bootstrap resampling methods something for nothing.pdf

THE STATISTICIAN’S PAGEBootstrap Resampling Methods: Something forNothing?Gary L. Grunkemeier,PhD, and YingXing Wu,MDProvidence Health System, Portland, OregonThe paper by Brunelli and colleagues[1]in this issueofThe Annals of Thoracic Surgeryused bootstrapresampling to select the ?nal variables for a logisticregression model to predict air leak after pulmonarylobectomy. Bootstrapping is a generic methodology,whose implementation involves a simple yet powerfulprinciple: creating many repeated data samples from thesingle one in hand, and making inference from thosesamples. The name derives from the expression “pullingoneself up by one’s bootstraps,” meaning no outside help(additional data, parametric assumptions) is involved. Itseems almost magical, like getting something for noth-ing. Bootstrapping can be used for many statistical pur-poses besides variable selection. One of the -mon is pute a con?dence interval (CI). As ademonstration of the technique, we pute a boot-strap CI for the O/E (observed-to-expected) ratio ofmortality after myocardial page 1205Background and RationaleBootstrap methods are more than 20 years old[2], butthey puter-intensive and have only recently e widely available in statistical programs. Powerfuland widely puting capability is transform-ing statistics. Originally, computers just speeded up datahandling and putations. Now they enablee