关于「看校准,不看自信:Brier 分数、Murphy 分解与达克效应图的统计伪迹」,《解释深度剃刀与校准》给出的定量结论是什么?

结论:第二章讲 Brier 分数与墨菲三分解、九成区间的真实命中率(区间只有应有宽度一半时命中率 0.589),以及四分位图的统计伪迹。

数据口径:书中参数为示意,数据全部合成,全部计算由书中程序生成。 第一章在合成数据里量出装置类自评下降 1.283 分,事实类下降区间从 -0.263 到 0.063,跨过零,并给出剃刀与检索的分流规则。

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: 0.589

k = s_stated / s_trueexact hitsimulated hitmisses of 400
0.250.3190.295282
0.400.4890.495202
0.500.5890.590164
0.700.7500.743103
1.000.9000.90538
1.300.9680.96315
observed hit 0.50 -> stated ranges are 0.410 of true sd; widen by 2.44
observed hit 0.60 -> stated ranges are 0.512 of true sd; widen by 1.95
observed hit 0.70 -> stated ranges are 0.630 of true sd; widen by 1.59
查看完整验算程序
# Chapter 2. p22: how often do "90 per cent" ranges contain the truth?
# A judge states a 90% range m +- 1.645 s_stated for a quantity whose
# real error is normal with sd s_true. If s_stated = k s_true, the true
# hit rate is 2 Phi(1.645 k) - 1. We also simulate 400 estimates of
# monthly order counts (log scale) to show the sampling noise around it.
# Shrink factors are illustrative; Alpert & Raiffa (1982) and Lichtenstein
# et al. (1982) report hit rates far below the nominal level.
import math
import random
from statistics import NormalDist

Z = NormalDist()
Z90 = 1.644854
rng = random.Random(39622)
N = 400
print("k = s_stated / s_true   exact hit   simulated hit   misses of 400")
for k in (0.25, 0.40, 0.50, 0.70, 1.00, 1.30):
    exact = 2.0 * Z.cdf(Z90 * k) - 1.0
    hits = 0
    for _ in range(N):
        err = rng.gauss(0.0, 1.0)
        if abs(err) <= Z90 * k:
            hits += 1
    print("        {:4.2f}            {:6.3f}      {:6.3f}         {:4d}"
          .format(k, exact, hits / N, N - hits))
# widening needed: observed hit rate h implies k = z((1+h)/2) / 1.645
for h in (0.50, 0.60, 0.70):
    k = Z.inv_cdf((1.0 + h) / 2.0) / Z90
    print("observed hit {:.2f} -> stated ranges are {:.3f} of true sd;"
          " widen by {:.2f}".format(h, k, 1.0 / k))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《解释深度错觉破除、认知边界校准与真知假知剥离剃刀母典》第 2 章「看校准,不看自信:Brier 分数、Murphy 分解与达克效应图的统计伪迹」(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-396 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026396,
  author    = {Fan, Yafu},
  title     = {Shattering the Illusion of Explanatory Depth, Epistemic Calibration and the Razor Separating True Knowledge from Assumed Familiarity},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论