From B resamples of the data the bootstrap gives replications T*1, …, T*B of a statistic Tn. Section 3.4 turns them into four intervals: the Normal interval Tn ± z seboot; the percentile interval (the α/2 and 1 − α/2 quantiles of the T*); the basic pivotal interval, the percentile interval reflected about Tn; and the studentized pivotal interval, which uses the quantiles of Z* = (T* − Tn)/se*, where se* is a standard error computed inside each resample (here by the influence function, as the book does in Example 3.17). Panel (a) shows the replications and the intervals, panel (b) the two standardised versions of the replications.
Resample b draws n indices floor(u·n) from the seeded generator, row after row. Quantiles interpolate linearly between order statistics. seboot is the standard deviation of the replications with divisor B, equation (3.11). For the skewness, se* is the nonparametric delta method with the influence function of the skewness; for the correlation, with the influence function uv − r(u² + v²)/2 of standardised coordinates; for the mean, σ̂*/√n. The jackknife standard error is shown for comparison (Section 3.1).
code/ch03_bootstrap_jackknife.py (intervals, si_demo; Figure 3.S2).