Date:         Mon, 8 Sep 2003 08:23:10 -0400
Reply-To:     Mark Davenport <madavenp@OFFICE.UNCG.EDU>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Mark Davenport <madavenp@OFFICE.UNCG.EDU>
Subject:      Re: Anomaly in logistic regression output
Comments: To: ban@zoology.ubc.ca
Content-Type: text/plain; charset=US-ASCII

Someone please correct me if I am wrong. I believe that what is happening is  SPSS is giving you the result it had at the time it stopped processing due to lack of memory. I believe that logistic reg is based on an maximum likelihood iterative process. That is, it runs a pass and produces results, then runs another pass using the prior result as a baseline. It continues to do this until the change in parameters from the prior run and the present run are smaller than a particular value. Next time you run logistic, go to the options window and ask for an iteration history. I bet you find that your memory stalled runs have fewer iterations.

Mark

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Mark A. Davenport Ph.D.
Asst to the Vice Chancellor for Student Affairs/Research and Evaluation
The University of North Carolina at Greensboro
149 Mossman Bldg.
Greensboro, NC  27402-6170
336.334.5099
madavenp@office.uncg.edu

'An approximate answer to the right problem is worth a good deal more than an exact
answer to an approximate problem'      -- J. W. Tukey

>>> Stephen Ban <ban@zoology.ubc.ca> 9/5/2003 4:13:14 PM >>>
I've run across a bit of a perplexing problem: I've run a logistic
regression exactly the same way on a data set four times; two of those
times I ran it with the SET MEMORY parameter set to 800 000; the other two
runs I left this command out. Otherwise, the production jobs were
identical. In the two runs missing the SET MEMORY command, I did get an
"out of memory error", but the run completed (apparently) successfully
nonetheless. However, if these runs were compared with the other two runs
that DID allocate memory, the results are completely different!

Normally, I would assume that I should "trust" the output from the runs
that didn't run into the out of memory problem, but the output from the
"memory deficient" runs actually makes more sense!

What is most disturbing, however, is that I can get completely different
(yet repeatable) results from the same command merely by changing the
memory allocation!

Any insights?

Thanks!
Stephen