[R-sig-ME] lmer logistic nested random effects

Ben Pelzer b.pelzer at maw.ru.nl
Mon Aug 13 23:37:45 CEST 2012

Dear list,

I would like to fit a logistic model with lmer, for nested factors like 
classes nested in schools, and Y being 0 or 1. So I have dichotomous Y 
values of pupils in classes in schools. My null-model reads:

lmer ( Y ~ (1|school/class), family=binomial(link=logit), nAGQ=7)

The options nAGQ=7 causes the following error:

    AGQ method requires a single grouping factor

Omitting the nAGQ=7 works fine and produces Laplace approximation of the 

Am I right in concluding that for hierarchical nested factors (like 
class nested in school) glmer can only use 1 quadrature point? Or am I 
overlooking something?

Of course, more quadrature points would be nicer, and in e.g. xtmelogit 
in stata and glimmix in sas I believe more quad. points are possible 
(i.e., for hierarchically nested factors, not for crossed factors) Hence 
I was surprised that in lmer, it seems impossible. Could anyone confirm 
this, please? Kind regards,


More information about the R-sig-mixed-models mailing list