Statalist


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

Re: st: [Non Stata] Testing nested models with clustering


From   Antoine Terracol <[email protected]>
To   [email protected]
Subject   Re: st: [Non Stata] Testing nested models with clustering
Date   Wed, 17 Jun 2009 09:18:48 +0200

Thanks Stas.

My problem was that, since my extra parameter is constrained in (0,1) (it is the inverse logistic of a parameter in |R), the lower CI limit is always larger than zero.

-streg- with frailty has a similar feature because the variance of the frailty is constrained to be strictly positive via an exponential transformation. The command does not do any test of the null that there is no heterogeneity (variance=0) whith clustered std. errs., but does a LRT without; hence my "fear" that there was no simple way to test such hypotheses with clustered std. errs.

Antoine

Stas Kolenikov wrote:
You can do a one-sided Wald test with the clustered standard errors
just fine. The LRT is indeed a disaster, and if you have clustering
you would probably have a pseudo-likelihood rather than true
likelihood, anyway (although I am not really familiar with frailty
models sufficiently well). This means that the distribution of the LRT
even without a boundary is not a chi-square but a sum of weighted
chi-squares, Satterthwaite-style. If you truncate this to the
boundary, all the hell breaks lose. The Wald test however is quite
simple, clustered standard errors, jackknife standard errors, or
whatever standard errors you trust in your situation. If the left
confidence limit covers zero, you don't reject, and can claim there is
no heterogeneity. I suspect there is one, anyways... otherwise you
won't be doing the more complicated analysis, right? :))

On 6/16/09, Antoine Terracol <[email protected]> wrote:
Dear _all,

 I have estimated two models using -ml-. The two models are nested, and
model B is equivalent to model A if some parameter delta is equal to zero.

 I have clustered standard errors, and thus cannot use a LR test to test B
against A. Moreover, delta is constrained in (0,1) through a logistic
transformation.

 In -heckman- with -cluster()-, the corresponding test is a Wald test of
atanh(rho)=0; but in -streg, frailty()- with -cluster()-, no test is
performed for the null that there is no heterogeneity (without clustering, a
LR test for theta=0 is performed).

 Should I infer  that there is no proper (and simple) way to test for nested
models with clustering when the additional parameter is constrained, and the
null is on the boundary?

 Best,

 Antoine

 --
 Ce message a ete verifie par MailScanner
 pour des virus ou des polluriels et rien de
 suspect n'a ete trouve.

 *
 *   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/




--
Ce message a ete verifie par MailScanner
pour des virus ou des polluriels et rien de
suspect n'a ete trouve.

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