[R] Error when trying to update cpglm model

[Peter Ehlers]

On 2012-06-19 08:31, Rubem Kaipper Ceratti wrote:
> Dear all,
>
> I've been having problems running update() to re-fit a cpglm model inside a function (as in the code below). The solution is probably simple, but I'm stuck. If anyone could help, I'd greatly appreciate it.
>
> Regards,
> Rubem
>
>
>
>
> ## R code
> library(cplm)
>
> ## Data simulation
> period<-factor(1:4)
> herd<-factor(1:50)
>
> dat<-expand.grid(period=period,herd=herd)
>
> beta<-c(-.3,1.7,2.5,3.4)
> X<-model.matrix(~period,dat)
>
> mu<-as.vector(exp(X%*%beta))
> phi<-1; p<-1.6
>
> dat$resp<-rtweedie(nrow(dat),p,mu,phi)
>
>
> ## cpglm model
> modPC1<-cpglm(resp~period,data=dat)
> head(modPC1 at model.frame)
>
>
> ## Simulating new response vector and re-fitting the model
> sim.hnp<-function(glmfit){
>    dat.1<-glmfit at model.frame
>    n<-nrow(dat.1)
>    dat.1[,1]<-rtweedie(n,glmfit$p,fitted(glmfit),glmfit$phi)
>
>    mfun<-update(object=glmfit,data=dat.1)
>
>    # rp<-resid(mfun,type='pearson')
>    # sort(abs(rp))
> }
>
> sim.hnp(modPC1) # Error in is.data.frame(data) : object 'dat.1' not found

Seems to be an environment problem. A quick fix might be
to insert

   assign("dat.1", dat.1, envir=.GlobalEnv)

before your call to update(), and

   rm(dat.1, envir=.GlobalEnv)

after the call. Presumably, you'll also ensure to return
something from the function.


Peter Ehlers