The question was do we test for outliers
"as we do when we run OLS". My answer
is much nearer "No" than "Yes",
or nearer "Yes" than "No", depending
on what is meant by this.
First, it seems possible that on a statistical list
"test" is being used in its correct statistical
sense. FWIW, I never test for outliers
before using OLS. I would examine the
data for outliers, using graphs and
to some extent summary measures, and then
consider what to do about any apparent
outliers, including
edit them out: they must be gross
data errors (subject-matter
expertise is needed here)
transform
use an appropriate link function (GLM etc.)
don't worry (yet)
This is testing in the sense of checking;
perhaps that is what is meant.
Second, the case of count models is perhaps
a little more complicated in that typically
with the distribution(s) entertained,
considerable right skewness and specifically
a few large counts may well be expected.
But the same broad advice applies, I
would have thought.
In either case, it is helpful to realise
that outliers on the marginal distribution
of the response could well not seem
that when placed in their appropriate
conditional distribution.
Nick
[email protected]
[email protected]
> Yes. Take a look at:
>
> Hardin & Hilbe (2003) Generalized Liner Models and
> Extensions, Stata Press
>
> Cameron & Trivedi (1998) Regression Analysis of Count Data,
> Cambridge
> University Press.
[email protected]
> I am new to STATA and Count models. I plan to test
> different count models. Before I do that I would like
> to know if we test for outliers in the dependent
> variable as we do when we run OLS.
*
* 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/
