[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Robust error variance and model selection

Subject   Re: st: Robust error variance and model selection
Date   Sun, 10 Feb 2008 19:57:30 +0100

No more serious answer that "Just believe Stata!". The output of -estat gof- is the same irrespective of the -robust- option.


At 02.33 08/02/2008 -0500, "Romain CASEY" wrote:
>Dear all,
>I want to estimate a relative risk in the same way that it is presented here:
><>. When I
>tried a log-binomial regression, some convergence problems emerged so
>I prefer to use a Poisson regression with a robust error variance.
>With this kind of estimation of the variance, no likelihood is
>calculated but only pseudo-likelihood. So, I don't know which methods
>to use for making a model selection : LRT, AIC, etc. are not usable.
>Moreover, how to test goodness-of-fit of the different models ? If I
>use the command "poisson y x, irr robust" instead of the command "glm
>y x, fam(poisson) link(log) robust", I can use the "estat gof,
>pearson" command but I'm note sure that this test can be apply when a
>robust variance is used.
>Thanks a lot for your help.
>Kind regards,
>Romain Casey 

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index