Sophia Kyriakou <sophia.kyriakou17 <at> gmail.com> writes:
>
> hello, I am using the optim function to maximize the log likelihood of a
> generalized linear mixed model and I am trying to replicate glmer's
> estimated components. If I set both the sample and subject size to q=m=100
> I replicate glmer's results for the random intercept model with parameters
> beta=-1 and sigma^2=1. But if I change beta to 2 glmer works and optim
> gives me the error message "function cannot be evaluated at initial
> parameters".
>
> If anyone could please help?
> Thanks
snip to make gmane happy.
It looks like you're getting floating-point under/overflow. If you do
all the computations on the log scale first and then exponentiate,
it seems to work, i.e.:
piYc_ir[i,] <- lchoose(m,Y[i]) + Y[i]*(z+beta) + (-z^2/(2*exp(psi))) -
m*(log1p(exp(z+beta))) - 0.5*(log(2*pi)+psi)
piYc_ir[i,] <- exp(piYc_ir[i,])
follow-ups should probably go to r-sig-mixed-mod...@r-project.org
instead ...
Ben Bolker
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
