Empirical Strategies in Economics: Illuminating the Path from Cause to Effect
Causal Question / Estimand
The estimand is the Local Average Treatment Effect (LATE) — the average causal effect for compliers, the subpopulation whose treatment status is moved by the instrument. The lecture’s point is that IV does not recover a universal ATE; it recovers an effect for a specific, instrument-defined group, and naming that group is what makes IV credible.
Identification Strategy
Instrumental variables under the LATE framework of Imbens–Angrist. An instrument that is (i) as-good-as-randomly assigned, (ii) affects outcomes only through treatment (Exclusion-Restriction), and (iii) moves everyone weakly the same direction (Monotonicity) identifies the complier average causal effect. Worked through with charter- and exam-school lottery and admission-cutoff examples; illustrates how a surprising exclusion restriction explains why Chicago exam-school enrollment can lower achievement.
Key Assumptions
Randomization (instrument independence / as-good-as-random assignment), Exclusion-Restriction, Monotonicity (no defiers), instrument relevance (first stage), and SUTVA. The estimand is LATE.
Threats to Validity
Violations of the exclusion restriction (the instrument affecting outcomes through other channels) and of monotonicity; weak first stages; and the external-validity limits of a complier-specific effect. Angrist reframes the exclusion restriction as a commitment device forcing a clear, consistent causal story for the reduced form.
Setting / Data
Education examples from Blueprint Labs: Boston/New York charter-school admission lotteries and Chicago/Boston exam-school admission cutoffs (a fuzzy-RD/IV design), plus the Tennessee STAR class-size RCT as a benchmark.
Key Claims
- IV’s value is interpretive: LATE reveals for whom an estimate is valid, separating independence (cheap, from random assignment) from the exclusion restriction (the controversial, substantive assumption).
- Exclusion restrictions formalize a commitment to clear explanations of reduced-form effects.
- The credibility revolution owes “at least as much to compelling empirical analyses as to methodological insights.”
Connections
- Builds on: AngristKrueger1999-EmpiricalStrategiesLaborEconomics, Potential-Outcomes
- See also: AngristPischke2010-CredibilityRevolution, IV, Foundations
- Shares the as-good-as-random logic with: Meyer1995-NaturalAndQuasiExperiments
Citation
Angrist, J. D. (2022). Empirical Strategies in Economics: Illuminating the Path from Cause to Effect. Econometrica, 90(6), 2509–2539. https://doi.org/10.3982/ECTA20640