关于「二十一天从哪里来:Maltz 的「至少」与 Lally 的六十六天」,《习惯渐近曲线与二十一天神话》给出的定量结论是什么?
结论:第一章追溯 21 天的来历:它源自 Maltz 对整形病人心理意象的临床印象,原话是「至少」;Lally 等人 2010 年的追踪研究给出的中位数是 66 天,个体从 18 天到 254 天。按其摘要重建的对数正态分布中,21 天内到达的人只占 0.027;把离散度放到综述允许的最大值,也只有 0.228。
数据口径:书中的人群参数、遗忘速率与线索指数均为示意;全部计算由书中程序生成;本书不是医疗或心理建议,也不是个人化的经营建议。 书中以标注框区分已发表的证据、可运行的 Python 计算(19 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: 0.027, 0.228, 18, 21, 66
| source | sigma | P(T95<=21) | P(T95<=21) |
| median 66 | median 59 | ||
| range 18-254, n=39 | 0.593 | 0.027 | 0.041 |
| drinking IQR 39-75 | 0.485 | 0.009 | 0.017 |
| exercise IQR 44-118 | 0.731 | 0.059 | 0.079 |
| eating IQR 35-106 | 0.821 | 0.082 | 0.104 |
| review 106 / 66 | 0.973 | 0.120 | 0.144 |
| review 154 / 59 | 1.385 | 0.204 | 0.228 |
| largest 21-day share in the scan: 0.228 | |||
| even then, odds against 21 days: 3.4 to 1 |
查看完整验算程序
# Chapter 1. p12: Sensitivity scan. How large could the 21-day share be?
# The log-sd sigma of T95 is not published, so we infer it four ways:
# (a) range 18 to 254 of 39 people (p11); (b) the quartiles of the eating,
# drinking and exercise subgroups in Lally et al. (2010): 35-106, 39-75,
# 44-118 days; (c) the review of Singh et al. (2024), whose medians run
# 59 to 66 days and means 106 to 154 days, read as a lognormal with
# mean / median = exp(sigma^2 / 2). Medians 59 and 66 are both scanned.
# Pairing (c) across studies is the author's assumption (illustrative).
import math
from statistics import NormalDist
Z = NormalDist()
zq = Z.inv_cdf(0.75)
zx = Z.inv_cdf((39 - 0.5) / 39)
cases = [
("range 18-254, n=39", math.log(254 / 18) / (2 * zx)),
("drinking IQR 39-75", math.log(75 / 39) / (2 * zq)),
("exercise IQR 44-118", math.log(118 / 44) / (2 * zq)),
("eating IQR 35-106", math.log(106 / 35) / (2 * zq)),
("review 106 / 66", math.sqrt(2 * math.log(106 / 66))),
("review 154 / 59", math.sqrt(2 * math.log(154 / 59))),
]
print(" source sigma P(T95<=21) P(T95<=21)")
print(" median 66 median 59")
worst = 0.0
for name, s in cases:
p66 = Z.cdf(math.log(21 / 66) / s)
p59 = Z.cdf(math.log(21 / 59) / s)
worst = max(worst, p59)
print(" %-21s %6.3f %9.3f %15.3f" % (name, s, p66, p59))
print("")
print("largest 21-day share in the scan: %.3f" % worst)
print("even then, odds against 21 days: %.1f to 1"
% ((1 - worst) / worst))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《习惯形成渐近曲线、二十一天神话证伪与情境线索自动化指数母典》第 1 章「二十一天从哪里来:Maltz 的「至少」与 Lally 的六十六天」(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-197 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026197,
author = {Fan, Yafu},
title = {Asymptotic Habit-Formation Curves, Debunking the 21-Day Myth and a Cue-Based Automaticity Index},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}