[R-sig-ME] Boosting computation time of glmmPQL when specifying spatial correlation structure

[alexandre villers]

Good afternoon,

I know how to speed up GAMM when specifying a spatial correlation structure by splitting up the dataset to compute the spatial correlation coefficients of corSpher.
such (if dataG is my dataset)
cutx<-cut(dataG$x,breaks=(4))                          cuty <- cut(dataG$y, breaks=(4))
cutxy <- paste(cutx, cuty)

and then 
gamm(Response~(var1)+s(var2),family=binomial, data=dataG,correlation=corSpher(form=~(x+y)|cutxy)).

the cutxy doesn't seem to work with glmmPQL and with 1500 points, it takes ages... 
Does anyone know if there is a way to apply the same "trick" ?

By the way (take a breath...), does the plotting of a spatial correlogram with residuals(model, type="pearson") from a glmmPQL model (where correlation structure was specified) makes sense to you ? I'm not sure if residuals of such model account for the stucture (and I can hear some of you, why don't you check this by yourself... yes , I will try !)

Best regards and thanks for any hint

Alex

Alexandre Villers
PhD. Candidate
Team Agripop
CEBC CNRS UPR 1934
79360 Beauvoir sur Niort

Phone: +33 (0)549 099 613 


__________ Information from ESET Mail Security, version of virus signature database 4777 (20100116) __________

The message was checked by ESET Mail Security.
http://www.eset.com