关于「动机性折扣:贝塔先验、伪样本量与证据权重」,《先验消解与反逆火沟通》给出的定量结论是什么?

结论:固执有两种:先验强与折扣重。二者在单人轨迹上无法区分,因为均值只依赖 \(w/n_0\)。伪样本量 80、打两成折扣的人,第一件材料只挪动 0.0020;二十四个正折扣格子里,只有 3 个能在一次沟通内把信念推过 0.5。分清打折与逆火需要的样本,在差别只有一两分时动辄数百人,而按期望算出的人数只有一半把握。

数据口径:书中参数为示意,全部计算由书中程序生成;本书非心理咨询、不提供任何临床服务、非个人化经营建议,例子只用商品与消费认知,不涉及政治、宗教与族群议题。 第六章把追问、肯定、证据与思想实验写成马尔可夫决策过程,示意参数下最难起点的期望收益 10.555,参数扰动两千次,完整路线不变的比例为 0.8270,沿用基准策略的平均相对遗憾只有 0.0003,所以只称模拟最优。

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: 0.0020, 80

posterior mean of 'fake' after k genuine items (m00.8)
n0 w k1 k=5 k=10 k=20 first step
51.00 0.6667 0.4000 0.2667 0.1600 0.1333
50.50 0.7273 0.5333 0.4000 0.2667 0.0727
50.20 0.7692 0.6667 0.5714 0.4444 0.0308
201.00 0.7619 0.6400 0.5333 0.4000 0.0381
200.50 0.7805 0.7111 0.6400 0.5333 0.0195
200.20 0.7921 0.7619 0.7273 0.6667 0.0079
801.00 0.7901 0.7529 0.7111 0.6400 0.0099
800.50 0.7950 0.7758 0.7529 0.7111 0.0050
800.20 0.7980 0.7901 0.7805 0.7619 0.0020
only w / n0 is identified: m_k = m0 / (1 + k w / n0)
k1 n0=20, w=1: 0.7619 n0=4, w=0.2: 0.7619
k10 n0=20, w=1: 0.5333 n0=4, w=0.2: 0.5333
ratio of 10th-step to 1st-step size, n05:
w1.00: 0.1429
w0.20: 0.5462
查看完整验算程序
# Chapter 2. p21: a Beta prior with a motivated discount weight w.
# Belief: "parcels from this cross-border buying channel are fakes".
# Prior Beta(a, b) with mean m0 = 0.8 and pseudo-sample n0 = a + b
# (the stubbornness of the prior). Each verified-genuine item is one
# observation against the claim, counted with weight w in (0, 1]:
# b <- b + w. Posterior mean after k items: a / (n0 + w k).
# Update size of item j = drop of the posterior mean caused by item j.
# Parameters are illustrative; no real customer is described.
from fractions import Fraction as F

M0 = F(4, 5)


def post_mean(n0, w, k):
    a = M0 * n0
    return a / (n0 + w * k)


if __name__ == "__main__":
    ks = (1, 5, 10, 20)
    print("posterior mean of 'fake' after k genuine items (m0 = 0.8)")
    print("   n0     w    k=1     k=5    k=10    k=20  first step")
    for n0 in (5, 20, 80):
        for w in (F(1), F(1, 2), F(1, 5)):
            row = [float(post_mean(n0, w, k)) for k in ks]
            step = float(M0 - post_mean(n0, w, 1))
            print("  {:3d}  {:4.2f}  {:6.4f}  {:6.4f}  {:6.4f}  {:6.4f}"
                  "  {:7.4f}".format(n0, float(w), *row, step))
    print("only w / n0 is identified: m_k = m0 / (1 + k w / n0)")
    for k in (1, 10):
        a = float(post_mean(20, F(1), k))
        b = float(post_mean(4, F(1, 5), k))
        print("   k={:2d}  n0=20, w=1: {:.4f}   n0=4, w=0.2: {:.4f}".format(
            k, a, b))
    print("ratio of 10th-step to 1st-step size, n0=5:")
    for w in (F(1), F(1, 5)):
        s1 = post_mean(5, w, 0) - post_mean(5, w, 1)
        s10 = post_mean(5, w, 9) - post_mean(5, w, 10)
        print("   w = {:.2f}: {:.4f}".format(float(w), float(s10 / s1)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《顽固先验偏见消解、解释深度错觉破除与反逆火效应沟通母典》第 2 章「动机性折扣:贝塔先验、伪样本量与证据权重」(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-387 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026387,
  author    = {Fan, Yafu},
  title     = {Dissolving Entrenched Priors, Shattering the Illusion of Explanatory Depth and Anti-Backfire Persuasion Mechanics},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论