You might want to try a logistic regression model to predict the binary
response as a function of your continuous predictor. You can then select a
cut point for your predictor variable to define a binary predictor variable
and fit the logistic regression model using this binary variable as the only
predictor. You could repeat the process for different cut points. The
optimal cut point can be selected as the one that maximizes the profile
likelihood function of the model.
Tableman and Kim (Survival Analysis using S; 2004) have presented a similar
approach for selecting cut points in the context of the Cox proportional
hazards model for survival analysis.
Hope this gives you some ideas to pursue.
-Christos
-----Original Message-----
From: s-news-owner@lists.biostat.wustl.edu
[mailto:s-news-owner@lists.biostat.wustl.edu] On Behalf Of Im, Kelly
Sent: Wednesday, January 25, 2006 2:45 PM
To: s-news@lists.biostat.wustl.edu
Subject: Re: [S] finding cutoff point for a continuous variable when outcome
is binary.
Hello,
I have a binary (0 or 1) outcome variable and a continuous predictor
variable. When I look at the smoothing plots I don't see much relationship
between these two variables, i.e., plots give me almost a horizontal line.
nonetheless, we're try to find a cutoff point. I prefer not to run such
analysis, but seems to have no choice. most of things I'm analyzing is
exploratory in a sense.
So, I'm using CART (classification and regression trees) to see if I can
pick up some useful cutoff points. but CART gives me a best tree with 40
some end nodes, which is not so useful for us. .. I'm wondering if there are
other ways to look at this data.
or I should stop here. I would appreciate any advice,
kelly
--------------------------------------------------------------------
This message was distributed by s-news@lists.biostat.wustl.edu. To
unsubscribe send e-mail to s-news-request@lists.biostat.wustl.edu with the
BODY of the message: unsubscribe s-news
|