关于「实践:四步核查与预注册实验方案」,《本草的分子与风土》给出的定量结论是什么?

结论:12 名评价员时,检验会漏掉大多数真实但不大的差别;36 名时,一半评价员能察觉的差别几乎每次都能检出。对日常来货核查,24 名评价员是合理的折中;对重要决定,应当用更多人。

数据口径:本书不是医疗建议,不主张任何植物或提取物能治疗或预防任何人类疾病。 书中以标注框区分已发表的证据、可运行的 Python 计算(36 段,全部核对)、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: 12, 24, 36

nneedpower if 30 % can tellpower if 50 % can tell
12826 %63 %
241355 %93 %
361872 %99 %
482288 %100 %
602792 %100 %
14/24 correct: one-sided p = 0.0103
查看完整验算程序
from math import comb

def tail(n, x, p):
    """P(at least x correct out of n) when each is correct with probability p."""
    return sum(comb(n, k) * p ** k * (1 - p) ** (n - k) for k in range(x, n + 1))

def critical(n, alpha=0.05):
    """Fewest correct answers that show a difference (guessing rate 1/3)."""
    return next(x for x in range(n + 1) if tail(n, x, 1 / 3) <= alpha)

print(" n   need   power if 30 % can tell   power if 50 % can tell")
for n in (12, 24, 36, 48, 60):
    x = critical(n)
    powers = [tail(n, x, 1 / 3 + 2 / 3 * pd) for pd in (0.3, 0.5)]
    print(f"{n:2d}   {x:4d}   {100 * powers[0]:22.0f} %   {100 * powers[1]:22.0f} %")

# One panel's result (invented): 24 assessors, 14 picked the odd sample.
n, correct = 24, 14
print(f"{correct}/{n} correct: one-sided p = {tail(n, correct, 1 / 3):.4f}")

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《本草的分子与风土》第 8 章「实践:四步核查与预注册实验方案」(版本 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23693057

DOI: 10.5281/zenodo.23693057 · 阅读本章所在专著 · 5DL-B-007 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026007,
  author    = {Fan, Yafu},
  title     = {Molecules and Terroir},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  doi       = {10.5281/zenodo.23693057},
  url       = {https://doi.org/10.5281/zenodo.23693057}
}

本书其他章节结论