s-news
[Top] [All Lists]

Re: logit interpretation

To: Ita.Cirovic-Donev@hypo-alpe-adria.com
Subject: Re: logit interpretation
From: Frank E Harrell Jr <f.harrell@vanderbilt.edu>
Date: Thu, 26 Oct 2006 07:59:14 -0500
Cc: s-news@lists.biostat.wustl.edu
In-reply-to: <OF322AD1C3.6AD66A03-ONC1257213.00310633-C1257213.003FED4F@arz.co.at>
References: <OF322AD1C3.6AD66A03-ONC1257213.00310633-C1257213.003FED4F@arz.co.at>
User-agent: Thunderbird 1.5.0.5 (X11/20060728)
Ita.Cirovic-Donev@hypo-alpe-adria.com wrote:



I have the following output from the logistic regression:

             Coef   S.E. Wald Z      P
Intercept -2.5159 0.3073  -8.19 0.0000
     P4.t -0.4595 0.2080  -2.21 0.0272
     P5.t -0.6054 0.2363  -2.56 0.0104
     L5.t  0.9877 0.3665   2.70 0.0070
    L12.t -1.6130 0.2577  -6.26 0.0000
    L17.t -3.4368 0.2876 -11.95 0.0000
     A1.t  0.4639 0.1906   2.43 0.0149
     A7.t  1.1285 0.2482   4.55 0.0000
  LEV21.t  1.1970 0.1887   6.34 0.0000
  LEV23.t -1.7330 0.2511  -6.90 0.0000

these are financial variables, rank transformed prior to runing logistic
regression. My question is about the intercept. Is it normal for it to be
so large as compared to other variables. thanks


Ita

Why would you expect the variables to act on the log odds in a way that is linear on the ranks? If you are concerned about linearity, then regression splines may be the answer. If you are concerned about robustness then curtailing the regression spline fits to extrapolate in a flat fashion will deal with overly influential observations. You will have difficulty getting predicted values when covariates are ranked.

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

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