[R] Overdispersion in count data
info at aghmed.fsnet.co.uk
Thu Apr 3 14:30:58 CEST 2008
At 12:54 03/04/2008, Wade Wall wrote:
>That is exactly how I am writing it. Glm works fine, but as I
>stated the residual deviance is much greater (10x) than the degrees
>of freedom. I want to take a look at using the negative binomial
>distribution, but I can't get glm.nb to work. I get the message
>Error: (subscript) logical subscript too long. I have used
>traceback() and it seems to be in the glm.fitter function, but as I
>say I am at the limit of my abilities here.
For Poisson models and for the negative binomial you have a single
outcome, a count.
For the binomial you can have two columns of counts of successes and
failures (there are other ways of arranging your data).
I think you might want to try the beta-binomial which is available I
think in aod.
However I still think reading the relevant section of MASS first
would be a good idea (or some equivalent text).
>On Thu, Apr 3, 2008 at 7:23 AM, Michael Dewey
><<mailto:info at aghmed.fsnet.co.uk>info at aghmed.fsnet.co.uk> wrote:
>At 17:03 02/04/2008, Wade Wall wrote:
>I have count data (number of flowering individuals plus total number of
>individuals) across 24 sites and 3 treatments (time since last burn).
>Following recommendations in the R Book, I used a glm with the model y~
>burn, with y being two columns (flowering, not flowering) and burn the time
>(category) since burn. However, the residual deviance is roughly 10 times
>the number of degrees of freedom, and using the quasibinomial distribution
>doesn't change this. Any suggestions as to why the quasibinomial
>distribution doesn't change the residual deviance and how I should proceed.
>I know that this level of residual deviance is unacceptable, but not sure is
>transformations are in order.
>You have received much helpful advice from Gavin and Achim and
>others but I wonder whether they are answering the quaestion in your
>title rather than in your post.
>Are you doing something like
>fit <- glm(cbind(flower, notflower) ~ burn, family = binomial)
>You might find it helpful to read the relevant section in MASS (see
>quasibinomial in the index) or in some other text.
>Needless to say that I am at the outer limits of my statistical knowledge.
>Thanks for any help,
> [[alternative HTML version deleted]]
More information about the R-help