ALR: Memory and time issues

Cande V Ananth (ananthcv@EPI.UMDNJ.EDU)
Thu, 15 Jan 1998 09:04:11 -0500


Fellow SPlus'ers:

I'm attempting to fit a (bivariate) logistic regression model on
a data set containing 10,500 obs, using the ALR routine. The
marginal model contains 4 covariates (plus an intercept) and the
pairwise association model contains 4 binary and 1 continuous
covariates.

The trouble that I'm encountering is that even after 12 hours, I
don't have the results -- the algorithm seems to be crunching away
to glory ! This is only one of my preliminary model, and I'll need
to fit several such models before I come up with a best fitting
model.

I have provided reasonably good starting values for the log OR model
(based on crude associations), and the ALR routine fits a GLM to
obtain the starting values for the marginal model parameters. I am
using a SUN Sparc workstation and Splus (version 3.4) operating on
SunOS 5.3

I will greatly appreciate comments on how to get the results quickly
(if at all possible !). Thanks in advance. If additional information
is required, I'll be glad to provide them.

---

Cande V. Ananth, PhD, MPH ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The Center for Perinatal Health Initiatives Dept of OB/GYN and Reproductive Sciences Tel: (732) 235-7940 Univ of Medicine and Dentistry of New Jersey Fax: (732) 235-6627 Robert Wood Johnson Medical School ananthcv@EPI.UMDNJ.EDU 125 Paterson St, New Brunswick NJ 08901-1977 http://epi.umdnj.edu/ananth ,,, (o-o) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ooO-(_)-Ooo