关于「出生年份不是界面变量:世代标签的方差与可测能力」,《客群分化界面与同价同条款》给出的定量结论是什么?

结论:出生世代是一个很弱的标签。元分析显示世代之间在工作态度上的均值差很小,一个标签哪怕对应 0.5 个标准差的差异,也只能解释不到百分之六的方差。年龄相关的能力变化真实存在,但它是连续的,个体差异很大,访客自己的动作比出生年份更接近我们要测的量。按年龄分流会把大量访客送错地方,一键自选则不必猜。

数据口径:书中参数为示意,全部计算由书中程序生成,不对应任何真实站点;非法律建议,非个人化经营建议。 第五章说明选择过载是有条件的效应:示意合并均值接近零,新情境的预测区间却从 -0.5195 到 +0.6687;筛选加买家自选排序让比较楼盘快 5.69 倍。

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

reported: Costanza et al. 2012 pooled 20 studies, 19,961 people
reported: Czaja et al. 2006 CREATE sample 1,204, aged 18-91
A. variance explained by a two-group label with mean gap d
d eta^2 share of variance NOT explained
0.050.0006 0.9994
0.100.0025 0.9975
0.200.0099 0.9901
0.500.0588 0.9412
0.800.1379 0.8621
B. simulated visitors: n20,000
band n mean L
18-293,612 0.139
30-444,476 0.484
45-594,459 0.871
60-857,453 1.361
R^2 of the four-band cohort label0.1757
R^2 of the observed signal S0.7746
ratio signal / label4.41
查看完整验算程序
# Chapter 1. p11: how much variance a birth-cohort label can carry.
# Part A is analytic: two equal groups whose means differ by d standard
# deviations; the label explains eta^2 = d^2 / (d^2 + 4) of the variance.
# Part B is a seeded simulation (illustrative parameters, no site data):
# a latent "needs a larger, calmer layout" score L rises slowly with age,
#   L = 0.025 * (age - 18) + e,  e ~ N(0, 1),
# and an observed interaction composite S (zoom, backtracks, dwell)
# measures L with noise, S = L + N(0, 0.6). We compare the share of the
# variance of L explained by a four-band cohort label with the share
# explained by S.
import math
import random
import statistics

print("reported: Costanza et al. 2012 pooled 20 studies, {:,} people"
      .format(19961))
print("reported: Czaja et al. 2006 CREATE sample {:,}, aged 18-91"
      .format(1204))
print("A. variance explained by a two-group label with mean gap d")
print("     d    eta^2   share of variance NOT explained")
for d in (0.05, 0.1, 0.2, 0.5, 0.8):
    eta2 = d * d / (d * d + 4.0)
    print("  {:>4.2f}   {:.4f}   {:.4f}".format(d, eta2, 1.0 - eta2))

rng = random.Random(20261120)
N = 20000
ages, lat, sig = [], [], []
for _ in range(N):
    age = rng.uniform(18.0, 85.0)
    L = 0.025 * (age - 18.0) + rng.gauss(0.0, 1.0)
    ages.append(age)
    lat.append(L)
    sig.append(L + rng.gauss(0.0, 0.6))

bands = [(18, 30), (30, 45), (45, 60), (60, 86)]
mu = statistics.fmean(lat)
tot = math.fsum((x - mu) ** 2 for x in lat)
between = []
print("B. simulated visitors: n = {:,}".format(N))
print("   band      n     mean L")
for lo, hi in bands:
    grp = [x for a, x in zip(ages, lat) if lo <= a < hi]
    m = statistics.fmean(grp)
    between.append(len(grp) * (m - mu) ** 2)
    print("   {}-{}   {:>6,}  {:>6.3f}".format(lo, hi - 1, len(grp), m))
r2_label = math.fsum(between) / tot
r = statistics.correlation(lat, sig)
print("   R^2 of the four-band cohort label: {:.4f}".format(r2_label))
print("   R^2 of the observed signal S:      {:.4f}".format(r * r))
print("   ratio signal / label:              {:.2f}".format(
    r * r / r2_label))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《客群认知画像分层、适老与新世代界面分化及千人千面转化交互母典》第 1 章「出生年份不是界面变量:世代标签的方差与可测能力」(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-363 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026363,
  author    = {Fan, Yafu},
  title     = {Demographic Cognitive Segmentation, Elder-to-Gen-Z Interface Divergence and Audience-Adaptive Conversion UX},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论