关于「自助法的区间:百分位、偏差校正加速与小样本的覆盖不足」,《门店小实验的重抽样推断》给出的定量结论是什么?

结论:第四章给出百分位与简化 BCa 自助区间,五个指数观测时名义百分之九十五的百分位区间实际覆盖 0.798。

数据口径:书中参数为示意;全部计算由书中程序生成;本书属于非个人化经营建议。 书中以标注框区分已发表的证据、可运行的 Python 计算(20 段程序)、原典与作者的假说。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 0.798

coverage of nominal 0.95 intervals, 500 data sets
populationntpctBCa| pct all below mu
exp(1)50.8920.7980.808 |0.178
exp(1)100.9040.8580.876 |0.122
exp(1)200.9320.9180.932 |0.066
logn(0,1)50.7940.7060.732 |0.278
logn(0,1)100.8460.8000.830 |0.184
logn(0,1)200.8880.8820.888 |0.116
查看完整验算程序
# Chapter 4. p42: do bootstrap intervals cover at small n?
# Model: draw n days from a skewed population with known mean
# (exponential(1), mean 1; lognormal(0, 1), mean exp(0.5)); build the
# nominal 95 percent t, percentile and simplified BCa intervals of
# p41 (B = 500 resamples each) and record whether the true mean is
# inside; the last column is the share of percentile intervals lying
# wholly below the true mean. R = 500 seeded data sets per
# cell, so a coverage near 0.95 carries a Monte Carlo standard error
# of about 0.01. Sizes and populations are illustrative.
import math
import random
from statistics import NormalDist

def betacf(a, b, x):
    tiny = 1e-300
    qab, qap, qam = a + b, a + 1.0, a - 1.0
    c = 1.0
    d = 1.0 - qab * x / qap
    d = 1.0 / (d if abs(d) > tiny else tiny)
    h = d
    for m in range(1, 300):
        m2 = 2 * m
        aa = m * (b - m) * x / ((qam + m2) * (a + m2))
        d = 1.0 + aa * d
        d = 1.0 / (d if abs(d) > tiny else tiny)
        c = 1.0 + aa / c
        c = c if abs(c) > tiny else tiny
        h *= d * c
        aa = -(a + m) * (qab + m) * x / ((a + m2) * (qap + m2))
        d = 1.0 + aa * d
        d = 1.0 / (d if abs(d) > tiny else tiny)
        c = 1.0 + aa / c
        c = c if abs(c) > tiny else tiny
        de = d * c
        h *= de
        if abs(de - 1.0) < 1e-15:
            break
    return h


def betai(a, b, x):
    if x <= 0.0:
        return 0.0
    if x >= 1.0:
        return 1.0
    lb = (math.lgamma(a + b) - math.lgamma(a) - math.lgamma(b)
          + a * math.log(x) + b * math.log(1.0 - x))
    bt = math.exp(lb)
    if x < (a + 1.0) / (a + b + 2.0):
        return bt * betacf(a, b, x) / a
    return 1.0 - bt * betacf(b, a, 1.0 - x) / b


def t_two_tail(t, v):
    return betai(v / 2.0, 0.5, v / (v + t * t))


def t_cdf(t, v):
    half = 0.5 * t_two_tail(t, v)
    return 1.0 - half if t > 0 else half


def t_crit(alpha, v):
    lo, hi = 0.0, 1000.0
    for _ in range(200):
        mid = 0.5 * (lo + hi)
        if t_two_tail(mid, v) > alpha:
            lo = mid
        else:
            hi = mid
    return 0.5 * (lo + hi)

N01 = NormalDist()


def mean(v):
    return math.fsum(v) / len(v)


def quantile(sv, q):
    pos = q * (len(sv) - 1)
    i = int(pos)
    if i >= len(sv) - 1:
        return sv[-1]
    return sv[i] + (pos - i) * (sv[i + 1] - sv[i])


def intervals(x, nboot, rng, tc):
    n = len(x)
    m = mean(x)
    s = math.sqrt(math.fsum((v - m) ** 2 for v in x) / (n - 1))
    out = [(m - tc * s / math.sqrt(n), m + tc * s / math.sqrt(n))]
    bm = sorted(mean(rng.choices(x, k=n)) for _ in range(nboot))
    out.append((quantile(bm, 0.025), quantile(bm, 0.975)))
    below = len([v for v in bm if v < m])
    z0 = N01.inv_cdf(min(max(below / nboot, 1e-6), 1 - 1e-6))
    jack = [mean(x[:i] + x[i + 1:]) for i in range(n)]
    jm = mean(jack)
    d = [jm - v for v in jack]
    den = math.fsum(v ** 2 for v in d) ** 1.5
    a = math.fsum(v ** 3 for v in d) / (6.0 * den) if den > 0 else 0.0
    ends = []
    for q in (0.025, 0.975):
        zq = N01.inv_cdf(q)
        ends.append(quantile(bm, N01.cdf(z0 + (z0 + zq)
                                         / (1 - a * (z0 + zq)))))
    out.append(tuple(ends))
    return out


if __name__ == "__main__":
    rng = random.Random(20261119)
    reps, nboot = 500, 500
    print("coverage of nominal 0.95 intervals, {} data sets".format(reps))
    print("  population   n    t      pct    BCa   | pct all below mu")
    for name, draw, mu in (
            ("exp(1)", lambda r: r.expovariate(1.0), 1.0),
            ("logn(0,1)", lambda r: r.lognormvariate(0.0, 1.0),
             math.exp(0.5))):
        for n in (5, 10, 20):
            tc = t_crit(0.05, n - 1)
            cov = [0, 0, 0]
            miss_hi = 0
            for _ in range(reps):
                x = [draw(rng) for _ in range(n)]
                ivs = intervals(x, nboot, rng, tc)
                for j, (lo, hi) in enumerate(ivs):
                    cov[j] += lo <= mu <= hi
                miss_hi += ivs[1][1] < mu
            print("  {:10s} {:3d}  {:.3f}  {:.3f}  {:.3f} |   {:.3f}".format(
                name, n, cov[0] / reps, cov[1] / reps, cov[2] / reps,
                miss_hi / reps))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《自助法与置换检验、小样本门店实验不靠正态假设的重抽样推断母典》第 4 章「自助法的区间:百分位、偏差校正加速与小样本的覆盖不足」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-308 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026308,
  author    = {Fan, Yafu},
  title     = {The Bootstrap and Permutation Tests: Resampling Inference for Small Shop Experiments without Normality Assumptions},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论