关于「东西方原典考辨:三人成虎与邦费罗尼的不等式」,《伪发现率与文案规则检验》给出的定量结论是什么?
结论:第七章考辨《战国策》「三人言而成虎」与邦费罗尼 1936 年的不等式:三份独立报告把后验推到 0.4219,同源转述则停在 0.0089。
数据口径:书中转化率、访客数与规则条数均为示意参数;全部计算由书中程序生成;本书属于非个人化经营建议。 本书用精确概率推翻这条常识:四十条全部无效的规则同时检验,平均每周冒出两条假冠军,至少一条的概率为 0.8715;模拟中每周最佳假冠军宣称的相对提升平均为 34.2 %,而真实提升为零。
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: 0.0089, 0.4219
| likelihood ratio of one report: 9 | ||
| reports | independent | all copies of one |
| 1 | 0.0089 | 0.0089 |
| 2 | 0.0750 | 0.0089 |
| 3 | 0.4219 | 0.0089 |
| 5 | 0.9834 | 0.0089 |
| independent reports to pass even odds: 4 | ||
| copy share c | effective reports | posterior after 3 |
| 0 | 3.00 | 0.4219 |
| 0.25 | 2.50 | 0.1957 |
| 0.5 | 2.00 | 0.0750 |
| 0.75 | 1.50 | 0.0263 |
| 1 | 1.00 | 0.0089 |
| 40 false rumours, each 3 reports at 0.1: E[believed] = 0.040 | ||
| P(at least one believed) = 0.0392 |
查看完整验算程序
# Chapter 7. p71: "three people say there is a tiger in the market".
# Model: prior P(tiger) = 0.001; a report "tiger" comes with
# probability 0.9 if there is one and 0.1 if not (likelihood ratio 9).
# Case A: the reports are independent, so three reports multiply the
# odds by 9^3. Case B: the three speakers repeat one rumour, so the
# evidence is one report (ratio 9) heard three times. Case C: a share
# c of speakers copy the first one. Then: the king hears a stream of
# rumours about 40 different ministers, all false; how many "three
# reports" events should he expect? All numbers are illustrative.
from fractions import Fraction
import math
PRIOR = Fraction(1, 1000)
SENS, FALSE = Fraction(9, 10), Fraction(1, 10)
def posterior(lr):
odds = PRIOR / (1 - PRIOR) * lr
return odds / (1 + odds)
if __name__ == "__main__":
lr = SENS / FALSE
print("likelihood ratio of one report: {}".format(lr))
print("reports independent all copies of one")
for k in (1, 2, 3, 5):
print(" {} {:.4f} {:.4f}".format(
k, float(posterior(lr ** k)), float(posterior(lr))))
need = math.ceil(math.log(float((1 - PRIOR) / PRIOR)) /
math.log(float(lr)))
print("independent reports to pass even odds: {}".format(need))
print("copy share c effective reports posterior after 3")
for c in ("0", "0.25", "0.5", "0.75", "1"):
cf = Fraction(c)
eff = 1 + 2 * (1 - cf)
post = posterior(lr ** float(eff))
print(" {:5s} {:.2f} {:.4f}".format(
c, float(eff), float(post)))
pf = FALSE ** 3
print("40 false rumours, each 3 reports at 0.1: "
"E[believed] = {:.3f}".format(float(40 * pf)))
print("P(at least one believed) = {:.4f}".format(
float(1 - (1 - pf) ** 40)))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《多重比较与伪发现率控制、同时检验数十条文案规则的本杰明尼霍赫伯格阶梯母典》第 7 章「东西方原典考辨:三人成虎与邦费罗尼的不等式」(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-307 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026307,
author = {Fan, Yafu},
title = {Multiple Comparisons and False-Discovery-Rate Control: the Benjamini-Hochberg Step-Up for Testing Dozens of Copy Rules at Once},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}