Too Young to Leave the Nest? The Effects of School Starting Age
Black, S. E., Devereux, P. J., & Salvanes, K. G. · 2011
grade Bnatural-experimentindependentmixednumbers spot-checked
Sample
Norwegian population registry; IQ regressions N=664,012 men; attainment N=514,662; teen pregnancy N=218,674
Population
Norwegian cohorts born 1962-1988, with military-conscription IQ tests at ~18-20.
Design
THE decomposition paper. Because Norwegian military test dates vary independently of the birth-cohort/start-age mapping, it can separate genuine STARTING-AGE effects from AGE-AT-TEST artifacts — which nothing else in this literature does cleanly.
Key findings
The single most important paper on starting age, because it shows most of the apparent benefit of starting older is an AGE-AT-TEST ARTIFACT: the naive IQ coefficient of -0.75 collapses to -0.06/-0.09 once age-at-test is separately identified, while age-at-test itself carries a coefficient 3-4x larger. Ultimate educational attainment is unaffected. Real residue: less teen pregnancy for later-starting girls.
Counterfactual
Children starting school slightly older vs younger due to the January 1 entry cutoff.
Genetic confound
Population registry with a cutoff-based instrument; family selection is not the identifying variation.
Replication notes
Sweden (Fredriksson & Öckert) finds a positive attainment effect where Norway finds none — institutional differences (tracking age, redshirting norms) may explain the divergence; unresolved.
DOI / URL
10.1162/REST_a_00081
Effects
| Outcome | Metric | Value | Measure | Timing | Vs | Horizon | Class |
|---|---|---|---|---|---|---|---|
| Adult IQ: naive (undecomposed) starting-age estimate | coefficient | -0.754 — the inflated, conflated number | standardized | adult | business-as-usual | adulthood | g |
| Adult IQ: genuine STARTING-AGE effect (2SLS) | coefficient | -0.06 to -0.09 stanine (~0.03-0.045 SD) — small and near-zero | standardized | adult | business-as-usual | adulthood | g |
| Adult IQ: AGE-AT-TEST effect (2SLS) | coefficient | +0.17 to +0.22 — 3-4x LARGER than the starting-age effect; this is the artifact | standardized | adult | business-as-usual | adulthood | g |
| Educational attainment | coefficient | ~0 (0.009 to -0.062 across specifications) — no effect | mixed | adult | business-as-usual | adulthood | attainment |
| Teen pregnancy (girls) | 2SLS | -0.018; starting ~3 months older reduces teen-birth probability ~0.5pp — a genuine effect | mixed | adult | business-as-usual | adulthood | behaviour |
| Earnings | coefficient | positive short-run, VANISHED by age 30 — pure life-cycle timing | mixed | adult | business-as-usual | adulthood | attainment |
Cited by
- School starting age, relative age, and academic redshirtingmixedconf: highgc: low