关于「从 0.62 到 0.04:自我损耗的预注册复制」,《自我损耗与饥饿法官》给出的定量结论是什么?
结论:第一章复算两次预注册多实验室检验:2016 年的 23 个实验室得到 0.04,2021 年的 36 个实验室得到 0.06,粗略合并为 0.0547。
数据口径:书中参数为示意;全部计算由书中程序生成;本书不是心理或医疗建议,书中流程与工具为非个人化经营建议。 本书把两项证据放到预注册复制、元分析偏差检验与案件排序模拟面前,看它们还剩多少。
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: 0.04, 0.0547, 0.06, 2016, 2021
study d SE weight share 2016 RRR 0.04 0.0561 0.265 2021 MSP 0.06 0.0337 0.735 fixed-effect pool d 0.0547, SE 0.0289, 95% CI -0.0019 to 0.1113 z 1.90, two-sided p 0.058 share of the 2010 estimate that survives: 0.088 d 0.6200: P(depleted worse than control) 0.669 d 0.0547: P(depleted worse than control) 0.515 d 0.0800: P(depleted worse than control) 0.523 per-group n for 80% power at the pooled d: 5245
查看完整验算程序
# Chapter 1. p12: Pooling the two preregistered multi-lab tests.
# Published inputs: Hagger et al. (2016) d = 0.04, CI -0.07 to 0.15;
# Vohs et al. (2021) 36 labs, N = 3531, confirmatory d = 0.06 and
# exploratory full-sample d = 0.08. The 2021 standard error is our
# approximation sqrt(4 / N); the pooled value is a fixed-effect sketch,
# not a published estimate.
# 两次预注册多实验室检验的合并:示意计算。
import math
from statistics import NormalDist
Z = NormalDist()
Z975 = Z.inv_cdf(0.975)
D16, LO16, HI16 = 0.04, -0.07, 0.15
D21, N21, LABS21, D21_ALL = 0.06, 3531, 36, 0.08
D_META = 0.62
se16 = (HI16 - LO16) / (2 * Z975)
se21 = math.sqrt(4 / N21)
w = [1 / se16 ** 2, 1 / se21 ** 2]
dd = [D16, D21]
pool = math.fsum(wi * di for wi, di in zip(w, dd)) / math.fsum(w)
se_p = 1 / math.sqrt(math.fsum(w))
zp = pool / se_p
print("study d SE weight share")
for name, d, s, wi in (("2016 RRR", D16, se16, w[0]),
("2021 MSP", D21, se21, w[1])):
print("{:9s} {:.2f} {:.4f} {:.3f}".format(
name, d, s, wi / math.fsum(w)))
print("fixed-effect pool d {:.4f}, SE {:.4f}, 95% CI {:.4f} to {:.4f}"
.format(pool, se_p, pool - Z975 * se_p, pool + Z975 * se_p))
print("z {:.2f}, two-sided p {:.3f}".format(
zp, 2 * (1 - Z.cdf(abs(zp)))))
print("share of the 2010 estimate that survives: {:.3f}"
.format(pool / D_META))
for d in (D_META, pool, D21_ALL):
# probability that a random depleted person scores worse than a
# random control (common-language effect size)
print("d {:.4f}: P(depleted worse than control) {:.3f}".format(
d, Z.cdf(d / math.sqrt(2))))
n = math.ceil(2 * (Z975 + Z.inv_cdf(0.8)) ** 2 / pool ** 2)
print("per-group n for 80% power at the pooled d: {}".format(n))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《自我损耗大规模复制失败、法官饥饿效应排序伪影与决策疲劳日程母典》第 1 章「从 0.62 到 0.04:自我损耗的预注册复制」(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-196 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026196,
author = {Fan, Yafu},
title = {The Failed Replication of Ego Depletion, the Ordering Artefact behind the Hungry Judges and Decision-Fatigue Scheduling},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}