Papers That Shaped My Thinking

Why it's a landmark: First to randomize families into lower-poverty neighborhoods, setting the causal benchmark that reframed the entire neighborhood-effects literature.

Why it's a landmark: Decomposed house prices into construction cost, land, and a regulatory 'zoning tax,' attributing high coastal prices to supply regulation rather than scarcity.

Why it's a landmark: Used sex-offender arrival and departure as a localized natural experiment to estimate a sharp house-price gradient in crime risk.

Why it's a landmark: Showed vehicle-kilometers traveled rise one-for-one with road capacity, establishing induced demand so new roads do not relieve congestion.

Why it's a landmark: Used the TVA to estimate agglomeration spillovers and evaluate big-push place-based policy, finding lasting manufacturing gains but modest aggregate efficiency effects.

Why it's a landmark: Used movers and sibling age-at-move variation to show neighborhood exposure effects on children accumulate roughly linearly with childhood years spent there.

Why it's a landmark: Quantified aggregate output lost to housing-supply restrictions in high-productivity cities, estimating large national gains from relaxing them.

Why it's a landmark: Showed a randomized bundle of housing-search assistance, not vouchers alone, sharply raised low-income families' moves to opportunity neighborhoods.

Why it's a landmark: Used federal hiring-grant variation to show added police reduce homicides while disproportionately raising low-level arrests of Black civilians.

Why it's a landmark: Estimated housing-supply elasticities at neighborhood scale, showing within-metro variation dominates and that renovations and teardowns, not just new construction, drive responses.

Why it's a landmark: Used the Mariel supply shock as a natural experiment, finding a sudden low-skill influx barely affected native wages or employment.

Why it's a landmark: Used a cross-border difference-in-differences on New Jersey's minimum-wage rise to find no employment loss, challenging competitive labor-demand predictions.

Why it's a landmark: Instrumented schooling with college proximity, estimating returns to education that exceeded OLS and reframed the ability-bias debate.

Why it's a landmark: Used the adoption of blind orchestra auditions as a natural experiment to identify sex bias in hiring evaluations.

Why it's a landmark: Introduced the routine-task framework, showing computers substitute for routine tasks and complement abstract ones, driving job polarization.

Why it's a landmark: Validated teacher value-added against quasi-random turnover and linked it to students' long-run earnings and life outcomes.

Why it's a landmark: Argued the residual gender pay gap stems from nonlinear pay for long, inflexible hours rather than discrimination or human capital alone.

Why it's a landmark: Documented that earnings gaps open sharply at first childbirth, showing cross-country child penalties track gender norms more than policy.

Why it's a landmark: Built real-time survey measures of remote work, documenting its persistent post-pandemic level and its estimated productivity and amenity value.

Why it's a landmark: Built a public real-time private-data tracker of spending, employment, and revenue to measure the pandemic's granular economic impact.

Why it's a landmark: Introduced reference-dependent utility with loss aversion and probability weighting, supplanting expected-utility theory as the descriptive model of risky choice.

Why it's a landmark: Consolidated experimental evidence that ownership raises valuations, documenting the endowment effect and status-quo bias as consequences of loss aversion.

Why it's a landmark: Showed randomly bidding 'zero-intelligence' traders achieve near-efficient outcomes, attributing market efficiency to institutions rather than trader rationality.

Why it's a landmark: Introduced the heterogeneous-agent general-equilibrium model with uninsurable idiosyncratic risk and borrowing constraints, founding the Bewley-Aiyagari framework.

Why it's a landmark: Used a field experiment to show that employers' costly information acquisition generates statistical discrimination through selective attention to applicants.

Why it's a landmark: Ran an early randomized trial showing a generative-AI writing tool raised productivity while compressing quality dispersion across workers.

Why it's a landmark: Formalized when LLM outputs yield valid inference, requiring no training-data leakage for prediction and a validation-sample correction for estimation.

Why it's a landmark: Formalized how low prior odds, bias, and multiple testing make most published positive findings likely false.

Why it's a landmark: Ran the first large-scale coordinated replication effort, finding under half of psychology studies replicated at reduced effect sizes.

Why it's a landmark: Documented a two-humped distribution of test statistics around significance thresholds in economics journals, evidencing p-hacking and selective reporting.

Why it's a landmark: Conducted coordinated high-powered replications of experimental economics studies, finding a majority replicated with somewhat attenuated effects.

Why it's a landmark: Developed a method to estimate the selection function governing publication and reweight literatures to correct for the resulting bias.

Why it's a landmark: Compared identification strategies across thousands of tests, showing IV and DiD exhibit more inflation from p-hacking and publication bias.

Why it's a landmark: Used the Medicare-at-65 age discontinuity to identify sharp jumps in insurance coverage and healthcare utilization.

Why it's a landmark: Used the age-65 Medicare eligibility discontinuity among emergency admissions to identify a reduction in patient mortality.

Why it's a landmark: Used a Starbucks natural experiment to show mandatory calorie posting modestly reduced calories purchased, concentrated among high-calorie consumers.

Why it's a landmark: Exploited a Medicaid lottery for the first randomized evidence that coverage raised utilization and financial security but not measured physical health.

Why it's a landmark: Introduced 'behavioral hazard,' showing cost-sharing can cut high-value care when patients underuse it, complicating the standard moral-hazard logic.

Why it's a landmark: Used Medicare movers to separate place from person, showing location causally affects mortality largely through healthcare use.

Why it's a landmark: Used Facebook-scale networks to measure social capital, showing cross-class friendship ('economic connectedness') strongly predicts upward mobility.

Why it's a landmark: Argued regression inference hinges on fragile priors, introducing extreme-bounds sensitivity analysis and catalyzing the drive toward credible identification.

Why it's a landmark: Introduced the LATE framework, showing instrumental variables identify treatment effects only for compliers under a monotonicity assumption.

Why it's a landmark: Recast IV within the Rubin potential-outcomes model, formalizing the exclusion, monotonicity, and independence assumptions underlying causal instrument estimates.

Why it's a landmark: Introduced the resume-audit correspondence experiment, documenting large callback gaps from randomly assigned racially distinctive names.

Why it's a landmark: Codified the design-based 'credibility revolution,' arguing research-design transparency, not structural modeling, secured empirical economics' reliability.

Why it's a landmark: Developed Neyman-orthogonal, cross-fitted estimators that let machine-learning nuisance estimates deliver valid inference on low-dimensional causal parameters.

Why it's a landmark: Introduced group-time average treatment effects with a clean control group, providing robust DiD estimators under staggered adoption and heterogeneity.

Why it's a landmark: Showed two-way fixed-effects estimates are contaminated 'negative-weight' averages under heterogeneous effects, and proposed a robust alternative estimator.

Why it's a landmark: Decomposed staggered DiD into all two-by-two comparisons, exposing the bias from using already-treated units as controls.

Why it's a landmark: Showed event-study coefficients mix effects across cohorts under heterogeneity, and proposed an interaction-weighted estimator for clean dynamic estimates.

Why it's a landmark: Replaced binary parallel-trends tests with a bounded sensitivity analysis, deriving robust confidence sets under formal restrictions on trend violations.

Why it's a landmark: Developed an imputation estimator that recovers untreated potential outcomes from controls, yielding efficient, robust event-study estimates under heterogeneity.