I don't know about the scaled decianve vs pearson deviance comparison, but
I don't think one can just reject a model because none of the explanatory
variables are significant. It may be that in the dataset you have none of
the explanatory variables are really explanatory. Rejecting a model just
because none of the coefficients are discernibly different from zero,
absent some strong theory, would seem to vitiate the value of statistical
analysis to reveal things we don't already know. But, I don't know your
substantive area, so maybe there is some strong theory to guide you.
For what it's worth (hopefully, more than you are paying to get it).
Sam
On Tue, 20 Aug 2002, Claire wrote:
> Hello,
>
> I have a large database of count data, 95% zeros & overdispersion (variance
> 25 x larger than the mean) - I have tried poisson, nbreg, zinb & zip with
> the Vuong test. The problem is that the results for the zinb, which the
> Vuong test shows is preferable over the nbreg, are poor in that very few
> explanatory variables are significant. Is there any other way to account
> for excess zeros/overdispersion using nbreg rather than zinb? The reason I
> can't accept my current nbreg model is that the scaled deviance is 1 but the
> pearson deviance is 212 - I assume that this is not acceptable?
>
> Thanks
>
> Claire
>
> *
> * 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/
>
*
* 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/
