关于「先证伪,后立界:贝叶斯理由及其限度」,《计算著作学与自校验专著》给出的定量结论是什么?
结论:第三章给出「先证伪,后立界」的贝叶斯理由:似然比为 12 时,信任变成三倍需要先验不高于 0.273;只发表显著小型研究时,真实比值 1.35 的发表中位数是 11.89 倍。
数据口径:书中的读者、书、错误率与成本均为示意模型;全部计算由书中程序生成;本书不是法律、财务或出版建议。 书中以标注框区分已发表的证据、可运行的 Python 计算(16 段程序)、原典与作者的假说。
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。 与结论吻合的打印值: 0.273, 12
| prior P(C) | LR | P(C) after | multiplier |
|---|---|---|---|
| signal: falsify+bound (LR 12.00) | |||
| 0.05 | 12.00 | 0.387 | 7.74 |
| 0.10 | 12.00 | 0.571 | 5.71 |
| 0.20 | 12.00 | 0.750 | 3.75 |
| 0.50 | 12.00 | 0.923 | 1.85 |
| 0.80 | 12.00 | 0.980 | 1.22 |
| signal: imitated (LR 2.00) | |||
| 0.05 | 2.00 | 0.095 | 1.90 |
| 0.10 | 2.00 | 0.182 | 1.82 |
| 0.20 | 2.00 | 0.333 | 1.67 |
| 0.50 | 2.00 | 0.667 | 1.33 |
| 0.80 | 2.00 | 0.889 | 1.11 |
| signal: self-praise (LR 0.33) | |||
| 0.05 | 0.33 | 0.017 | 0.34 |
| 0.10 | 0.33 | 0.036 | 0.36 |
| 0.20 | 0.33 | 0.077 | 0.38 |
| 0.50 | 0.33 | 0.250 | 0.50 |
| 0.80 | 0.33 | 0.571 | 0.71 |
| a 3x jump with LR 12 needs a prior at or below 0.273 | |||
| a 5x jump with LR 12 needs a prior at or below 0.127 |
查看完整验算程序
# Chapter 3. p05: The Bayesian arithmetic of "falsify first". A reader
# weighs two kinds of author: calibrated (C) and hyping (H). An opening
# that refutes a popular overclaim in the author's own field, and states
# where the corrected result holds, is a signal: C does it with chance
# 0.60, H with chance 0.05 (likelihood ratio 12). Self-praise is the
# reverse signal: C 0.30, H 0.90 (ratio 1/3). If hypers learn to imitate
# the opening, the ratio falls. We print P(C) before and after, and the
# multiplier, for several priors.
def post(prior, lr):
odds = prior / (1 - prior) * lr
return odds / (1 + odds)
if __name__ == "__main__":
print(" prior P(C) LR P(C) after multiplier")
for lr, label in ((12.0, "falsify+bound"), (2.0, "imitated"),
(1 / 3, "self-praise")):
print(" signal: {} (LR {:.2f})".format(label, lr))
for p in (0.05, 0.10, 0.20, 0.50, 0.80):
q = post(p, lr)
print(" {:9.2f} {:5.2f} {:10.3f} {:10.2f}".format(
p, lr, q, q / p))
for m in (3, 5):
# prior needed for a multiplier m with LR 12: q = m p
lo, hi = 1e-9, 1 / m
for _ in range(200):
mid = (lo + hi) / 2
if post(mid, 12.0) / mid > m:
lo = mid
else:
hi = mid
print(" a {}x jump with LR 12 needs a prior at or below {:.3f}"
.format(m, lo))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《计算著作学、先证伪后立界修辞与自校验专著编译器》第 3 章「先证伪,后立界:贝叶斯理由及其限度」(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-101 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026101,
author = {Fan, Yafu},
title = {Computational Bibliometrics, Falsification-First Rhetoric and Self-Verifying Monograph Compilers},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}