Matthias Richard wrote:
> This is a question concerning the bootstrap function of the HMISC-library
> written by Harrell (I use SPLUS 3.3 under windows95). I am interested in
> the corrected R-square of a final fitted model (logistic regression using
> the lrm()-function) to have an estimate 'how stabile' the model is, i.e.
> how much the uncorrected R-squared may be too optimistic, assuming that
> this model is used for other samples.
>
> 1) is the corrected R-square ('index.corrected') the mean of the
> R-squares over the bootstrap replications or is it sort of a lower bound
> of the `true` R-square (like a confidence interval)? Is there a confidence
> interval for the index.corrected too?
Matthias - your speaking of the Design library (not Hmisc). The documentation
in
Design along with our Stat in Med papers in '96 and '98 should tell you what
you want. Corrected R2 = original R2 - average over bootstrap resamples of
optimism (R2 in bootstrap sample minus R2 when model is frozen and applied to
the original sample).
>
>
> 2) Could I get an answer to the same question by doing cross validation
> and what would be the major (dis-)advantage?
Cross validation has much high mean squared error of estimates such as
corrected R2 (see Efron). You can do 10-fold cross validation 20 times
and take an average and get MSE about equal to the bootstrap. A disadvantage
of cross-val is that you are validating a "fit 90% of the data" model. Brian
Ripley will argue that "what exactly are you validating when you fit on
samples
with replacement" but in expectation I think the bootstrap does what you want.
>
>
> I know I have to read the Efron and Tibshirani book on it - probably
> someone knows where to start best here.
>
> Thanks for any comments!
>
> yours Matthias
>
> ----------------------------------------
> Matthias Richard (doctoral candidate)
>
> Center for Research on Psychotherapy
> Christian-Belser-Str. 79a
> 70593 Suttgart
> Germany
> email: richard@psyres-stuttgart.de
> telephone: ++49-711-6781-408
> http://www.psyres-stuttgart.de
>
> COST Action B6:
> 'Efficient Psychotherapy of Eating Disorders'
> (COoperation of Science and Technology in the European Union)
> -----------------------------------------
>
> -----------------------------------------------------------------------
> This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
> send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
> message: unsubscribe s-news
--
Frank E Harrell Jr
Professor of Biostatistics and Statistics
Division of Biostatistics and Epidemiology
Department of Health Evaluation Sciences
University of Virginia School of Medicine
http://hesweb1.med.virginia.edu/biostat
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
|