Date:         Fri, 29 Apr 2005 16:18:24 -0400
Reply-To:     Tim Wade <wadetj@gmail.com>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Tim Wade <wadetj@GMAIL.COM>
Subject:      Proc NLMIXED--binomial family with identity link
Content-Type: text/plain; charset=ISO-8859-1

Hello: I have the following Proc NLMIXED code which works perfectly:

proc nlmixed data=body1;
parms a1=-2.80 beta1=0.2539 logs2u=-0.1;
eta=a1+beta1*m_l10count_dy+u;
expeta=exp(eta);
p=expeta/(1+expeta);
s2u = exp(2*logs2u);
model hcgi~binary(p);
random u~normal(0, s2u) subject=beachnum;
run;

I would like to try the same model with an identity link. I tried to
use the following, with initial parameter estimates from Proc Genmod
(dist=family, link=identity)


proc nlmixed data=body1;
parms a1=0.0244 beta1=0.0290 beta2=0.1202  beta3=0.0607 s2u=-0.1;
eta=a1+beta1*m_l10count_dy+beta2*gicontact_any+beta3*venfest+u;
*expeta=exp(eta);
p=eta;
s2u = exp(2*logs2u);
model hcgi~binary(p);
random u~normal(0, s2u) subject=beachnum;
run;

but it won't converge. Am I on the right track here? Any suggestions?
Thanks much. Tim