Van der Klaauw 2014 — From Micro Data to Causality Forty Years of Empirical Labor Economics

---

title: Van der Klaauw 2014 — From Micro Data to Causality: Forty Years of Empirical Labor Economics aliases: ["Van der Klaauw 2014 — From Micro Data to Causality: Forty Years of Empirical Labor Economics"] tags: [labor-economics, causal-inference, natural-experiments, instrumental-variables, difference-in-differences, regression-discontinuity, sample-selection, potential-outcomes, LATE, microeconometrics, survey, weak-instruments, sufficient-statistics] sources: [] updated: 2026-07-13 kind: article author: Bas van der Klaauw date: 2014-06-19 url: https://doi.org/10.1016/j.labeco.2014.06.009

Summary

A retrospective survey of causal identification methods in labor economics from the early 1970s to 2014, tracing the arc from ordinary-least-squares (OLS)-based labor market analysis through the credibility revolution (natural experiments, instrumental variables (IV), difference-in-differences (DiD), regression discontinuity design (RDD)) to current debates about interpretation, welfare relevance, and the sufficient statistics approach. Written as an overview for the 2013 EALE (European Association of Labour Economists) meeting; intended as a compact pedagogical roadmap rather than an original contribution.

Key Claims

• The foundational problem: Heckman (1974) showed that standard regression methods suffer from endogeneity and selection; LaLonde (1986) showed empirically that non-experimental methods cannot reliably recover the true causal effect even with many controls.
• Mincerian wage equation: $\beta_1$ (returns to schooling) is endogenous because ability drives both schooling and wages; OLS is inconsistent regardless of how many observables are added (kitchen-sink failure).
• Sample selection model (Heckman 1979): deal with selective observation via a two-step correction using the inverse Mills ratio (IMR); exclusion restriction is essential because the IMR is nearly linear locally, making identification nonparametric without one.
• Potential outcomes: average treatment effect (ATE) and average treatment effect on the treated (ATET) defined; the fundamental problem of causal inference (Holland 1986) — only one potential outcome is ever observed; randomized experiments provide the gold standard but are rare in economics.
• Social experiments surveyed: LaLonde (1986) JTPA, Rand HI (Manning et al. 1987), STAR class size (Krueger 1999), Canadian Self-Sufficiency Project (Card & Robins 1998), Progresa (Schultz 2004).
• Nonparametric bounds (Manski 2003): without assumptions, identification region for ATE = width 1; Roy assumption narrows it; partial identification is credible where point identification requires strong assumptions.
• DiD: Ashenfelter dip as canonical parallel-trends violation; Card & Krueger (1994) minimum wage; Bertrand et al. (2004) serial correlation → clustered standard errors (SEs); synthetic controls (Abadie et al. 2010) for violations.
• IV: Angrist (1990) Vietnam lottery; Angrist & Krueger (1991) quarter-of-birth; local average treatment effect (LATE) interpretation (Imbens & Angrist 1994); weak instruments (Bound et al. 1995; Staiger & Stock 1997): F > 10 rule of thumb; limited-information maximum likelihood (LIML) preferred for multiple weak instruments.
• RDD: Van der Klaauw (2002) GRE scholarship threshold; Angrist & Lavy (1999) Maimonides rule; Hahn, Todd & Van der Klaauw (2001) theoretical formalization; sharp vs. fuzzy; McCrary (2007) density test; marginal treatment effect at the threshold.
• Sufficient statistics approach (Chetty 2009): bridge between reduced-form and structural; derives welfare-relevant statistics from theory that can be estimated via treatment-effect methods; advantages include fewer assumptions, robust to behavioral misspecification; disadvantages include policy-specificity and limited structural testing.
• Spillover effects: the stable unit treatment value assumption (SUTVA) fails when equilibrium effects are present; Manski (1993) reflection problem for peer effects (three sources: endogenous, contextual, correlated); Angrist (2013): regressing on peer mean includes own observation → coefficient should be unity by construction; experiments manipulating peer composition are the strongest evidence.
• Dynamic treatment evaluation (Abbring & van den Berg 2003): no-anticipation assumption as the key identifying restriction; mixed proportional hazard framework; ex-post effect on treated survivors.

Concepts Introduced or Extended

• Sample Selection Model — Heckman (1979) two-step correction for selective observation
• Instrumental Variables — weak instruments criterion (F > 10); Staiger-Stock formula
• Natural Experiments — survey of four decades of natural experiment methods in labor economics
• Difference-in-Differences — Ashenfelter dip; Bertrand et al. (2004) clustering solution
• Local Average Treatment Effect — pedagogical exposition of LATE and complier typology

Entities Mentioned

• Bas van der Klaauw

Quotes

"LaLonde (1986) showed empirically that endogeneity can be a major problem in microeconometric research... The non-experimental results are often different from those obtained from a randomized experiment."

"Heckman and Vytlacil (2001) stress that many of the conventional treatment parameters lack a direct link to an interpretable economic framework or, for example, a cost–benefit analysis."

My Take

A clear, accessible survey — useful as a reading list and narrative skeleton for the causal inference literature in labor economics. Best used as a complementary reference alongside the more detailed concept pages in this wiki. The paper deliberately avoids depth in favor of breadth; it doesn't adjudicate debates (e.g., LATE vs. structural) but signals them. The dynamic treatment evaluation section (Section 6.3) is the most original contribution relative to existing surveys, covering material not easily found in textbook-length treatments.