[R] solved mystery of difference between glmmPQL and lme

[Paul Johnson]

I asked a few days ago about the difference in results I saw between the 
MASS function glmmPQL (due to Venables and Ripley) and the lme function 
from the package nlme (due to Pinheiro and Bates).  When the two tools 
apply to the same model (gaussian, link=identity, correlation=AR1), I 
was getting different results and wondered if there was an argument in 
favor of one or the other.

Several list readers emailed me to point out that glmmPQL is repeatedly 
calling lme, so if a model really can be estimated by lme, then lme is 
the more appropriate one because it is maximum likelihood, rather than 
quasi-likelihood.

That did not explain the difference in results, so I read the source 
code for  glmmPQL and learned that it sets the method for lme fitting to 
 "ML".  In contrast, lme defaults to "REML".  The estimates from 
glmmPQL and lme (method="ML") are identical in my test case.

pj
--
Paul E. Johnson                       email: [EMAIL PROTECTED]
Dept. of Political Science            http://lark.cc.ku.edu/~pauljohn
1541 Lilac Lane, Rm 504
University of Kansas                  Office: (785) 864-9086
Lawrence, Kansas 66044-3177           FAX: (785) 864-5700
______________________________________________
[EMAIL PROTECTED] mailing list
https://www.stat.math.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html