关于「过程追踪四类检验:把线索写成似然比」,《推理轨迹识人与结果运气剥离》给出的定量结论是什么?

结论:ρ 为零时,两条线索同时出现的真实似然比是 4.000,与朴素相乘一致,后验 0.800。ρ 升到 0.25,真实似然比降到 3.024,后验 0.752;ρ 为 0.5 时是 2.519 与 0.716;ρ 为 0.75 时是 2.209 与 0.688;ρ 为一时,第二条线索纯属复读,真实似然比回到单条的 2.000,后验 0.667。朴素相乘却始终报告 0.800。

数据口径:事后解释常常是一套通俗理论:合成实验里,事后自述与真实权重的相关只有 -0.013,与旁观者凭常识的猜测却高达 0.826;同步出声思考保留了次序与跳跃,要求边做边解释则会改变被测过程。 核对文献后本书做了蓝图修正:人能表演推理,旁人判别真假的正确率只比抛硬币高一点,因此不主张「不可伪造的思维指纹」,只主张伪造成本随追问深度上升,并写出上下界与推翻条件;甄选效度以修订后的估计为锚,结构化访谈居首也只有 0.42 上下,单一方法有限,须按规则多方法合成。

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: 0.25, 0.5, 0.667, 0.688, 0.716, 0.75

two clues, both seen; prior P(H)0.5
rho true LR naive LR P(H) true P(H) naive eff. clues
0.004.000 4.000 0.800 0.800 2.00
0.253.024 4.000 0.752 0.800 1.60
0.502.519 4.000 0.716 0.800 1.33
0.752.209 4.000 0.688 0.800 1.14
1.002.000 4.000 0.667 0.800 1.00
k clues all seen, rho0.5
k true LR naive LR eff. clues
12.000 2.0 1.00
22.519 4.0 1.33
43.994 16.0 2.00
810.042 256.0 3.33
查看完整验算程序
# Chapter 2. p23: correlated clues are counted twice by naive Bayes.
# Two clues with the same margins a = 0.70 under H and b = 0.35 under R
# (single LR = 2). Dependence model: with prob rho the second clue just
# copies the first (same preparation drives both); otherwise it is an
# independent draw. Then P(both | H) = a (rho + (1 - rho) a), likewise
# under R. Naive multiplication uses a^2 / b^2 = 4 whatever rho is.
# Effective clue count = log(true joint LR) / log(single LR).
# A second block repeats the clue k times with copy prob rho = 0.5.
import math

A, B = 0.70, 0.35


def joint(p, rho, k):
    # prob that k chained clues all show, each copies the previous one
    # with prob rho, else draws fresh with margin p
    return p * (rho + (1 - rho) * p) ** (k - 1)


def post(lr):
    return lr / (1 + lr)


def main():
    single = A / B
    print("two clues, both seen; prior P(H) = 0.5")
    print("rho   true LR  naive LR  P(H) true  P(H) naive  eff. clues")
    for rho in (0.0, 0.25, 0.5, 0.75, 1.0):
        lr = joint(A, rho, 2) / joint(B, rho, 2)
        naive = single ** 2
        eff = math.log(lr) / math.log(single)
        print("{:<5.2f}{:>8.3f}{:>10.3f}{:>11.3f}{:>12.3f}{:>11.2f}".format(
            rho, lr, naive, post(lr), post(naive), eff))
    print("k clues all seen, rho = 0.5")
    print("k   true LR   naive LR  eff. clues")
    for k in (1, 2, 4, 8):
        lr = joint(A, 0.5, k) / joint(B, 0.5, k)
        print("{:<3d}{:>8.3f}{:>11.1f}{:>10.2f}".format(
            k, lr, single ** k, math.log(lr) / math.log(single)))


main()

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《思维脉络溯源识人、过程追踪博弈与剥离意图结果假象母典》第 2 章「过程追踪四类检验:把线索写成似然比」(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-385 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026385,
  author    = {Fan, Yafu},
  title     = {Reasoning-Trajectory Profiling, Process-Tracing Games and Separating Cognitive Structure from Stated Intentions and Outcome Luck},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论