Date: Mon, 16 Jun 2003 10:43:18 -0400
Reply-To: "Handel, Richard W." <HandelRW@EVMS.EDU>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: "Handel, Richard W." <HandelRW@EVMS.EDU>
Subject: Re: Nagelkerke R Square
Content-Type: text/plain; charset="iso-8859-1"
From: Christina Cutshaw [mailto:ccutsha1@JHEM.JHMI.EDU]
Sent: Thursday, June 12, 2003 11:08 PM
Subject: Nagelkerke R Square
As part of my dissertation, I ran a binary logistic regression model
(N=73, Outcome present n=22, outcome absent n=53). I have three
variables in the model: one nominal (a) one dichotomous (b)and one
continuous (c) . The model contains the three main effects a, b,and c,
and one interaction term a*b. The Nagelkerke R Square is 0.479. I
understand that this is a "pseudo R-squared" measure because this is a
logistic versus linear regression, but could someone explain to me how
to understand what it is calculating and how to explain variance
accounted for in logistic regression? I have never seen this measure
used in journal articles using logistic regression.
I recommend two introductory texts that are both outstanding:
Applied Logistic Regression Analysis, Second Edition, Scott Menard
Logistic Regression, A Primer, Fred Pampel
Both are from the SAGE series and are inexpensive. My understanding is that
there is no index in logistic regression that is "the same as" variance
accounted for in OLS regression. For pseudo-R^2, Menard recommends the use
of McFadden's R^2 instead of Naglekerke's, but you have to calculate it from
the SPSS output.
Richard W. Handel, Ph.D.
Eastern Virginia Medical School
Department of Psychiatry and Behavioral Sciences
P.O. Box 1980
Norfolk, VA 23501