http://www.stern.nyu.edu/~wgreene
> 212.998.0876
>
> ----- Original Message -----
> From: "Dr. Anat Tchetchik" <manes at agri.huji.ac.il>
> To: Limdep at limdep.itls.usyd.edu.au
> Sent: Wednesday, October 17, 2007 9:52:04 AM (GMT-0500) America/Bogota
> Subject: [Limdep Nlogit List] A selection qeustion
>
> Hi  All,
>
> We want to run a selection model which resemble the sample selection one
> but
> the second equation is not a regression one, rather it is also a probit (a
> "yes/no" decision that depends on a realization of "yes" on the first
> probit
> equation). Any idea how to handle it? Does Limdep have a code for that?
>
> Thanks in advance for any help
>
> Anat Tchetchik, Ph.D.
>
> The Department of Agricultural Economics
>
> The Hebrew University of Jerusalem
>
> Tel: 08-9489231
>
> cell: 054-4928740
>
> _______________________________________________
> Limdep site list
> Limdep at limdep.itls.usyd.edu.au
> http://limdep.itls.usyd.edu.au
>
> _______________________________________________
> Limdep site list
> Limdep at limdep.itls.usyd.edu.au
> http://limdep.itls.usyd.edu.au
>
> _______________________________________________
> Limdep site list
> Limdep at limdep.itls.usyd.edu.au
> http://limdep.itls.usyd.edu.au



_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au




More information about the Limdep mailing list