[R] GAM and Poisson distribution

[Simon Wood]

Sally,

Is Kept.CPUE discrete valued? If it isn't then the poisson family won't like 
it, and it would be better to use quasipoisson, if the variance proportional 
to mean is the right assumption. (mgcv 1.5 includes a Tweedie family, which 
would be an alternative, but it's not ready for release yet).

best,
Simon

On Friday 14 November 2008 21:29, sallyr wrote:
> Hi -I'm running a GAM with 7 explanatory variables with a Poisson error
> structure.  All of the variables are continuous so I'm getting error
> messages in R.
>
> cod.fall.full.gam.model<-gam(Kept.CPUE~s(HOUR)+s(LAT_dec)+s(LONG_dec)+s(mea
>ntemp_C)+s(meandepth_fa)+s(change_depth)+s(seds),
> data=cod.fall.version2,family=poisson)
>
> In dpois(y, mu, log = TRUE) ... : non-integer x = 5.325517
>
> Thanks Sally

-- 
> Simon Wood, Mathematical Sciences, University of Bath, Bath, BA2 7AY UK
> +44 1225 386603  www.maths.bath.ac.uk/~sw283