Date:         Thu, 28 Jun 2007 07:40:09 -0400
Reply-To:     Steve Denham <steven.c.denham@MONSANTO.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Steve Denham <steven.c.denham@MONSANTO.COM>
Subject:      Re: Logistic Regression with Random effects

On Wed, 27 Jun 2007 14:29:32 -0400, Brad Heins <hein0106@UMN.EDU> wrote:

>I am trying to analyze a data set of 300 observations.  The response
>variable is a 0 or 1 (binary).  I also want to include a random effect in
>the model.
>
>What would be the best option PROC GLIMMIX or NLMIXED?
>
>The model that I would like to fit is below:
>
>CLASS BREED  HY herd service cow;
>MODEL CONCEIVE = HERD  breed hy(herd) service;
>random cow(breed);
>
>
>I can not seem to get PROC GLIMMIX to converge with binary data and I am
not
>sure where to get starting values to use NLMIXED?
>
>If anyone could offer suggestions, I would appreciate it.
>
>Brad H.

Brad,

What symptoms are you getting with GLIMMIX?  Long, long iterations?  Or is
it cutting out at 20?  If so, simply upping the MAXITER may solve the
problem.

Also, in GLIMMIX both R and G side covariance structures are specified with
the RANDOM statement.  From over here, it looks like you have multiple
breedings per cow, but each is specified separately (service).  You might
need to think of this more as a repeated measures design, with cow(breed)
as a subject.  (Here, I'm assuming that cow is an id number).

Good luck.

Steve Denham
Mathematical Biologist
Monsanto Co.