One of the great new features in JMP Pro, Version 10, is the one-click bootstrap. Most tables of results in analytical platforms have a new item in their context (right-click) menus, called Bootstrap. Selecting this item will bring up a small dialog with some options. Once you click OK, JMP will perform a bootstrap analysis of the report table that you clicked on and produce a table of the bootstrap results. You can then run a Distribution analysis on that table and get bootstrap confidence intervals for the statistic in question.
The statistical bootstrap is a resampling technique that estimates the sampling distribution for a statistic, which allows you to estimate the sampling variability even when there is no analytical solution for the standard error of a particular statistic. This is accomplished by treating the sample that you have as a population and repeatedly sampling from that sample. Each time you get a bootstrap sample, you calculate the statistic(s) of interest. You can then use the variability of the statistics calculated during the bootstrap replication as a proxy for the variability of the statistics calculated from your original sample. For more information on the statistical bootstrap, I recommend the 1993 text by Efron & Tibshirani (reference below).
Prior to Version 10 of JMP Pro, you could perform a bootstrap by writing your own JSL script to do the resampling and then collecting all the resulting statistics from each bootstrap sample. Now, JMP Pro version 10 will do all that work for you! The results can then be analyzed in the Distribution platform, where a special Bootstrap Confidence Limits outline node will appear containing percentile-based confidence limits of varying coverage levels. For a quick video demo of the bootstrap feature in action, you can check out the JMP Pro product page.
The bootstrap is also something that we believe should be promoted in introductory statistics courses. It allows students to see the effects of sampling variability without worrying about distribution theory or having to make distributional assumptions for your data. It also avoids the feeling of introductory statistics courses being "just a bunch of formulas." In the past, it has required significant computational resources as well as scripting/programming ability. But with today's computers and the one-click bootstrap in JMP Pro, there's no coding necessary to start using bootstrap examples to help students learn about sampling variation.
This blog post is based on my recent presentations at JMP Discovery Summit 2012 in Cary, NC, and SESUG 2012 in Durham, NC. Some of the examples I featured in those presentations will be the subjects of future blog posts. One of the purposes of the presentations was to get JMP users thinking about how the new bootstrap feature in JMP Pro would be useful to them. If you have interesting use cases of the bootstrap feature, please share them below.