s-news
[Top] [All Lists]

Re: generic factor levels

To: "Schneck, Ryan" <ryan_schneck@standardandpoors.com>
Subject: Re: generic factor levels
From: Prof Brian Ripley <ripley@stats.ox.ac.uk>
Date: Wed, 24 Mar 2004 07:48:04 +0000 (GMT)
Cc: s-news@wubios.wustl.edu
In-reply-to: <4B435DE9D3AA7245BD290D8D580ABC9DB2A1B2@nycmsx03.mhf2.mhf.mhc>
On Tue, 23 Mar 2004, Schneck, Ryan wrote:

> Do you know if there is a way to show the specific names of the factor
> levels in the coefficients matrix that is produced by:
> 
> fit1<-glm(DEFAULT~RATING+SECTOR,family=binomial(link=logit),data=RC)
> summary(fit1)
> 
> The default coefficient matrix just has generic labels:  "RATING1,
> RATING2, . . . SECTOR1, SECTOR2, . . . " And, so, it is hard to use the
> results in a practical way, to tell the story of the data.

They are not directly related to factor levels, but to particular (e.g. 
Helmert) contrasts.  You might prefer to use treatment contrasts: Venables 
& Ripley (2002) discuss this in depth, and recommend treatment contrasts 
for logistic regressions.

The real answer is to use predictions to explain the fitted model (rather 
than `the story of the data', for the model may not be an adequate 
summary, and based on your earlier postings it may be an over-elaborate 
one).  The coefficients are only a means to an end, predictions.

-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595


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