[R-sig-ME] [Fwd: testing quasi-likelihood in glmer with simulations]
bolker at zoology.ufl.edu
Mon Sep 29 21:01:20 CEST 2008
I spent a bit of time this weekend running simulations
of quasi-likelihood estimation in glmer. The results
are posted at
I confirmed my suspicion that glmer is scaling the
random-effects variance-covariance matrix by the
scale parameter when it shouldn't; I also discovered
that the quasi-likelihood fit to lognormal-Poisson-distributed
data is alarmingly (?) biased, although a lognormal-Poisson
fit with individual-level random effects works nicely.
Any comments are welcome.
More information about the R-sig-mixed-models