[R-sig-ME] Correlation in glmmPQL

[Mariangela Sciandra]

Hi all,

I have a question.
I'm analyzing a dataset using the glmmPQL function including only one
fixed effect (treatment), a random intercept and a within group residual
autocorrelation structure of order 1 (with coefficient Phi).

mod<-glmmPQL(y~1+treatn,random=~1|idnew,family=binomial,data=Toenail,correlation=corAR1())

If I try to write step by step the linear predictor like below...

X.mod<-model.matrix(~1+treatn,data=Toenail)
beta.mod<-as.matrix(fixef(mod))
eta1.mod<-X.mod%*%beta.mod
BLUP.mod<-as.matrix(ranef(mod))
eta2.mod<-Z%*%BLUP.mod
eta.mod<-eta1.mod+eta.mod2

...eta.mod will be qual to fitted(mod,level=1).

i don't understand this result. Why in the linear pedictor eta.mod it
doesn't need to add  Phi*elag, where elag is tha lag 1 within-group
residual?

How does Phi get incorporated into the linear predictor?

Thanks in advance,

Mariangela


*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
Mariangela Sciandra
Assegnista di ricerca in "Statistica"
Dipartimento Scienze Statistiche e Matematiche "Silvio Vianelli"
Viale delle Scienze, Edificio 13
90128, Palermo
Tel:0039-0916626242