Date:         Tue, 1 May 2007 09:55:25 -0400
Reply-To:     "H. Lawrence Hotchkiss" <larryh@zebra.us.udel.edu>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         "H. Lawrence Hotchkiss" <larryh@ZEBRA.US.UDEL.EDU>
Subject:      a high odds ratio an artefact ? (2)
Content-Type: TEXT/plain; charset=us-ascii

I'm responding to a question about a large odds ratio in logistic regression
submitted by Frank Thomas. Large odds ratios occur fairly frequently in logistic
regression. They often are due to a near zero entry in the crosstab between the
dependent variable and the independent variable associated with the large odds
ratio. When one cell in this crosstab contains a zero, the coefficient estimate
tries to go to positive or negative infinity, and the algorithm doesn't
converge. A good marker for this occurs when a very large (or small) odds ratio
is associated with an excessively large standard error estimate.

So the 1st check is to run the crosstab. If it shows a near zero cell, you might
have to delete the variable from the regressor set, or (if feasible) collapse
categories of it.

Larry Hotchkiss


Larry Hotchkiss
University of Delaware
IT User Services -- Smith Hall
Newark, DE 19716
302-831-1989  larryh@zebra.us.udel.edu