关于「剥离结果运气:成败记录、序贯检验与多方法合成」,《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:成败记录混合了决策质量与运气。好决策者与差决策者的成功率相差一成时,十次决策下差者战绩反超的概率仍有三成,要压到一成以下需要八十多次;序贯检验能省一半,仍需一百多次。经典甄选效度经修订后下调,结构化访谈居首也只有 0.42 上下;多方法按规则合成优于整体印象,推理图评分的增量取决于它与其他方法的相关,最乐观时也不大。所以,结果成败单独记、永不进推理评分,推理评分只作合成中的一项。
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。
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: 0.42
| anchors: structured 1998 0.51, revised 2022 0.42 | |
| validity structured conversation 0.42 | |
| validity work sample | 0.30 |
| validity reasoning graph | 0.25 |
| rho R opt(3) r unit(3) R opt(2) gain of graph | |
| 0.1 | 0.528 0.511 0.494 0.034 |
| 0.3 | 0.467 0.443 0.458 0.009 |
| 0.5 | 0.433 0.396 0.433 0.000 |
| variance explained, best case rho 0.1 | 0.278 |
查看完整验算程序
# Chapter 6. p63: what a multi-method composite can and cannot add.
# Validity anchors: structured interview 0.51 (Schmidt and Hunter 1998)
# and 0.42 after the range-restriction revision (Sackett et al. 2022).
# A work-sample exercise at 0.30 and a reasoning-graph score at 0.25
# are illustrative, as are the common intercorrelations 0.1 / 0.3 /
# 0.5. Optimal composite R = sqrt(v' Rxx^-1 v); unit-weighted
# composite r = sum(v) / sqrt(sum of all Rxx entries). Increment of the
# reasoning score = R(all three) - R(interview and work sample).
import math
V_SI_1998, V_SI_2022 = 0.51, 0.42
V = [V_SI_2022, 0.30, 0.25]
NAMES = ["structured conversation", "work sample", "reasoning graph"]
def solve(a, b):
n = len(b)
m = [row[:] + [b[i]] for i, row in enumerate(a)]
for c in range(n):
piv = max(range(c, n), key=lambda r: abs(m[r][c]))
m[c], m[piv] = m[piv], m[c]
for r in range(n):
if r != c:
f = m[r][c] / m[c][c]
m[r] = [x - f * y for x, y in zip(m[r], m[c])]
return [m[i][n] / m[i][i] for i in range(n)]
def opt_r(v, rxx):
w = solve(rxx, v)
return math.sqrt(math.fsum(a * b for a, b in zip(w, v)))
def unit_r(v, rxx):
tot = math.fsum(x for row in rxx for x in row)
return math.fsum(v) / math.sqrt(tot)
def mat(k, rho):
return [[1.0 if i == j else rho for j in range(k)] for i in range(k)]
def main():
print("anchors: structured 1998 {:.2f}, revised 2022 {:.2f}".format(
V_SI_1998, V_SI_2022))
for n, v in zip(NAMES, V):
print("validity {:<24}{:.2f}".format(n, v))
print("rho R opt(3) r unit(3) R opt(2) gain of graph")
for rho in (0.1, 0.3, 0.5):
r3, u3 = opt_r(V, mat(3, rho)), unit_r(V, mat(3, rho))
r2 = opt_r(V[:2], mat(2, rho))
print("{:<5.1f}{:>9.3f}{:>11.3f}{:>10.3f}{:>15.3f}".format(
rho, r3, u3, r2, r3 - r2))
print("variance explained, best case rho 0.1: {:.3f}".format(
opt_r(V, mat(3, 0.1)) ** 2))
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 6 章「剥离结果运气:成败记录、序贯检验与多方法合成」(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}
}