Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Heckman model with gllamm and ordinal/multinomial response


From   Stas Kolenikov <skolenik@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Heckman model with gllamm and ordinal/multinomial response
Date   Thu, 10 Nov 2011 10:14:30 -0500

Paul Silcocks asked about tweaking -gllamm- to work with a
Heckman-type model... well, what matters is that he has several
equations in -gllamm- that relate to different response variables:

On Thu, Nov 10, 2011 at 7:53 AM, Silcocks, Paul wrote:
>
> Hi - I'm trying to run a Heckman selection type of model where the response variable may be ordinal (or even multinomial, though I've not got that far yet).  As a test data set I've used the data set womenwk as in the example given by Sophia Rabe-Hesketh "Multilevel selection models using gllamm"  Stata User Group Meeting, Maastricht, May 2002 - but converting wage into 4 ordered categories.
>
> The problem is that the syntax for the linear predictor in the example specifies "no constant" (as I understand it, this is because the intercepts are already specified by the command eq load: i1 i2.  However the ordinal logit model for the response (and I assume the multinomial one too)  will not permit a "no constant" option.

The original motivation for specifying -noconst- option must have been
that the intercepts of separate equations are specified via dummies.
So if -i1- and -i2- are the dummies corresponding to the two equations
of the model, then they will act as the intercepts of these equations
when they are entered into the linear predictor part (between the
response variable and the options in -gllamm- syntax). The -eq()-
option relates to the random effects/latent variables rather than the
linear predictor: it says that the random effect enters both
equations, that's all. In ordinal models (depending on the
specification, of course), the intercept is excessive to begin with if
you are estimating all the thresholds. So you need to figure out which
equations need the intercepts, enter these intercepts as dummies of
these equations in the linear predictor part, and still specify the
-noconstant- option, as the overall constant is  meaningless in this
situation.


--
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index