Overview
James J. Heckman is the Henry Schultz Distinguished Service Professor of Economics at the University of Chicago and a Nobel laureate (2000, shared with Daniel McFadden) for the development of econometric methods for analyzing selective samples. His contributions span econometric theory (the Heckman selection correction), causal inference (Marginal Treatment Effects, treatment effect heterogeneity under essential heterogeneity), and the economics of human capital (returns to education under uncertainty, technology of skill formation, early childhood investment). He is one of the most-cited economists in history.
Key Contributions
- Heckman (1979) — "Sample Selection Bias as a Specification Error." Econometrica 47(1): 153–161. Introduced the Heckman two-step estimator (lambda correction) for correcting OLS bias when the sample is non-randomly selected; foundational for labor economics, DI application studies, and program evaluation.
- Marginal Treatment Effects (MTE) — With Edward Vytlacil: showed that IV estimates can be decomposed as weighted averages of individual-level treatment effects, with weights determined by the instrument's propensity score. Clarified that LATE (Imbens-Angrist) is a special case of MTE, and that different instruments yield different LATEs because they move different people.
- Heckman (1996) JASA Comment — Four-point critique of the AIR (1996) framework: (1) potential outcomes = econometric switching regression, independently invented; (2) mean independence assumptions (A-1, A-2, A-3) suffice to identify ATT without AIR's full independence or monotonicity; (3) AIR's assumptions impose behavioral restrictions (Granger noncausality) violated under essential heterogeneity and in Roy/competing-risks/labor-supply models; (4) draft lottery instrument is questionable on exclusion restriction and relevance grounds. Argues ATT (for an observable subpopulation) is preferable to LATE (for an unobservable one).
- Heckman (2008) "Econometric Causality" — Survey for International Statistical Review contrasting the econometric and Neyman-Rubin statistical approaches to causal inference. Introduces the three-task distinction (defining counterfactuals / identifying from ideal data / identifying from real data); the P1/P2/P3 policy hierarchy (historical evaluation / new environments / new policies); the Marginal Treatment Effect (MTE) synthesis showing OLS, IV, matching, and DiD are different weighted averages of individual treatment effects; the fixing-vs-conditioning distinction (Haavelmo 1943); and Marschak's Maxim (use the minimum model needed for the policy question). Argues structural models are necessary for P2/P3 questions and that LATE identifies only a policy-irrelevant, instrument-specific complier average.
- Cunha and Heckman (2007) — Distinguished ex ante from ex post returns to schooling; showed that selection on private information about returns ("essential heterogeneity") explains why IV estimates can exceed OLS estimates without violating instrument validity; identified distributions of both ex ante and ex post returns using a factor model.
- Technology of Skill Formation — With Flavio Cunha and others: formalized a multi-period skill investment model showing that early childhood is the highest-return period for human capital investment; cognitive and non-cognitive skills are complements and self-reinforce across stages. Empirical basis for evaluations of Perry Preschool and Abecedarian programs showing large long-run returns to early childhood intervention.
Related
Sources