DAVID ARTETA GARCIA wrote:> Dear list,
>
> sorry for asking a more statistical question. I have been reading on
> penalyzing estimates of regression and bootstrapping regression,
> trying to include both or either in my analysis. But it is not clear
> to me (mainly due to my non-statistics background) whether they aim to
> do similar things in the case of regression, i.e. to get robust
> estimates, or there is a completely different goal to applying them.
> Reading and using "logistf" and "penalized" packages
for penalization
> and "boot" and "bootstrap" and John Fox?s append on
bootstrapping, I
> think penalization aims at the coefficients, but bootstrapping aims at
> standard errors and CIs.
>
> Can anyone tell me whether I am far from understanding both concepts?
>
> Thanks in advance
>
>
> David
>
David,
I think you have contrasted the two approaches fairly accurately.
Penalization is an estimation technique used up front to improve the
mean squared error for predictions and parameter estimates.
Bootstrapping (except for bootstrap bumping, bagging, and a few other
variations) is used primarily to inform you of the variation or bias of
a specific estimator.
Frank
--
Frank E Harrell Jr Professor and Chair School of Medicine
Department of Biostatistics Vanderbilt University