Prev: reading raw file
Next: survival analysis: setting up longitudinal data question
From: Nitish on 12 Apr 2010 18:09
Hi,

I am trying to model count data as Poisson with normal random effect.
I am assuming that the data is coming out of Poisson arrival rate with
normal random effect. Since this form does not have closed form
likelihood, I have couple of options. One of them being using SAS'
ability to code the log likelihood, since SAS does have this coded.
This one does not converge for my data.

The other option is write out the approximate likelihood using
quadratures or use maximum simulated likelihood. Has anybody else
tried such likelihood before? Appreciate any help.

proc nlmixed data=lnbemedata_v1 TECH=CONGRA;
parms alpha1 -9 gamm 0.67 sigm 1.6;
BOUNDS sigm > 0;
eta = (alpha1 + gamm * lnw)+e;
lambda = exp(eta);
model newscount ~ poisson(lambda);
random e ~ normal(0,sigm*sigm) subject=id;
run;

I have also tried coding the sigma as logsigma instead, I get similar
issues.
Regards
Nitish Sinha
 | 
Pages: 1
Prev: reading raw file
Next: survival analysis: setting up longitudinal data question