[R-sig-ME] Restrictions on the class of GLMMs

John Maindonald john.maindonald at anu.edu.au
Mon Dec 7 10:10:27 CET 2009


Reasonably recently, it used to be possible, with glmer, to
associate one random term with each observation.  With the
current version of glmer(), I find that this is not allowed.  In the
case I was using it for, this allowed me to fit a random between
observations variance that was additive on the scale of the
linear predictor, rather than as with the dispersion estimate
fudge which estimates a constant multiplier for the theoretical
variance.  I am wondering what the reason may be for disallowing
this; does it unduly complicate code somewhere or other?

I am using lme4_0.999375-32;    Matrix_0.999375-32

Here is what I had been doing:

library(DAAG)
moths$transect <- 1:41  # Each row is from a different transect
moths$habitat <- relevel(moths$habitat, ref="Lowerside")
(A.glmer <-  glmer(A~habitat+log(meters)+(1|transect), 
                                family=poisson, data=moths))


Generalized linear mixed model fit by the Laplace approximation 
Formula: A ~ habitat + log(meters) + (1 | transect) 
  Data: moths 
AIC BIC logLik deviance
 95 112  -37.5       75
Random effects:
Groups   Name        Variance Std.Dev.
transect (Intercept) 0.234    0.483   
Number of obs: 41, groups: transect, 41

...

Thanks

John Maindonald             email: john.maindonald at anu.edu.au
phone : +61 2 (6125)3473    fax  : +61 2(6125)5549
Centre for Mathematics & Its Applications, Room 1194,
John Dedman Mathematical Sciences Building (Building 27)
Australian National University, Canberra ACT 0200.
http://www.maths.anu.edu.au/~johnm


John Maindonald             email: john.maindonald at anu.edu.au
phone : +61 2 (6125)3473    fax  : +61 2(6125)5549
Centre for Mathematics & Its Applications, Room 1194,
John Dedman Mathematical Sciences Building (Building 27)
Australian National University, Canberra ACT 0200.
http://www.maths.anu.edu.au/~johnm




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