One interesting consideration of goodness-of-fit (as opposed to
significance) is J.S. Cramer (1999), 'Predictive performance of the
binary logit model in unbalanced samples', The Statistician, 48, Part 1,
pp. 85-94.
He proposes a measure 'lambda' which is a measure of how much better
your model does than if you just used uniform probabilities to predict
the binary outcome. If you do no better than the naive 'uniform
probability' assumption lambda is zero; if you improve by 100 per cent
(i.e. predict perfectly) lambda is 1. He argues that lambda is
analagous to R-squared in linear regression.
Hope this helps
Jim
--------Original message--------------------
Date: Sun, 6 Feb 2005 09:15:24 -0800 (PST)
From: Ricardo Ovaldia <[email protected]>
Subject: st: Hosmer-Lemeshow statistic
Dear all,
I though that some years back there was a posting
regarding criticism of the Hosmer-Lemeshow GOF
statistic. After a frustrating and unsuccessful search
through the archives and FAQ, I am writing to the list
to see if anyone recalls the criticisms or can point
me to a source with this information.
Thank you in advance,
Ricardo.
=====
Ricardo Ovaldia, MS
Statistician
Oklahoma City, OK
__________________________________________________
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
