you might want to take a look at hurdle model, since hurdle model is able to take care of both over- and under-dispersion.

On Wed, Sep 30, 2009 at 3:33 PM, Shawn Haskell <shawn.haskell@state.vt.us> wrote: > On Sep 29, 8:00 pm, liuwen...@GMAIL.COM (Wensui Liu) wrote: >> Derek, >> You might want to take a look at my sugi paper in 2008. In my paper, I >> provided several statistical tests for over-dispersion. >> >> wensui >> >> >> >> >> >> On Tue, Sep 29, 2009 at 10:40 AM, Reed, Derek <derek.r...@sinclair.edu> wrote: >> > Morning Folks, >> >> > In performing a Poisson regression using GENMOD, does anyone have a >> > metric for how far is too far away from one for the scale parameter? I >> > am trying to decide when the data are too over dispersed. When I re-run >> > an analysis using the negative binomial distribution my 1.26 scale >> > parameter becomes .80. Any help/guidance would be appreciated. Many >> > thanks! >> >> > Derek >> >> -- >> ============================== >> WenSui Liu >> Blog : statcompute.spaces.live.com >> Tough Times Never Last. But Tough People Do. - Robert Schuller >> ==============================- Hide quoted text - >> >> - Show quoted text - > > very good - thanks for that WenSui. What's the latest on under- > dispersion if you can help there too? I have counts (response > variable) of 0-4. It is not a ZIP situation, and deviance/df < 0.5. > The scale parameter being <1 artificially reduces Type-3 SEs and p- > values. So, I switched to an ordinal logistic model that seems to fit > the data much better according to AICc. Any other thoughts? thanks. > Shawn >

-- ============================== WenSui Liu Blog : statcompute.spaces.live.com Tough Times Never Last. But Tough People Do. - Robert Schuller ==============================