关于「实证闭环:会员等级门槛效果评估、蒸馏小工具与预注册方案」,《会员门槛的断点回归》给出的定量结论是什么?

结论:第八章给出会员等级门槛效果评估的完整流程、蒸馏小工具「门槛断点效应估计器」与预注册检验方案。书中含 23 段可运行的 Python 程序。

数据口径:落笔前本书核对原文并作蓝图修正:1960 年的论文研究的是奖学金荣誉证书而非会员体系,2008 年的指南是方法论文,二者都没有给出任何会员权益的效应量,书中的高估倍数全部来自固定种子的示意数据;命题也必须带边界陈述:只识别门槛处的局部效应,客人能精确凑单时设计失效,估计对带宽敏感,不宜用全局高阶多项式。 书中参数为示意;全部计算由书中程序生成。

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

pilot se at n = 20000: 0.2451
effect mcustomers neededwithin h of the line
0.53772212992
1.094313249
1.541921444
2.51509520
simulated power at n = 9431 for m = 1: 0.78
查看完整验算程序
# Chapter 8. p81: how many customers a threshold study needs.
# The local linear RD standard error shrinks like 1 / sqrt(n). From one
# pilot sample of 20000 customers (seed 20261113, line c = 60, h = 8)
# we take se0 and compute the sample needed to detect a local effect of
# size m with a two-sided 5% test and 80% power:
#   n(m) = 20000 * ((1.959964 + 0.841621) * se0 / m)^2.
# A Monte Carlo check (100 samples, seed 20261181) sets the benefit
# so that the local effect equals m = 1 and measures the power at n(1).
# Illustrative design; a real pilot must use the shop's own data.
import math
import random

C, H, N0 = 60.0, 8.0, 20000
ZA, ZB = 1.959964, 0.841621


def draw(rng, n, base=2.0, c=C):
    rows = []
    for _ in range(n):
        u = rng.gauss(0.0, 1.0)
        x = 50.0 + 12.0 * u + 8.0 * rng.gauss(0.0, 1.0)
        d = 1 if x >= c else 0
        g = 0.15 * (x - c) + 0.0003 * (x - c) ** 3
        y = 20.0 + 6.0 * u + g + (base + u) * d
        rows.append((x, y + rng.gauss(0.0, 3.0), u))
    return rows


def wls(pts, c, h):
    xs, ys, ws = [], [], []
    for x, y in pts:
        t = abs(x - c) / h
        if t < 1.0:
            xs.append(x - c)
            ys.append(y)
            ws.append(1.0 - t)
    s0 = math.fsum(ws)
    s1 = math.fsum(w * x for w, x in zip(ws, xs))
    s2 = math.fsum(w * x * x for w, x in zip(ws, xs))
    t0 = math.fsum(w * y for w, y in zip(ws, ys))
    t1 = math.fsum(w * x * y for w, x, y in zip(ws, xs, ys))
    det = s0 * s2 - s1 * s1
    a = (s2 * t0 - s1 * t1) / det
    b = (s0 * t1 - s1 * t0) / det
    va = math.fsum((w * (s2 - s1 * x) / det * (y - a - b * x)) ** 2
                   for w, x, y in zip(ws, xs, ys))
    return a, va


def rd(rows, c, h):
    al, vl = wls([(r[0], r[1]) for r in rows if r[0] < c], c, h)
    ar, vr = wls([(r[0], r[1]) for r in rows if r[0] >= c], c, h)
    return ar - al, math.sqrt(vl + vr)


if __name__ == "__main__":
    _, se0 = rd(draw(random.Random(20261113), N0), C, H)
    print("pilot se at n = {}: {:.4f}".format(N0, se0))
    print("effect m   customers needed   within h of the line")
    share = 6888 / N0
    need = {}
    for m in (0.5, 1.0, 1.5, 2.5):
        n = int(math.ceil(N0 * ((ZA + ZB) * se0 / m) ** 2))
        need[m] = n
        print("{:8.1f}   {:16d}   {:20d}".format(
            m, n, int(math.ceil(n * share))))
    rng = random.Random(20261181)
    base = 1.0 - 12.0 / 208.0 * (C - 50.0)
    hit = 0
    for _ in range(100):
        est, se = rd(draw(rng, need[1.0], base), C, H)
        if abs(est / se) > ZA:
            hit = hit + 1
    print("simulated power at n = {} for m = 1: {:.2f}".format(
        need[1.0], hit / 100))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《断点回归设计与会员等级门槛、消费满额升级线两侧的局部因果识别母典》第 8 章「实证闭环:会员等级门槛效果评估、蒸馏小工具与预注册方案」(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-303 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026303,
  author    = {Fan, Yafu},
  title     = {Regression-Discontinuity Designs at Membership-Tier Thresholds: Local Causal Identification on Either Side of a Spend-Based Upgrade Line},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论