Date: Wed, 11 Oct 2006 21:59:30 -0700
Reply-To: David L Cassell <davidlcassell@MSN.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: David L Cassell <davidlcassell@MSN.COM>
Subject: Re: Hypergeometric distribution
Content-Type: text/plain; format=flowed
>Anyone familiar with Zelterman book, "Advanced Log-Linear Models Using
>SAS"? I'm looking for help with Program 8.2 from the book. This program
>fits a model to a stratified 2x2 table with the hypergeometric
>distribution, and Prof. Zelterman's macros appearing in Program 8.4.
>I'm attempting to use the published programs to estimate the model.
>Basic problem is that the model fails to converge. No problem with the
>code; it's copied and pasted correctly from the SAS companion web site
>into my data editor. It runs correctly with the supplied sample data. I
>suspect a problem with my data and, probably, the appropriateness of
>this model for my data. Error message in the log are these:
>WARNING: The specified model did not converge.
>ERROR: Error in computing the link function, its derivatives, or the
>NOTE: The scale parameter was held fixed.
>ERROR: Negative variance from user-defined variance function.ERROR:
>Where could/should I begin looking for solutions? Thanks for all
I think you are right. I think that your data do not match the
underlying assumptions of the model, so that SAS simply cannot
fit the model to your data in any decent way.
First, when you say a stratified 2x2 table, are *your* data from
a survey sample? If so, then don't use PROC FREQ anyway. You
need to move to PROC SURVEYFREQ and focus on the design
effects instead of hypergeometric distributions.
Then, assuming that you don't have a survey sample, try a simpler
model and see if you can get convergence. Analyze the fitted
model and see if it looks anything like it would meet the assumptions
of the hypergeometric.
David L. Cassell
3115 NW Norwood Pl.
Corvallis OR 97330
The next generation of Search—say hello!