s-news
[Top] [All Lists]

Re: logistic regression - no R squared or F output in long format

To: Prof Brian Ripley <ripley@stats.ox.ac.uk>
Subject: Re: logistic regression - no R squared or F output in long format
From: Frank E Harrell Jr <f.harrell@vanderbilt.edu>
Date: Thu, 24 Feb 2005 09:28:09 -0500
Cc: Scott Clark <scott.clark@adelaide.edu.au>, s-news@lists.biostat.wustl.edu
In-reply-to: <Pine.LNX.4.61.0502241348150.11628@gannet.stats>
References: <000001c51a72$c27245f0$5cf07f81@Scott> <Pine.LNX.4.61.0502241348150.11628@gannet.stats>
User-agent: Mozilla Thunderbird 0.9 (X11/20041105)
Prof Brian Ripley wrote:
R^2 is a measure of fit by least-squares, so not appropriate to logistic regression.

But is occasionally still useful:

@ARTICLE{cox92,
  author = {Cox, D. R. and Wermuth, N.},
  year = 1992,
title = {A comment on the coefficient of determination for binary responses},
  journal = {The American Statistician},
  volume = 46,
  pages = {1-4}
}

Frank


An F-test is not really appropriate either, unless there is evidence of over-dispersion and a quasi-likelihood is acceptable. You can that from the command line, using anova.glm.


On Thu, 24 Feb 2005, Scott Clark wrote:

Hi all,

Just wondering if anyone can help a beginner - having trouble getting an
R squared output from logistic regression in splus 6.2 student version.
I get Deviance Residuals, Min 1Q Median 3Q Max, Coefficients, Value Std.
Error,  t value, intercepts, null deviance, Residual deviance, degrees
of freedom, missing values and fisher score iterations BUT NO R squared
or F test figures using the long format, no matter how I change the
dialog box.




--
Frank E Harrell Jr   Professor and Chair           School of Medicine
                     Department of Biostatistics   Vanderbilt University

<Prev in Thread] Current Thread [Next in Thread>