Message-ID: <49EDC50C.5030900@vanderbilt.edu>
Date: 2009-04-21T13:07:24Z
From: Frank E Harrell Jr
Subject: automatic exploration of all possible loglinear models?
In-Reply-To: <loom.20090421T063645-287@post.gmane.org>
Dieter Menne wrote:
> Christopher W. Ryan <cryan <at> binghamton.edu> writes:
>
>> Is there a way to automate fitting and assessing loglinear models for
>> several nominal variables . . . something akin to step or drop1 or add1
>> for linear or logistic regression?
>
> Not strictly for loglinear, but glm works with stepAIC. Make sure that
> in the field you are working this approach is an accepted ritual.
>
> Dieter
>
Dieter was kind. I would say the following: Make sure your colleagues
are interested in non-reproducible results, overfitting, incorrect
P-values, and incorrect confidence limits before proceeding.
Frank
--
Frank E Harrell Jr Professor and Chair School of Medicine
Department of Biostatistics Vanderbilt University