The bootstrap is a general resampling procedure which can be applied to estimate the sampling distribution of a statistic. From the statistical practitioner's point of view it has attractive properties because it requires few assumptions, little modeling or analysis, and can be applied in an automatic way in a wide variety of situations regardless of their theoretical complexity. The bootstrap can provide answers to questions that are too complicated for traditional statistical analyses, which are usually based on asymptotic normal approximations. A brief discussion of the non-parametric bootstrap is presented, followed by examples and illustrations. Possible suggestions regarding the teaching of these concepts at various levels are made. The key requirements for computer implementation of the bootstrap method include a flexible programming language with a collection of reliable quasi-random number generators, a wide range of built-in statistical bootstrap procedures and a reasonably fast processor. The use of the statistical languages S and Fortran, using the current commercial versions S-Plus 4.5 and Digital Fortran 6.0, are illustrated.
The CAUSE Research Group is supported in part by a member initiative grant from the American Statistical Association’s Section on Statistics and Data Science Education