关于「后门集合的枚举:d 分离、M 型偏差与因果图的计数」,《辛普森悖论与后门调整》给出的定量结论是什么?

结论:第五章枚举后门集合,并说明对一个处理前的对撞变量作调整会带来相当于效应 66.0 % 的偏差。

数据口径:书中门店、方案与成交数均为示意;全部计算由书中程序生成。 第一章在伯克利六个最大院系上复算反转:合并后男性录取率 0.445、女性 0.304,按申请者构成标准化后男性 0.387、女性 0.430;三店示意报表中甲在每店都高 0.050,合并后却落后 0.2493,并给出反转可能的充要条件。

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

descendants of X: C,M,Y
subsets tried 32  admissible 3  rejected 29
  admissible: {R,S}
  admissible: {S,F}
  admissible: {R,S,F}
minimal: {R,S}  {S,F}
  {S}: open back door
  {R,F}: open back door
  {S,R,M}: descendant of X
  {S,R,C}: descendant of X
查看完整验算程序
# Chapter 5. p51: enumerate every back-door admissible set in a graph.
# Back-door criterion (Pearl 2009, 3.3.1): a set S is admissible for
# (X, Y) if no member of S descends from X and S d-separates X from Y
# in the graph with the arrows out of X removed. d-separation is
# tested by the moralised ancestral graph. Illustrative branch graph:
#   R region, S size, F footfall, X plan, M basket, Y sales,
#   C review (a collider of X and Y).
from itertools import combinations

EDGES = [("R", "X"), ("R", "F"), ("F", "Y"), ("S", "X"), ("S", "Y"),
         ("X", "M"), ("M", "Y"), ("X", "Y"), ("X", "C"), ("Y", "C"),
         ("R", "S")]
CANDS = ["R", "S", "F", "M", "C"]


def parents(edges, v):
    return [a for a, b in edges if b == v]


def ancestors(edges, nodes):
    out = set(nodes)
    stack = list(nodes)
    while stack:
        v = stack.pop()
        for p in parents(edges, v):
            if p not in out:
                out.add(p)
                stack.append(p)
    return out


def descendants(edges, v):
    out, stack = set(), [v]
    while stack:
        u = stack.pop()
        for a, b in edges:
            if a == u and b not in out:
                out.add(b)
                stack.append(b)
    return out


def dsep(edges, x, y, given):
    keep = ancestors(edges, [x, y] + list(given))
    sub = [(a, b) for a, b in edges if a in keep and b in keep]
    nb = {v: set() for v in keep}
    for a, b in sub:
        nb[a].add(b)
        nb[b].add(a)
    for v in sorted(keep):
        ps = sorted(parents(sub, v))
        for i, j in combinations(ps, 2):
            nb[i].add(j)
            nb[j].add(i)
    seen, stack = {x}, [x]
    while stack:
        u = stack.pop()
        for w in sorted(nb[u]):
            if w in given or w in seen:
                continue
            if w == y:
                return False
            seen.add(w)
            stack.append(w)
    return True


def main():
    cut = [(a, b) for a, b in EDGES if a != "X"]
    desc = descendants(EDGES, "X")
    print("descendants of X: " + ",".join(sorted(desc)))
    ok, bad = [], 0
    for r in range(len(CANDS) + 1):
        for s in combinations(CANDS, r):
            if set(s) & desc:
                bad = bad + 1
                continue
            if dsep(cut, "X", "Y", s):
                ok.append(s)
            else:
                bad = bad + 1
    print("subsets tried {}  admissible {}  rejected {}".format(
        2 ** len(CANDS), len(ok), bad))
    for s in ok:
        print("  admissible: {" + ",".join(s) + "}")
    mins = [s for s in ok if not any(set(t) < set(s) for t in ok)]
    print("minimal: " + "  ".join("{" + ",".join(s) + "}" for s in mins))
    for s in (("S",), ("R", "F"), ("S", "R", "M"), ("S", "R", "C")):
        why = "descendant of X" if set(s) & desc else (
            "ok" if dsep(cut, "X", "Y", s) else "open back door")
        print("  {" + ",".join(s) + "}: " + why)


if __name__ == "__main__":
    main()

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《辛普森悖论与后门调整准则、多分店数据合并结论反转与因果图去混杂母典》第 5 章「后门集合的枚举:d 分离、M 型偏差与因果图的计数」(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-302 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026302,
  author    = {Fan, Yafu},
  title     = {Simpson's Paradox and the Back-Door Criterion: Reversals When Branch Data Are Pooled and De-Confounding with Causal Graphs},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论