M1-060M1: Mathematics & Statistics FundamentalsCausal InferenceHard
⚡Flashcard Drill➔Mastery:

Explain Instrumental Variables (IV), Two-Stage Least Squares (2SLS), and the Local Average Treatment Effect (LATE)

🎯 Active Recall Self-Check (Draft First, Verify After)
Login-free practice tracker: recall core answer, audit pitfalls, and log review status.
Step 1 · Draft your concise answer & assumptions
Step 2 · Check Pitfalls & Follow-up Inquiries
Next: M1-061 →
📐 Mathematical Definition
Y=βT+ε,T=πZ+ν,β^2SLS=Cov(Z,Y)Cov(Z,T)Y=\beta T+\varepsilon,\quad T=\pi Z+\nu,\qquad \hat\beta_{2SLS}=\frac{\mathrm{Cov}(Z,Y)}{\mathrm{Cov}(Z,T)}
⚡ One-Sentence Core Answer
An Instrumental Variable ZZ influences treatment XX without directly affecting outcome YY except through XX; 2SLS purges endogeneity, estimating the Local Average Treatment Effect (LATE) specifically for 'compliers'.

📌 Key Takeaways

  • •
    Core IV Assumptions: (1) Relevance (Cov(Z,X)≠0\text{Cov}(Z, X) \ne 0), (2) Exclusion Restriction (Z→YZ \to Y only through XX), and (3) Independence (Z⊥unobserved confoundersZ \perp \text{unobserved confounders}).
  • •
    Two-Stage Least Squares (2SLS): Stage 1 regresses XX on ZZ to get exogenous fitted values X^\hat{X}; Stage 2 regresses YY on X^\hat{X}.
  • •
    LATE (Angrist & Imbens): The estimated effect applies strictly to 'compliers' (units whose treatment adoption was caused by instrument ZZ).

📐 Mathematical Derivations

Consider structural model Y=βX+UY = \beta X + U where Cov(X,U)≠0\text{Cov}(X, U) \ne 0 (omitted variable bias / endogeneity). OLS is biased: β^OLS=β+Cov(X,U)Var(X)≠β\hat{\beta}_{\text{OLS}} = \beta + \frac{\text{Cov}(X, U)}{\text{Var}(X)} \ne \beta. Let ZZ be a valid instrument such that Cov(Z,U)=0\text{Cov}(Z, U) = 0 and Cov(Z,X)≠0\text{Cov}(Z, X) \ne 0. Taking covariance with ZZ on both sides: Cov(Z,Y)=Cov(Z,βX+U)=βCov(Z,X)+Cov(Z,U)=βCov(Z,X)\text{Cov}(Z, Y) = \text{Cov}(Z, \beta X + U) = \beta \text{Cov}(Z, X) + \text{Cov}(Z, U) = \beta \text{Cov}(Z, X). Dividing gives the Wald estimator: β^IV=Cov(Z,Y)Cov(Z,X)=Reduced Form SlopeFirst Stage Slope\hat{\beta}_{\text{IV}} = \frac{\text{Cov}(Z, Y)}{\text{Cov}(Z, X)} = \frac{\text{Reduced Form Slope}}{\text{First Stage Slope}}. In 2SLS with multiple instruments: Stage 1 fits X^=Z(ZTZ)−1ZTX=PZX\hat{X} = Z(Z^T Z)^{-1} Z^T X = P_Z X. Stage 2 fits Y=X^β+ϵY = \hat{X}\beta + \epsilon, yielding β^2SLS=(XTPZX)−1XTPZY\hat{\beta}_{\text{2SLS}} = (X^T P_Z X)^{-1} X^T P_Z Y.

🏭 Production Trade-offs

In tech industry experiments, IV is essential for non-compliance (encouragement designs): An experiment randomizes discount coupons (Z∈{0,1}Z \in \{0, 1\}), but only some users redeem it and make a purchase (X∈{0,1}X \in \{0, 1\}). Intention-to-treat (ITT) measures impact of the coupon; IV/2SLS estimates the causal impact of the purchase itself among compliers (LATE). If the first stage is weak (F<10F < 10), IV standard errors explode and 2SLS becomes heavily biased toward OLS.
⚠️ Common Interview Pitfalls
  • ✕
    Weak instruments (FF-statistic <10< 10), which magnifies bias and ruins confidence interval coverage.
  • ✕
    Violating the exclusion restriction: If instrument ZZ directly impacts YY through any channel other than XX, IV estimates are invalid.
🎯 Interviewer Follow-up Inquiries
  • ?
    Why does a weak instrument (F<10F < 10) cause 2SLS estimates to be biased toward OLS?
  • ?
    Who are the four compliance cohorts (Compliers, Always-Takers, Never-Takers, Defiers) and why does Monotonicity rule out Defiers?
⚡
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

← PreviousM1-059: Explain Difference-in-Differences (DID) and Its Foundational Parallel Trends Assumption📋 Back to BankNext →M1-061: Explain Covariance vs. Correlation Coefficient and Whether Zero Correlation Implies Independence