关于「逆火之辨:纠正平均有效,逆火罕见」,《先验消解与反逆火沟通》给出的定量结论是什么?
结论:第一章说明,没有任何人真逆火时,重测信度 0.50 的量表也会让 0.351 的人看上去在逆火,信度 0.95 时仍有 0.179;五十二个议题里光凭偶然平均就有 2.60 个显著逆火;混合证据下,不合意材料权重低于 0.4286,两派就会拉开,不需要任何逆火。
数据口径:书中参数为示意,全部计算由书中程序生成;本书非心理咨询、不提供任何临床服务、非个人化经营建议,例子只用商品与消费认知,不涉及政治、宗教与族群议题。 第六章把追问、肯定、证据与思想实验写成马尔可夫决策过程,示意参数下最难起点的期望收益 10.555,参数扰动两千次,完整路线不变的比例为 0.8270,沿用基准策略的平均相对遗憾只有 0.0003,所以只称模拟最优。
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: 0.179, 0.351, 0.50, 0.95
| true change N(-8.0, 6.0) for everyone; nobody truly backfires | |
| r SEM sd(obs) P(rise) sim RCI P(rel) sim | |
| 0.50 | 14.14 20.88 0.351 0.347 39.20 0.0119 0.0088 |
| 0.60 | 12.65 18.87 0.336 0.338 35.06 0.0112 0.0138 |
| 0.70 | 10.95 16.61 0.315 0.315 30.36 0.0105 0.0100 |
| 0.80 | 8.94 14.00 0.284 0.287 24.79 0.0096 0.0110 |
| 0.90 | 6.32 10.77 0.229 0.232 17.53 0.0089 0.0088 |
| 0.95 | 4.47 8.72 0.179 0.175 12.40 0.0097 0.0105 |
| apparent backfire share falls from 0.351 to 0.179 | |
| ratio of the two shares | 1.96 |
| reported item-level rho | -0.61 and -0.73 |
查看完整验算程序
# Chapter 1. p11: measurement noise alone manufactures "backfire" cases.
# Model (illustrative): every respondent truly moves toward the correction.
# True change D ~ N(MU, TAU) on a 0-100 confidence scale, MU < 0 means the
# false belief weakens. Observed pre and post scores add independent error
# with SD SEM = SD_X * sqrt(1 - r), r = test-retest reliability.
# Apparent backfire: observed change > 0. Reliable backfire: observed
# change above the reliable change index 1.96 * sqrt(2) * SEM.
# All parameters are illustrative, not estimates from any study.
import math
import random
from statistics import NormalDist
MU, TAU, SD_X = -8.0, 6.0, 20.0
N_SIM = 4000
Z = NormalDist()
def rates(r):
sem = SD_X * math.sqrt(1.0 - r)
sd_obs = math.sqrt(TAU ** 2 + 2.0 * sem ** 2)
p_any = 1.0 - Z.cdf((0.0 - MU) / sd_obs)
rci = 1.96 * math.sqrt(2.0) * sem
p_rel = 1.0 - Z.cdf((rci - MU) / sd_obs)
return sem, sd_obs, p_any, p_rel, rci
def simulate(r, rng):
sem = SD_X * math.sqrt(1.0 - r)
rci = 1.96 * math.sqrt(2.0) * sem
n_any = n_rel = 0
for _ in range(N_SIM):
d = rng.gauss(MU, TAU)
obs = d + rng.gauss(0.0, sem) - rng.gauss(0.0, sem)
if obs > 0.0:
n_any += 1
if obs > rci:
n_rel += 1
return n_any / N_SIM, n_rel / N_SIM
if __name__ == "__main__":
rng = random.Random(38711)
print("true change N({}, {}) for everyone; nobody truly backfires".format(
MU, TAU))
print(" r SEM sd(obs) P(rise) sim RCI P(rel) sim")
for r in (0.50, 0.60, 0.70, 0.80, 0.90, 0.95):
sem, sdo, pa, pr, rci = rates(r)
sa, sr = simulate(r, rng)
fmt = " {:.2f} {:6.2f} {:7.2f} {:6.3f} {:6.3f} {:6.2f} {:7.4f}"
print((fmt + " {:6.4f}")
.format(r, sem, sdo, pa, sa, rci, pr, sr))
lo, hi = rates(0.50)[2], rates(0.95)[2]
print("apparent backfire share falls from {:.3f} to {:.3f}".format(
lo, hi))
print("ratio of the two shares: {:.2f}".format(lo / hi))
# Reported in Swire-Thompson et al. (2022), two samples of 388 and
# 532 people, 21 items: item backfire rate vs test-retest reliability.
print("reported item-level rho: {:.2f} and {:.2f}".format(-0.61, -0.73))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《顽固先验偏见消解、解释深度错觉破除与反逆火效应沟通母典》第 1 章「逆火之辨:纠正平均有效,逆火罕见」(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-387 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026387,
author = {Fan, Yafu},
title = {Dissolving Entrenched Priors, Shattering the Illusion of Explanatory Depth and Anti-Backfire Persuasion Mechanics},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}