关于「认知能量预算与自我损耗之争」,《行动的力学》给出的定量结论是什么?
结论:在这个模拟中,所有研究的平均效应正确地落在 0.05 附近,但只有约 4% 的研究达到显著;而这些被发表的研究,平均效应量高达 0.79,比 2010 年元分析的估计还高。小样本加上选择性发表,就足以从一个几乎不存在的效应中制造出一篇「稳健」的文献。这不证明自我损耗研究中一定发生了这样的选择,但说明大规模预注册重复为什么比许多小研究的汇总更可信。
数据口径:本书不是心理治疗或医疗建议,不作任何诊断或疗效声称。 书中以标注框区分已发表的证据、可运行的 Python 计算(15 段程序,全部运行核对)、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: 0.05, 0.79, 2010
| true effect d = 0.05; 4000 studies with 20 per group | |
| mean d of all studies run: | 0.05 |
| share reaching significance: | 3.9 % |
| mean d of the published ones: | 0.79 |
| for comparison: 2010 meta-analysis d = 0.62; 2016 23-lab replication d = 0.04; 2021 36-lab test d = 0.06 |
查看完整验算程序
# How a near-zero effect can look large in the published record.
# Suppose the true ego-depletion effect is d = 0.05. Many small labs run studies with 20 people per group;
# only studies with a positive, significant result (p < .05, one-tailed direction) reach print.
import math, random
from statistics import NormalDist, mean, stdev
random.seed(4)
TRUE_D, N, STUDIES = 0.05, 20, 4000
z_crit = NormalDist().inv_cdf(0.975)
all_d, published = [], []
for _ in range(STUDIES):
a = [random.gauss(TRUE_D, 1) for _ in range(N)]
b = [random.gauss(0, 1) for _ in range(N)]
sp = math.sqrt((stdev(a) ** 2 + stdev(b) ** 2) / 2)
d = (mean(a) - mean(b)) / sp
se = math.sqrt(2 / N + d * d / (4 * N))
all_d.append(d)
if d / se > z_crit:
published.append(d)
print(f"true effect d = {TRUE_D}; {STUDIES} studies with {N} per group")
print(f"mean d of all studies run: {mean(all_d):5.2f}")
print(f"share reaching significance: {100 * len(published) / STUDIES:4.1f} %")
print(f"mean d of the published ones: {mean(published):5.2f}")
print("for comparison: 2010 meta-analysis d = 0.62; 2016 23-lab replication d = 0.04; 2021 36-lab test d = 0.06")唯一数据源与引用
范雅福(Fan, Y.). (2026). 《行动的经典与统计力学》第 6 章「认知能量预算与自我损耗之争」(版本 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23738774
DOI: 10.5281/zenodo.23738774 · 阅读本章所在专著 · 5DL-B-013 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026013,
author = {Fan, Yafu},
title = {Classical and Statistical Mechanics of Human Action},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
doi = {10.5281/zenodo.23738774},
url = {https://doi.org/10.5281/zenodo.23738774}
}