s-news
[Top] [All Lists]

underdispersion in Poisson glm

To: s-news@wubios.wustl.edu
Subject: underdispersion in Poisson glm
From: Henrik Parn <parn@nt.ntnu.no>
Date: Mon, 31 Jan 2005 13:04:25 +0100
Reply-to: parn@nt.ntnu.no
User-agent: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.6) Gecko/20040113
Dear s-plus users,

I run a glm( , family=poisson). The response variable is number of eggs in a bird's nest (1-7). The explanatory variables in the full model are two continous variables, and two factors with two levels each.


In the summary.glm of the models I see that the residual deviance is far from equal to the residual degree of freedom (residual dev. 12.70, 151 df). If I use the 'common way' to discover underdispersion, this seems to be a serious case. From the help and MASS I understood that I might specify the dispersion in 'summary.glm(object, dispersion=' and it is refered to equation (7.8) on p. 187. in MASS.

So now I wonder:

Lets assume that I really have an underdispersion of a factor around 0.1. How serious is it? Does this mean that I have broken all possible assumption of glm(...family=poisson) and should specify models in a completely different way?

Can I correct for it by calculating the scale parameter obtained from equation 7.8 and put this value as the dispersion parameter in summary.glm? Are there any 'short cuts' in s plus to obtain the scale directly or do I need to calculat it from eg. 7.8? If, so I am not entirely sure of from where I extract the necessary terms in the equation 7.8 so I better ask for all of them...
The residuals: is this the residuals$glm.object?
V(ûi): is this the same as var(ûi)?
Ai: 'is a /known/ prior weight' but where do I find it?
Sorry for my ignorance on glm...

Thanks a lot in advance for any kind of help!

Sincerely,

Henrik

--
************************
Henrik Pärn
Department of Biology
NTNU
7491 Trondheim
Norway

+47 735 96282 (office)
+47 909 89 255 (mobile)
+47 735 96100 (fax)
************************


<Prev in Thread] Current Thread [Next in Thread>
  • underdispersion in Poisson glm, Henrik Parn <=