I have a question regarding the method used to fit the generalized linear
mixed model in Splus 6.2 WinXp. The default is REPQL. I have a 4-level
mixed model with a dichotomous outcome. When I use the default estimation
method I receive error messages indicating false convergence.
I have tried REMQL but I notice that the random effect (level 2) is
significantly smaller [by a factor of 100]. Given that I have a 4-level
model I can't use AGQUAD or LAPLACE. Any thoughts on how to facilitate
convergence?
Respectfully,
Frank Lawrence
|