[R] GLMM: MEEM error due to dichotomous variables

[lorenz.gygax at art.admin.ch]

> I am trying to run a GLMM on some binomial data. My fixed 
> factors include 2 
> dichotomous variables, day, and distance. When I run the model:
> 
> modelA<-glmmPQL(Leaving~Trial*Day*Dist,random=~1|Indiv,family=
>"binomial")
> 
> I get the error:
> 
> iteration 1
> Error in MEEM(object, conLin, control$niterEM) :
>         Singularity in backsolve at level 0, block 1
> 
> >From looking at previous help topics,( 
> http://tolstoy.newcastle.edu.au/R/help/02a/4473.html)
> I gather this is because of the dichotomous predictor 
> variables - what approach should I take to avoid this problem?

Are you sure? I have never had problems including factors in a glmmPQL so far. More likely, the combination of your explanatory variables leads to a fragmentation in your response such that each combination of your factor levels only contain 0s or 1s. Thus, your model is 'too good' (it has too many predictors given the amount of data). Try e.g. to fit a model without the interactions.

Cheers, Lorenz
- 
Lorenz Gygax
Centre for proper housing of ruminants and pigs
Agroscope Reckenholz-Tänikon Research Station ART
Tänikon, CH-8356 Ettenhausen / Switzerland