关于「相关误差与上限:贝塔二项模型、意见领袖与同源共错」,《相关误差下的多模型投票》给出的定量结论是什么?

结论:只要错误有共同来源,多数准确率就有一个低于 1 的上限。这个上限在贝塔二项模型里等于难度低于一半的题所占的比例,在意见领袖模型里取决于跟随概率是否越过 \(1-1/(2p)\)。弱相关时,越投越准仍然成立,但收益被打了折扣;相关稍强时,上限远低于 1,再多的票也只是在原地踏步,甚至越投越差。下一章讨论一个实际问题:店主手里的模型池,相关系数到底是多少,怎样从数据里量出来。

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

A. majority accuracy at p0.70, n odd, by correlation rho
rho n1 n=5 n=15 n=51 n=301 limit
0.00.7000 0.8369 0.9500 0.9986 1.0000 1.0000
0.050.7000 0.8132 0.8927 0.9409 0.9607 0.9649
0.10.7000 0.7940 0.8507 0.8829 0.8966 0.8996
0.20.7000 0.7649 0.7955 0.8107 0.8167 0.8180
0.30.7000 0.7443 0.7617 0.7695 0.7725 0.7731
0.50.7000 0.7186 0.7243 0.7266 0.7274 0.7276
B. Beta parameters behind each row
rho 0.05 a13.300 b = 5.700
rho 0.1 a6.300 b = 2.700
rho 0.2 a2.800 b = 1.200
rho 0.3 a1.633 b = 0.700
rho 0.5 a0.700 b = 0.300
C. check: rho0.3 limit by midpoint rule, 200000 cells
continued fraction 0.7731, midpoint 0.7731
查看完整验算程序
# Chapter 2. p21: majority accuracy when errors are correlated.
# Model: exchangeable Beta-binomial (the Polya-Eggenberger family used
# for dependent juries). On each question a shared difficulty theta is
# drawn from Beta(a, b); given theta the n voters are right
# independently with probability theta. Mean accuracy p = a / (a + b);
# correlation of any two voters' correctness rho = 1 / (a + b + 1).
# As n grows the majority is right exactly when theta > 1/2, so the
# limit is P(theta > 1/2) = 1 - I(1/2; a, b), not 1. Exact finite-n
# tails by log-gamma terms and math.fsum; the incomplete beta by the
# Lentz continued fraction. p and rho are illustrative.
import math


def lbeta(a, b):
    return math.lgamma(a) + math.lgamma(b) - math.lgamma(a + b)


def ab(p, rho):
    s = 1 / rho - 1
    return p * s, (1 - p) * s


def bb_major(n, p, rho):
    if rho == 0:
        lp, lq = math.log(p), math.log(1 - p)
        terms = [math.exp(math.lgamma(n + 1) - math.lgamma(k + 1)
                          - math.lgamma(n - k + 1) + k * lp
                          + (n - k) * lq)
                 for k in range(n // 2 + 1, n + 1)]
        return math.fsum(terms)
    a, b = ab(p, rho)
    terms = [math.exp(math.lgamma(n + 1) - math.lgamma(k + 1)
                      - math.lgamma(n - k + 1) + lbeta(k + a, n - k + b)
                      - lbeta(a, b))
             for k in range(n // 2 + 1, n + 1)]
    return math.fsum(terms)


def betacf(a, b, x):
    tiny = 1e-300
    c, d = 1.0, 1.0 - (a + b) * x / (a + 1)
    d = 1.0 / (d if abs(d) > tiny else tiny)
    h = d
    for m in range(1, 400):
        m2 = 2 * m
        for num in (m * (b - m) * x / ((a + m2 - 1) * (a + m2)),
                    -(a + m) * (a + b + m) * x / ((a + m2) * (a + m2 + 1))):
            d = 1.0 + num * d
            d = 1.0 / (d if abs(d) > tiny else tiny)
            c = 1.0 + num / c
            c = c if abs(c) > tiny else tiny
            h *= d * c
        if abs(d * c - 1.0) < 1e-15:
            break
    return h


def betai(a, b, x):
    if x <= 0:
        return 0.0
    if x >= 1:
        return 1.0
    lead = math.exp(a * math.log(x) + b * math.log(1 - x) - lbeta(a, b))
    if x < (a + 1) / (a + b + 2):
        return lead * betacf(a, b, x) / a
    return 1.0 - lead * betacf(b, a, 1 - x) / b


def limit(p, rho):
    if rho == 0:
        return 1.0 if p > 0.5 else 0.0
    a, b = ab(p, rho)
    return 1.0 - betai(a, b, 0.5)


if __name__ == "__main__":
    p = 0.7
    print("A. majority accuracy at p = 0.70, n odd, by correlation rho")
    print("   rho      n=1     n=5    n=15    n=51   n=301   limit")
    for rho in (0.0, 0.05, 0.1, 0.2, 0.3, 0.5):
        cells = [bb_major(n, p, rho) for n in (1, 5, 15, 51, 301)]
        print("  {:<5} {}  {:.4f}".format(
            rho, " ".join("{:7.4f}".format(c) for c in cells),
            limit(p, rho)))
    print("B. Beta parameters behind each row")
    for rho in (0.05, 0.1, 0.2, 0.3, 0.5):
        a, b = ab(p, rho)
        print("  rho {:<5} a = {:6.3f}  b = {:6.3f}".format(rho, a, b))
    print("C. check: rho = 0.3 limit by midpoint rule, 200000 cells")
    a, b = ab(p, 0.3)
    h = 0.5 / 200000
    mids = [math.exp((a - 1) * math.log(0.5 + (i + 0.5) * h)
                     + (b - 1) * math.log(0.5 - (i + 0.5) * h)
                     - lbeta(a, b)) * h for i in range(200000)]
    print("  continued fraction {:.4f}, midpoint {:.4f}".format(
        limit(p, 0.3), math.fsum(mids)))

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 2 章「相关误差与上限:贝塔二项模型、意见领袖与同源共错」(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