Interactive demo · Statistical Methods for Modern Biology · Chapter 4

The nonparametric bootstrap on Darwin's Zea mays data

The sample is treated as a mixture of 15 point masses, the empirical distribution F̂n (top panel). A bootstrap resample draws 15 values from it with replacement; the statistic computed on many resamples approximates its sampling distribution (bottom panel). Two errors remain: the Monte Carlo error from using B resamples instead of all of them, which shrinks as B grows, and the bootstrap error from using F̂n instead of the true distribution, which does not. For the median the exact bootstrap distribution is known in closed form (gold dots), so the first error can be seen directly.

Statistic

Number of resamples B

Interval level

Resample

Scoreboard

Try these

Percentile intervals use the inverse of the empirical distribution function of the bootstrap values (the smallest value whose cumulative proportion reaches the level), which is also how the exact interval for the median is read from its exact distribution. Resamples come from a seeded generator, so a class sees the same pictures; "New seed" draws a fresh set.

In the lecture notes