Re: st: How to conclude? Poisson Regression: Prob > chi2 = .00000, estat gof prob > chi2(4) = .0164

[Maarten Buis <maartenlbuis@gmail.com>]

On Mon, Jul 11, 2011 at 3:19 AM, Herberto Gomez wrote:
> In a poisson regression how to decide if the model is adequate?
>
> If prob > chi2 from top part of the Poisson regression conflicts with
> estat gof conclusion,
> what should I decide.

The null-hypotheses in these tests are never "this model is adequate"
instead they tend to test a very specific hypothesis, e.g. in the case
of chi2 test at the top of the output the null hypothesis is that all
estimated coefficients except the constant are simultaneously equal to
0. So the first step is to find out what exactly these tests test, and
whether you care about that.

More importantly never use (only) statistical tests to decide whether
a model is adequate. Statistical tests are an extremely useful tool in
our toolbox, but that does not mean it can be applied to all problems.
Remember that models are supposed to be simplifications of reality,
which by necessity means that they are supposed be useful not true. So
to decide whether a model is adequate you (and not "Statistics",
Stata, the computer or some guru) need to make a tradeoff between
simplification and realism. Statistical test are not designed for that
tradeoff.

Some people fear that this is too subjective or non-scientific. The
answer to that is that you can and should document and report how and
why you made these decisions. That way people check what you have done
and see for themselves whether they agree with you, and have a
reasoned debate with you in case they do not. This is what makes a
study scientific, not the fact that you have literally followed the
instructions set out in some statistics "cookbook".

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany


http://www.maartenbuis.nl
--------------------------
*
*   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/