Instrumental Variables: An Econometrician’s Perspective
Causal Question / Estimand
What IV identifies under treatment-effect heterogeneity: the LATE — the average effect for compliers — rather than an overall ATE; presented to bridge the econometric and statistical literatures.
Identification Strategy
Reviews and connects perspectives on IV using the Potential-Outcomes framework. Lays out the modern (Imbens–Angrist) result that, with heterogeneous effects, IV identifies a complier-specific LATE under instrument independence, the Exclusion-Restriction, and Monotonicity. Contrasts this “local” view with the traditional constant-effects/structural reading where IV recovers a single structural parameter, and discusses settings (randomized experiments with noncompliance, natural experiments) where each view applies. Emphasizes the credibility hierarchy of assumptions and the weak-instrument problem.
Key Assumptions
Instrument independence, Exclusion-Restriction, Monotonicity (for LATE), relevance / strong first stage (Weak-Instruments), SUTVA.
Threats to Validity
Failure of exclusion or monotonicity; weak instruments; conflating LATE with ATE when effects are heterogeneous (external-validity limits).
Setting / Data
n/a — methodological review (Statistical Science).
Key Claims
- Under heterogeneity, IV estimates a LATE for compliers, not a universal average effect — the central reinterpretation of IV.
- The potential-outcomes framework reconciles the econometric and statistical accounts of instrumental variables.
- Monotonicity and exclusion are the substantive, largely untestable commitments; relevance is testable but weak instruments still threaten inference.
Connections
- Formalizes the LATE view summarized by Angrist2022-EmpiricalStrategiesIlluminatingPath and AngristKrueger2001-SearchForIdentification.
- Shares its potential-outcomes basis with Rubin1974-EstimatingCausalEffects and the program-evaluation survey ImbensWooldridge2009-ProgramEvaluation.
- Weak-instrument practice: Murray2006-WeakAndInvalidInstruments. See also IV
Citation
Imbens, G. W. (2014). Instrumental Variables: An Econometrician’s Perspective. Statistical Science, 29(3), 323–358.