Message-ID: <20030515063810.07ff7acb.fharrell@virginia.edu>
Date: 2003-05-15T10:38:10Z
From: Frank E Harrell Jr
Subject: Manly's randomization analysis of multiple regression
In-Reply-To: <E19G8r9-0006ji-00@server.family>
On Wed, 14 May 2003 22:52:33 -0400
Nurnberg-LaZerte <mail at fwr.on.ca> wrote:
> My wife has been using a diagnostic from Manley (1991; "Randomization and MonteCarlo Methods in Biology") that compares a normal multiple regression's performance with that using random predicted variables.
>
> Is there something like this already available in R?
>
> If not, the "boot" package looks like a good place to start looking for methods, no?
>
> Thanks in advance
> Bruce L.
This is related to that: Look at the validate function in the Design package (http://hesweb1.med.virginia.edu/biostat/s/Design.html) which has a "randomize" option to get the apparent and validated performance of models fitted by randomly permuting the vector of response variable values. This is not a high-precision way to estimate bias in your original R^2 etc. but rather is a teaching tool. For efficient estimation of future model performance use the validation bootstrap option (the default for validate( )).
---
Frank E Harrell Jr Prof. of Biostatistics & Statistics
Div. of Biostatistics & Epidem. Dept. of Health Evaluation Sciences
U. Virginia School of Medicine http://hesweb1.med.virginia.edu/biostat