Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: significance of model improvement using log pseudolikelihood?


From   Stas Kolenikov <[email protected]>
To   [email protected]
Subject   Re: st: significance of model improvement using log pseudolikelihood?
Date   Wed, 28 Dec 2005 18:12:53 -0600

The issue here is that this test is not always straightforward. For
instance one might want to compare tobit to linear regression with no
corner solutions, and that sort of a test may be rather intricate and
involve non-linear constraints. Of course you can do -nltest- to get
Wald test out of that, but you would need to do quite a bit of
analytical work figuring out what those constraints are, and there are
complications due to nonlinear character that were described in one of
the early SJ's. If all you need is to test a significance of a few
coefficients, then go ahead with the Wald test; if you want to compare
tobit to Heckman, or something of that kind... well that's still
doable along the above lines. -lrtest- assumes nice convergence to the
chi^2; the test statistic following pseudolikelihood estimation does
converge to a distribution of a quadratic form, but my guess will be
that it will be a mixture of chi^2's with different weights, and its
distribution will be more involved than a chi^2 with a given number of
degrees of freedom.

On 12/28/05, Maarten buis <[email protected]> wrote:
> Dear Bart:
>
> Testing the improvement of fit between two nested models is
> equivalent to testing one or more constraints. Say I have the following nested models:
> Model 1: y = a0 + a1*x1
> Model 2: y = b0 + b1*x1 + b2*x2 + b3*x3
> Model 2 is exactly the same as model 1 if b2 and b3 are zero. In which case both x2 and x3 added
> nothing to the model and could have been left out. In other words model 2 is not an improvement
> over model 1 if we cannot reject the constraint that b2 and b3 are zero.
>
> Sometimes the constraints are more complicated, or involve the shape parameter (e.g. some -streg-
> models are nested that way) but the principle remains the same.
>
> So the advise given to you by the Stata FAQ also applies to testing improvements of fit.
>
> HTH,
> Maarten
>
> -----------------------------------------
> Maarten L. Buis
> Department of Social Research Methodology
> Vrije Universiteit Amsterdam
> Boelelaan 1081
> 1081 HV Amsterdam
> The Netherlands
>
> visiting adress:
> Buitenveldertselaan 3 (Metropolitan), room Z214
>
> +31 20 5986715
>
> http://home.fsw.vu.nl/m.buis/
> -----------------------------------------
>
> Bart Vanneste wrote:
> > How can I test the significance of model improvement if only
> > the log pseudolikelihood is given?
> >
> > Can I still use LRtest?
>
> <snip>
>
> > On the STATA website, it is argued that lrtest is not valid
> > when using pseudolikelihoods. The argument seems to focus on
> > testing of a group of coefficients, and it's not clear whether
> > the lrtest is also invalid when assessing (increases in) model
> > fit. See:
> > http://www.stata.com/support/faqs/stat/lrtest.html
>
>
>
> ___________________________________________________________
> To help you stay safe and secure online, we've developed the all new Yahoo! Security Centre. http://uk.security.yahoo.com
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>


--
Stas Kolenikov
http://stas.kolenikov.name

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index