关于「漏斗里的空气:选择性发表、Egger 检验与 PET-PEESE」,《自我损耗与饥饿法官》给出的定量结论是什么?

结论:第二章模拟一个真实为零、经选择性发表筛选的文献,随机效应元分析仍得到 0.286,并用 Egger 检验、PET 与 PEESE 检测与校正。

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。

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: 0.286

all run k 1000, significant25 (0.025)
mean d -0.017, fixed -0.012, random -0.013 [-0.031, 0.006]
tau^2 0.0040, I^2 0.047
published k72, significant 25 (0.347)
mean d 0.308, fixed 0.276, random 0.286 [0.207, 0.365]
tau^2 0.0357, I^2 0.315
share of run studies that were published0.072
published mean d: n<20 0.352 (k 22), n>30 0.191 (k 25)
查看完整验算程序
# Chapter 2. p21: A literature built on a true effect of zero.
# 1000 two-group studies, n per group drawn from 10 to 40, true d = 0.
# Every significant positive study is published; a non-significant one
# with probability 0.10 if its sign is positive, 0.02 if negative.
# The published set is pooled by DerSimonian-Laird random effects.
# All parameters are illustrative. Seed fixed.
# 真实效应为零的文献,经选择性发表后被元分析。
import math
import random
from statistics import NormalDist

Z975 = NormalDist().inv_cdf(0.975)
rng = random.Random(20261112)
TRUE_D, K_RUN = 0.0, 1000
P_NS_POS, P_NS_NEG = 0.10, 0.02


def one_study(n, true_d):
    a = [rng.gauss(true_d, 1.0) for _ in range(n)]
    b = [rng.gauss(0.0, 1.0) for _ in range(n)]
    ma, mb = math.fsum(a) / n, math.fsum(b) / n
    va = math.fsum((x - ma) ** 2 for x in a) / (n - 1)
    vb = math.fsum((x - mb) ** 2 for x in b) / (n - 1)
    d = (ma - mb) / math.sqrt((va + vb) / 2)
    se = math.sqrt(2 / n + d * d / (4 * n))
    return d, se


def dl(ds, ses):
    w = [1 / s ** 2 for s in ses]
    sw = math.fsum(w)
    fe = math.fsum(wi * di for wi, di in zip(w, ds)) / sw
    q = math.fsum(wi * (di - fe) ** 2 for wi, di in zip(w, ds))
    k = len(ds)
    c = sw - math.fsum(wi * wi for wi in w) / sw
    tau2 = max(0.0, (q - (k - 1)) / c)
    wr = [1 / (s ** 2 + tau2) for s in ses]
    re = math.fsum(wi * di for wi, di in zip(wr, ds)) / math.fsum(wr)
    se_re = 1 / math.sqrt(math.fsum(wr))
    i2 = max(0.0, (q - (k - 1)) / q) if q > 0 else 0.0
    return fe, re, se_re, tau2, i2


pub, allst = [], []
for _ in range(K_RUN):
    n = rng.randint(10, 40)
    d, se = one_study(n, TRUE_D)
    allst.append((d, se, n))
    sig = d / se > Z975
    if sig or rng.random() < (P_NS_POS if d > 0 else P_NS_NEG):
        pub.append((d, se, n))
for label, st in (("all run", allst), ("published", pub)):
    ds = [s[0] for s in st]
    ses = [s[1] for s in st]
    fe, re, se_re, tau2, i2 = dl(ds, ses)
    nsig = len([1 for d, s in zip(ds, ses) if d / s > Z975])
    print("{:9s} k {:3d}, significant {:3d} ({:.3f})".format(
        label, len(st), nsig, nsig / len(st)))
    print("   mean d {:.3f}, fixed {:.3f}, random {:.3f} [{:.3f}, {:.3f}]"
          .format(math.fsum(ds) / len(ds), fe, re,
                  re - Z975 * se_re, re + Z975 * se_re))
    print("   tau^2 {:.4f}, I^2 {:.3f}".format(tau2, i2))
print("share of run studies that were published: {:.3f}".format(
    len(pub) / K_RUN))
small = [s for s in pub if s[2] <= 20]
large = [s for s in pub if s[2] > 30]
print("published mean d: n<=20 {:.3f} (k {}), n>30 {:.3f} (k {})".format(
    math.fsum(s[0] for s in small) / len(small), len(small),
    math.fsum(s[0] for s in large) / len(large), len(large)))

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 2 章「漏斗里的空气:选择性发表、Egger 检验与 PET-PEESE」(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}
}

本书其他章节结论