Hi there,
I've been wanting to fit a GLMM and I'm not completely sure I'm
doing
things right. As I said in a previous message my response variable is
continuous with many zeros, so I was having a hard time finding an
appropriate error distribution. I read some previous help mails given to
other people advising them to use the Tweedie distribution. I'm still
not sure if this would be appropriate for my data set, for I'm a
beginner and really don't follow all the details. So I ran a GLMM using
this distribution. I ran it for several models to do later model
selection with AIC. I used the following script, where the file
"GLMM_tweedie" (line 2) has a list of all the models I want to run,
each
one in the form [ x=lmer(GGgiv ~ Rank_1 + Rank_diff + DAI +
Gen_dy*Rank_diff + Gen_dy*DAI + Gen_dy + (1| D_1) + (1| D_2), family =
tweedie(var.power=1,link.power=0), offset=log(Dt), data=data) ]
data=read.csv(file="GLMM_data.csv")
> models<-read.table("GLMM_tweedie.txt", sep="\t")
> data$Ggrec_Dtlog = log(data$Ggrec_Dt+1)
> models<-as.vector(models[,1])
> totres=c()
> for (i in 1:79) {model=models[i]
+ res=eval(parse(text=model))
+ res=AIC(logLik(x))
+ res=as.vector(res)
+ totres=rbind(totres,res)}
The output would then be just a list of all the AIC of each model. For 1
of the models (the one in the [] above) I'm getting the following error
message, which I don't know what it means:
CHOLMOD warning: matrix not positive definite
Error in objective(.par, ...) : Cholmod error `matrix not positive
definite' at file:../Supernodal/t_cholmod_super_numeric.c, line 614
Could anybody give me some advice on using Tweedie distributions and
does anybody have an idea what this error message means.
Thanks a lot in advance,
Cheers,
Cristina.
