[R] New lmer: How to recode random effect ?

[jerome lemaitre]

Dear all,

I ran the following model without any problem previously to the update of
lme4:

fm2<-lmer(data=NGud,family=poisson,
	seed~hab*seedtray
	+(1|site)+(1|site:hab))

 
I have 25 sites, 2 habitats ("hab") per site, 8 seedtrays per habitat (4 as
control, and 4 as treatment), and I'm interested in comparing the number of
seed in seedtrays as a function of the treatment and its interaction with
habitat.

My problem is that I have just updated lme4 with the version posted in the
last days. Running the same model implies that R crashes rather than
printing an error message. I read the post of Douglas Bates in the "what
news" section but I don't know what is the new appropriate formulation for
this model. 

Any help would be very appreciated.

Many thanks by advance.

PS: I tested 

fm2<-lmer(data=NGud,family=poisson,
	seed~hab*seedtray
	+(1|site)+(1|site/hab))

which also made R crashing.

And,

fm2<-lmer(data=NGud,family=poisson,
	seed~hab*seedtray
	+(1|site/hab))

and for this, I obtained :

CHOLMOD warning: matrix not positive definite
Error in devLaplace(PQLpars) : Cholmod error `matrix not positive definite'
at file:../Supernodal/t_cholmod_super_numeric.c, line 614




Jérôme Lemaître


Étudiant au doctorat
Chaire de recherche industrielle CRSNG - Université Laval en sylviculture et
faune
& Département de biologie,
Faculté des sciences et de génie
Pavillon Alexandre-Vachon
Université Laval
Québec, QC  G1K 7P4
tél : (418) 656-2131 poste 2917
Local : VCH-2044
Courriel: jerome.lemaitre.1 at ulaval.ca