[R-sig-ME] FW: random slopes with lme4

[Julie Morand-Ferron]

Dear all,
I would like to share this reply from Dr Bolker with members of the list:

I wrote: 
I cannot find a confirmation that the logic of the procedure one would use to test the significance of random slopes with linear mixed-models in package nlme (i.e. comparing 2 identical models differing only in the random slope component with anova(model1,model2)) is also correct for glmm in lme4?

Dr Bolker:
  Yes, the logic of comparison should be the same in both cases.  The
usual caveats apply: (1) the likelihood ratio test
is asymptotic (analogous to assuming infinite 'denominator degrees of
freedom' in an F test); (2) likelihood ratio test
comparisons where the null hypothesis is at the boundary of the feasible
set (in this case NH = (variance of 'centcond' response
across individuals equals 0)) are conservative (increased type II error)
by approximately a factor of 2.  See <http://glmm.wikidot.com/faq> ...



Julie Morand-Ferron

Postdoctoral researcher
Edward Grey Institute
Department of Zoology
Oxford University
South Parks Road
Oxford, OX1 3PS

Tel: 01865 281999
Fax: 01865 271168