Statalist


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

Re: RE: st: gllamm log-likelihood cycle?


From   Claudio Cruz Cazares <Claudio.Cruz@uab.cat>
To   statalist@hsphsun2.harvard.edu
Subject   Re: RE: st: gllamm log-likelihood cycle?
Date   Wed, 29 Oct 2008 16:55:19 +0100

Dear Arne,

Thank you very much for your valuable help.

Greetings,

Claudio



----- Mensaje original -----
De: Arne Risa Hole <arnehole@gmail.com>
Fecha: Miércoles, Octubre 29, 2008 4:00 pm
Asunto: Re: RE: st: gllamm log-likelihood cycle?

> Claudio
> 
> -mixlogit- will let you run the same model as your -gllamm- model, the
> only difference is that -mixlogit- uses simulation rather than
> quadrature to approximate the log likelihood function. As pointed out
> by Jay quadrature can be very slow when the dimension of the integral
> is high (and unfortunately 3 is already 'high'). With your problem
> -mixlogit- is likely to be relatively slow too, however, as you've got
> a fairly large sample, but should be considerably faster than
> -gllamm-.
> 
> I have a paper describing the -mixlogit- command which was published
> in the Stata Journal last year (7-3). The second example in the paper
> is very similar to your model. If you don't have access to the SJ a
> pre-publication draft is available here:
> <http://www-users.york.ac.uk/~ah522/mixlogit.pdf>
> 
> Arne (author of -mixlogit-)
> 
> On 29/10/2008, Stas Kolenikov <skolenik@gmail.com> wrote:
> > On 10/29/08, Claudio Cruz Cazares <Claudio.Cruz@uab.cat> wrote:
> > >  When using mprobit with the cluster option if enough for 
> controlling for the unobservable hetereogenity in the panel? Is 
> the unobservable hetereogenity and correlation between 
> observations are controlled in mixlogit since the "group" and "id" 
> options are specified?
> >
> > No, all the -cluster- options do is they correct the estimating
> > equations for possible correlations between individuals. If you want
> > to take those into account more accurately you need to come up with
> > more complicated models. Think of the difference between linear
> > regression and panel data models: you can do -cluster- with regress,
> > but that will still give you biased point estimates if you have 
> random> effects correlated with regressors. Same story here.
> >
> > --
> > 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/
> >
> *
> *   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/
> 


*
*   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   |   What's new   |   Site index