--- [email protected] wrote:
> Dear Listers,
>
> I have a somewhat technical question.
>
> At the UCLA Stata FAQ site, it can be read that Hendrickx's
> DESMAT be used to generate a saturated model for
> Agresti's afterlife data.
>
> (see
> http://www.ats.ucla.edu/stat/stata/faq/desmat_saturated_Model.htm)
>
> However, isn't this inappropriate when fitting a saturated model
> for a contingency table? Using DESMAT POISSON would indicate
> that count is the dependent variable with error terms poisson
> distributed.
The faq is referring to a saturated loglinear model. A loglinear
model is applied to contingency table data. It assumes the
frequencies have either a poisson or a multinomial distribution.
>
> In addition to that, the output shows "model degrees of
> freedom=11",
> however
> the saturated model has 0 df by definition as it fits the data
> perfectly.
A saturated model has 0 residual degrees of freedom. If you run:
desmat: glm count race*gender*belief, family(poisson) link(log)
desrep(exp all)
This produces the same estimates but addition model statistics. The
value for deviance is 0, as are the residual degrees of freedom.
There are 11 model degrees of freedom are equal to the number of
predictors (excluding the intercept). There are 11 dummy variables,
so 11 model df.
>
> Maybe I am so wrong or stupid or both that I don't get it, but I'd
> really
> appreciate your comments.
>
You're just wrong.
John Hendrickx
_______________________________
Do you Yahoo!?
Shop for Back-to-School deals on Yahoo! Shopping.
http://shopping.yahoo.com/backtoschool
*
* 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/
