[R-sig-Epi] R-sig-Epi Digest, Vol 41, Issue 1

[David Evans]

Olà Mirian,

Are you using Poisson models with an offset?  I've never tried this  
with the Epi package so giving us some sample code and sample data  
structure would help.

But if you simply use glm, one way to account for over-dispersion is  
by specifying family=quasipoisson.  It maximizes the quasilikelihood  
rather than the likelihood (don't ask me the maths of this, maybe  
start with the Wikipedia...).

Another approach might be hierarchical modelling but we'd need to know  
your data structure.

Cheers.

David Evans
-----------------
Epidemiologist, Kappa Santé


>
> Message: 1
> Date: Wed, 19 May 2010 10:10:58 -0300
> From: "Mirian Carvalho de Souza" <miriancs at inca.gov.br>
> To: <bxc at steno.dk>
> Cc: r-sig-epi at stat.math.ethz.ch, Mirian Gmail <miriancs99 at gmail.com>
> Subject: [R-sig-Epi] Age-period-cohort models for lung cancer
> 	mortality rates
> Message-ID: <E587D8FA5FB8444EBC572547CAFEA18B at inca.local>
> Content-Type: text/plain; charset="iso-8859-1"
>
> GeleiraHello!
> I work at National Cancer Institute, Brazil and I'm interested on  
> fit age-period-cohort models for lung cancer mortality rates.
> I'm using the Epi package, but I have a problem. There is over- 
> dispersion on my data. How can I deal with this problem? Some one  
> can suggest a solution?
> Thanks
> --
> Mirian Carvalho de Souza
> Instituto Nacional de C?ncer/MS