s-news
[Top] [All Lists]

Re: Tree for ordinal variable with uneven priors

To: Peter Flom <flom@ndri.org>
Subject: Re: Tree for ordinal variable with uneven priors
From: ripley@stats.ox.ac.uk
Date: Fri, 28 Jun 2002 19:19:36 +0100 (BST)
Cc: <s-news@lists.biostat.wustl.edu>
In-reply-to: <sd1c6618.084@MAIL.NDRI.ORG>
On Fri, 28 Jun 2002, Peter Flom wrote:

> Hello
>
> I am interested in fitting a tree model for a variable which is ordinal, and 
> where the distribution is far from uniform.
>
> More specifically, I would like to investigate drug use, which we have 
> operationalized as hardest drug used in the last year (none, marijuana,
> cocaine/heroin (not crack, and not injected), crack, and any injected drug.
>
> The distribution in our sample is
> none - 186  marijuana - 136   cocaine/heroin 137   crack - 41  injection - 28
>
>
> The ordinality means that the misclassification costs  are uneven (e.g., 
> calling a marijuana user a nonuser is no big deal.  Calling an injector a 
> nonuser is much more serious).
>
> I have been unable to find out how to do this in the S-Plus manuals; although 
> the classic reference (Breiman et al., 1984) offers some hints, they do not 
> provide any code (nor do they indicate that they used S-Plus at all).

That predates S-PLUS!

Use rpart and set up a loss matrix in the `parm' argument.  You can
find rpart at the Mayo site, and pre-compiled for Windows at mine
(S+2000 and S+6 in separate areas).


-- 
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 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595


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