M1-058M1: Mathematics & Statistics FundamentalsCausal InferenceMedium
⚡Flashcard Drill➔Mastery:

Explain Propensity Score Matching (PSM) and Inverse Probability Weighting (IPW) for Observational Causal Inference

📐 Mathematical Definition
IPW: τ^=1n∑[TiYie(Xi)−(1−Ti)Yi1−e(Xi)]\mathrm{IPW}:\ \hat\tau=\frac1n\sum\Big[\frac{T_iY_i}{e(X_i)}-\frac{(1-T_i)Y_i}{1-e(X_i)}\Big]
⚡ One-Sentence Core Answer
The propensity score e(X)=P(T=1∣X)e(X) = P(T=1\mid X) condenses high-dimensional confounders into a scalar; PSM matches treated and control units with identical scores, while IPW weights units by 1/e(X)1/e(X) to create a synthetic pseudo-randomized population.

📌 Key Takeaways

  • •
    Propensity Score: By Rosenbaum & Rubin (1983), conditioning on scalar e(X)e(X) achieves conditional independence: (Y(0),Y(1))⊥T∣e(X)(Y(0), Y(1)) \perp T \mid e(X).
  • •
    Propensity Score Matching (PSM): Pairs each treated unit with one or more nearest-neighbor control units having similar e(X)e(X).
  • •
    Inverse Probability Weighting (IPW): Weights treated units by 1e(X)\frac{1}{e(X)} and control units by 11−e(X)\frac{1}{1 - e(X)}; Horvitz-Thompson estimator.
  • •
    Positivity Assumption: 0<e(X)<10 < e(X) < 1 for all XX; every unit must have non-zero probability of receiving either treatment.

📐 Mathematical Derivations

Proof of balancing score property: P(T=1∣X,e(X))=P(T=1∣X)=e(X)P(T=1 \mid X, e(X)) = P(T=1\mid X) = e(X). Thus, TT and XX are conditionally independent given e(X)e(X). For IPW Horvitz-Thompson estimator: μ^1=E[TYe(X)]=EX[E[TY(1)e(X)∣X]]=EX[E[T∣X]E[Y(1)∣X]e(X)]=EX[e(X)E[Y(1)∣X]e(X)]=EX[E[Y(1)∣X]]=E[Y(1)]\hat{\mu}_1 = E\left[\frac{T Y}{e(X)}\right] = E_X\left[E\left[\frac{T Y(1)}{e(X)} \mid X\right]\right] = E_X\left[\frac{E[T\mid X] E[Y(1)\mid X]}{e(X)}\right] = E_X\left[\frac{e(X) E[Y(1)\mid X]}{e(X)}\right] = E_X[E[Y(1)\mid X]] = E[Y(1)]. Similarly, E[(1−T)Y1−e(X)]=E[Y(0)]E\left[\frac{(1-T)Y}{1-e(X)}\right] = E[Y(0)]. The IPW ATE estimator is τ^IPW=1N∑i=1N(TiYie(Xi)−(1−Ti)Yi1−e(Xi))\hat{\tau}_{\text{IPW}} = \frac{1}{N}\sum_{i=1}^N \left( \frac{T_i Y_i}{e(X_i)} - \frac{(1-T_i)Y_i}{1-e(X_i)} \right).

🏭 Production Trade-offs

Comparison: (1) PSM drawbacks: Discards unmatched control units (reducing effective sample size), suffers from caliper tuning sensitivity, and cannot easily handle continuous treatments. (2) IPW drawbacks: Extreme propensity scores (e(X)<0.01e(X) < 0.01 or >0.99> 0.99) blow up weights, causing massive estimator variance. In practice, scientists use Doubly Robust Estimators (AIPW), which combine propensity scores with outcome regression models; if EITHER model is correctly specified, the estimator remains consistent.
⚠️ Common Interview Pitfalls
  • ✕
    Violating positivity (common support): Trying to match treated units in regions of feature space where zero control units ever exist.
  • ✕
    Unnormalized IPW weights creating massive variance spikes when propensity scores approach zero or one (mitigated by weight trimming / clipping).
🎯 Interviewer Follow-up Inquiries
  • ?
    How does the Augmented Inverse Probability Weighting (AIPW) estimator achieve double robustness?
  • ?
    Why is covariate balance checking (standardized mean difference < 0.1) mandatory after propensity score matching?
⚡
Test muscle memory on this topic via 3-Layer Flashcards
Active Recall: Mentally formulate Prompt ➔ Math Derivation ➔ Benchmark Implementation in 3 minutes.
🎯 Drill This Card Now➔
📚

Associated Knowledge Base Guides & Mindmaps

Explore the comprehensive technical article, exam cards, and global architecture tree.

← PreviousM1-057: Explain Confounders, Mediators, and Colliders in Causal DAGs and Which Must or Must Not Be Controlled📋Back to BankNext →M1-059: Explain Difference-in-Differences (DID) and Its Foundational Parallel Trends Assumption
中文速览与面试要点

解释倾向得分匹配(PSM)与逆概率加权(IPW)。

PSM 用倾向得分匹配相似样本;IPW 用 1/倾向得分重加权,构造伪随机化总体。

核心要点

  • 倾向得分 e(X)=P(T=1|X) 是'在给定协变量下接受处理的概率',它把高维协变量压缩为一维——这是 Rosenbaum & Rubin (1983) 的核心定理:若控制 X 后处理可忽略,则控制 e(X) 同样可忽略。 PSM 通过匹配倾向得分相近的处理/对照个体来构造可比组,直观但丢弃未匹配样本且匹配质量难控。 IPW 的思路完全不同:给每个观测赋予权重 1/e(X)(处理组)或 1/(1−e(X))(对照组),使得加权后的样本在协变量分布上 模拟随机化 ——直觉是'倾向得分低却接受了处理的人代表了很多没被处理的人,故应放大其权重'。IPW 估计量可写为加权均值差,且有两组正确性条件:倾向得分模型正确,或结果模型正确(后者对应 DR)。
  • 实践要点:① 假设与诊断 ——需满足 条件可忽略性 (无未观测混淆,不可检验)、 重叠性/正性 (0 SUTVA 。诊断手段:检查倾向得分的重叠区域(画两组的 e(X) 分布)、用标准化偏差(standardized bias)检查加权/匹配后的协变量平衡、检查有效样本量(ESS)。② IPW 的方差问题 ——权重可能极大(当 e(X)→0 或 1 时),导致方差爆炸;缓解手段是 稳定化权重 (乘以边际处理概率,SNIPS)、 权重截断 (trimming,如限制到 99 分位)、或用 overlap weights (一种自带降方差性质且不需截断的加权方案)。③ PSM 的常见误用 ——只匹配倾向得分而不检查协变量平衡、用匹配后样本直接做 t 检验(未考虑匹配的不确定性)、丢弃大量样本导致外部效度下降(估的是 ATT 而非 ATE)。④ PSM vs IPW 选择 ——IPW 用全部样本、理论基础更清晰(可推导渐近方差)、更易扩展(DR、TMLE),现代实践中更受推荐。
  • 倾向得分 e(X)=P(T=1|X)
  • 重叠性假设:0<e(X)<1
  • IPW 对倾向得分模型误设敏感

常见陷阱

  • 只匹配倾向得分但不检查协变量平衡
  • 忽视重叠性假设(e(X) 接近 0 或 1 时权重爆炸)