关于「孔多塞的二项多数:独立投票为什么越投越准」,《相关误差下的多模型投票》给出的定量结论是什么?

结论:A 段的五列分别对应单票准确率 0.45、0.55、0.60、0.70 与 0.80。看 0.70 这一列:一票时是 0.7000,三票时升到 0.7840,五票时 0.8369,十五票时 0.9500,五十一票时 0.9986。只要投票者够多,多数意见几乎不会出错。这就是孔多塞定理的正面:独立的、略好于抛硬币的判断,积少成多,可以变成近乎确定的判断。

数据口径:书中参数为示意;全部计算由书中程序生成;本书是非个人化的经营分析,不构成经营建议。 书中以标注框区分已发表的证据、可运行的 Python 计算(19 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: 0.45, 0.55, 0.60, 0.70, 0.7000, 0.7840

A. majority accuracy, n independent voters (exact)
n p0.45 p=0.55 p=0.60 p=0.70 p=0.80
10.4500 0.5500 0.6000 0.7000 0.8000
30.4253 0.5747 0.6480 0.7840 0.8960
50.4069 0.5931 0.6826 0.8369 0.9421
90.3786 0.6214 0.7334 0.9012 0.9804
150.3465 0.6535 0.7869 0.9500 0.9958
250.3063 0.6937 0.8462 0.9825 0.9996
510.2359 0.7641 0.9265 0.9986 1.0000
1010.1562 0.8438 0.9791 1.0000 1.0000
B. gain from two more voters at p0.70
13: +0.0840
35: +0.0529
57: +0.0370
911: +0.0206
1517: +0.0097
C. even juries: n2k with a coin for ties equals n = 2k - 1
n4: 0.784000 n = 3: 0.784000
n10: 0.901191 n = 9: 0.901191

核心公式

\[P_n = \sum_{k > n/2} \binom{n}{k} p^k (1-p)^{n-k}.\]
查看完整验算程序
# Chapter 1. p11: exact majority accuracy of n independent voters.
# Model: Condorcet's binomial case. Each of n voters (n odd) is right
# with the same probability p, independently of the others, on a yes/no
# question; the majority is right when more than n/2 are right.
# Exact rational arithmetic with fractions.Fraction; p values are
# illustrative accuracies of a single model, not measurements.
from fractions import Fraction
import functools
import math
import operator


def majority(n, p):
    # P(more than n/2 of n independent voters are right), exact
    q = 1 - p
    terms = [math.comb(n, k) * p ** k * q ** (n - k)
             for k in range(n // 2 + 1, n + 1)]
    return functools.reduce(operator.add, terms, Fraction(0))


if __name__ == "__main__":
    ps = [Fraction(45, 100), Fraction(55, 100), Fraction(6, 10),
          Fraction(7, 10), Fraction(8, 10)]
    print("A. majority accuracy, n independent voters (exact)")
    print("    n   p=0.45   p=0.55   p=0.60   p=0.70   p=0.80")
    for n in (1, 3, 5, 9, 15, 25, 51, 101):
        row = "  ".join("{:.4f}".format(float(majority(n, p)))
                        for p in ps)
        print("  {:>3}   {}".format(n, row))
    print("B. gain from two more voters at p = 0.70")
    p = Fraction(7, 10)
    for n in (1, 3, 5, 9, 15):
        g = majority(n + 2, p) - majority(n, p)
        print("  {:>2} -> {:>2}: +{:.4f}".format(n, n + 2, float(g)))
    print("C. even juries: n = 2k with a coin for ties equals n = 2k - 1")
    for k in (2, 5):
        n = 2 * k
        tie = math.comb(n, k) * p ** k * (1 - p) ** k
        even = majority(n, p) + tie / 2
        print("  n = {:>2}: {:.6f}   n = {:>2}: {:.6f}".format(
            n, float(even), n - 1, float(majority(n - 1, p))))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《孔多塞陪审团定理与相关误差、多模型投票何时越投越准何时同源共错母典》第 1 章「孔多塞的二项多数:独立投票为什么越投越准」(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-309 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026309,
  author    = {Fan, Yafu},
  title     = {Condorcet's Jury Theorem with Correlated Errors: When Multi-Model Voting Gets More Accurate and When Shared Origins Make Everyone Wrong Together},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论