关于「一项被撤回的截止期实验:等距截止期优势的证据与反证」,《截止期与时间动机》给出的定量结论是什么?
结论:第一章把原实验与重复实验放在同一把尺子上:重复实验找错数的方差分析统计量只有 0.181,与之相容的最大组间差距是 0.132 个标准差;即使模型偏向截止期,单一期限组与等距组的得分比最低也只到 0.885,而原论文报告的是 0.522。
数据口径:书中参数为示意;全部计算由书中程序生成;本书不是个人化的经营、法律或理财建议。 书中以标注框区分已发表的证据、可运行的 Python 计算(18 段程序)、原典与作者的假说。
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: 0.132, 0.181, 0.522
| 2002 study: corrections 136.1 vs 71.1, ratio 0.522 | |
| d | 2.5 with n = 20 per arm -> t = 7.91, r = 0.781 |
| 2026 replication, 41.3 per arm on average: | |
| metric F p eta^2 largest |d| allowed | |
| errors | 0.181 0.835 0.0030 0.132 |
| days late | 0.353 0.703 0.0058 0.185 |
| payment | 0.215 0.807 0.0035 0.144 |
| power of the replication, pairwise, alpha 0.05: | |
| d | 2.5 power 1.000 |
| d | 0.8 power 0.953 |
| d | 0.5 power 0.623 |
| d | 0.3 power 0.276 |
| self-set deadlines fall on days 9.4, 13.2, 16.8 of 21 |
查看完整验算程序
# Chapter 1. p11: The 2002 deadline claim and its 2026 replication on one
# scale. Sources: Ariely & Wertenbroch 2002, Study 2 (retracted 2026):
# n = 20 per arm, 136.1 vs 71.1 corrections; effect d = 2.5 and r = .79
# as computed by Data Colada post 138. Hyndman & Bisin 2026: N = 124,
# one-way ANOVA F(2, 121) = 0.181 (errors), 0.353 (days late), 0.215
# (payment); 93.33 % and 89.33 % self-imposed deadlines; mean self-set
# deadlines 11.6, 7.8 and 4.2 days before the end of the 21 days.
import math
def phi(z):
return 0.5 * (1.0 + math.erf(z / math.sqrt(2.0)))
def f2_sf(f, df2):
# survival function of F(2, df2), exact closed form
return (df2 / (df2 + 2.0 * f)) ** (df2 / 2.0)
def power(d, n):
# two-sided 5 % test of two arms with n each, normal approximation
return phi(d * math.sqrt(n / 2.0) - 1.959964)
if __name__ == "__main__":
d0, n0 = 2.5, 20
print("2002 study: corrections 136.1 vs 71.1, ratio %.3f"
% (71.1 / 136.1))
print(" d = %.1f with n = %d per arm -> t = %.2f, r = %.3f"
% (d0, n0, d0 * math.sqrt(n0 / 2.0), d0 / math.sqrt(d0 ** 2 + 4)))
n1 = 124 / 3.0
print("2026 replication, %.1f per arm on average:" % n1)
print(" metric F p eta^2 largest |d| allowed")
for name, f in (("errors", 0.181), ("days late", 0.353),
("payment", 0.215)):
eta = 2 * f / (2 * f + 121)
dmax = math.sqrt(4 * f / n1)
print(" %-10s %6.3f %6.3f %7.4f %6.3f"
% (name, f, f2_sf(f, 121), eta, dmax))
print("power of the replication, pairwise, alpha 0.05:")
for d in (2.5, 0.8, 0.5, 0.3):
print(" d = %.1f power %.3f" % (d, power(d, n1)))
days = [21 - g for g in (11.6, 7.8, 4.2)]
print("self-set deadlines fall on days %.1f, %.1f, %.1f of 21"
% tuple(days))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《双曲贴现偏好反转、截止期承诺装置与时间动机方程母典》第 1 章「一项被撤回的截止期实验:等距截止期优势的证据与反证」(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-199 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026199,
author = {Fan, Yafu},
title = {Hyperbolic-Discounting Preference Reversals, Deadline Commitment Devices and the Temporal Motivation Equation},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}