The plots look reasonable to me. The plot of residuals against linear
predictor always looks scary when many of the fitted values are very
close to zero, so I tend to look at residuals against sqrt(fitted) in
such cases. I don't think that the presence of the zero curve is a
reason to reject the model --- it's easy to produce such plots by
fitting a completely correct model to simulated count data.
best,
Simon
On 08/06/11 15:50, Samuel Turgeon wrote:> Dear list,
>
> i'm checking the residuals plots of a gam model after a processus of
model
> selection. I found the "best" model, all my terms are
significant, the
> r-square and the deviance explained are good, but I have strange residuals
> plots:
>
> http://dl.dropbox.com/u/1169100/gam.check.png
> http://dl.dropbox.com/u/1169100/residuals_vs_fitted.png
>
> The curve is caused by the zeroes in my data.
>
> I've also plotted each explanatory variables included in the model
> against residuals
> and everything looks fine.
>
> Is this curve does not allow me to accept this model?
>
> Does the use of an other family (eg negbin) would be the solution for
fixing
> this problem? Currently I use the poisson family (and quasipoisson). I
have
> a lot of 0 in my response variable, almost 65 %.... Should I use a specific
> function that allows me to use zero-inflated data??
>
> Kind regards,
>
> Sam
>
> [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
--
Simon Wood, Mathematical Science, University of Bath BA2 7AY UK
+44 (0)1225 386603 http://people.bath.ac.uk/sw283
