[Limdep Nlogit List] bivariate probit

William Greene wgreene at stern.nyu.edu
Mon Nov 5 04:38:42 EST 2007

I agree with Fred.  The .99+ almost surely indicates some kind of model
misspecification.  The bivariate probit with selection is using a pretty
accurate quadrature method for the bivariate normal (a routine written
by Jerry Hausman), but you can't adjust the number of quadrature points.
I would look at the variables in the two equations. There is a problem 
variable in there somewhere.
/Bill Greene

----- Original Message -----
From: "Fred Feinberg" <feinf at umich.edu>
To: "Limdep and Nlogit Mailing List" <limdep at limdep.itls.usyd.edu.au>
Sent: Sunday, November 4, 2007 11:38:20 AM (GMT-0500) America/Bogota
Subject: Re: [Limdep Nlogit List] bivariate probit

You *could* report the results, but probably shouldn't. Rho up near 0.99 almost
certainly indicated one of three things: (1) an important omitted variable; (2)
a model misspecification; (3) a flat-out error.

It is very unlikely that the fit at 0.99 is a true maximum. It may be a holdover
from the estimation's approximation method, which I believe is either some form
of quadrature or quasi-Monte-Carlo. Those methods work well IF enough estimation
points are used. Have you tried using a larger value? You may discover that your
0.99 value changes. [See the documentation for how to do this.]

A telltale sign is that the *substantive* results -- particularly the size of
Rho -- change radically when any variables are subtracted. It's actually more
common to have Rho go up when you leave out a variable, because that variable is
now unavailable to 'remove' its effects from the model's two equations, whose
error would then be more correlated. Of course, taking out a variable can
increase the overall error in the model, so Rho can go down, too; but this is
less common.

I'm curiour what Prof. Greene might say, but my own experience with Heckman-type
models indicates that this particular solution probably isn't an accurate
representation of what's going on in your data.



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

"Dr. Anat Tchetchik" wrote:

> We are running via  NLOGIT a bivariate probit with selection (which  uses
> the atanh(rho) during estimation) and keep receiving very high and
> significant Rho (0.99 but never 1), The model results, other than that, are
> reasonable and fit our expectations.  when we omit variables from the model,
> rho decreases substantially and become insignificant.  Can we report our
> results with such rho?
> Thanks,
> 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

Professor William Greene
Department of Economics
Stern School of Business
New York University
44 West 4th St., Rm. 7-78
New York, NY   10012

More information about the Limdep mailing list