The Evidence on Teaching

Long-Term Effects of Class Size (Swedish RD, grades 4-6)

Fredriksson, P., Ockert, B., & Oosterbeek, H. · 2013

grade Bnatural-experimentindependentunreplicatednumbers spot-checked
Sample
30,818
Population
Swedish students, grades 4-6 (1980s), followed to ages 27-42.
Design
Max-class-size rule RD; registry outcomes.
Key findings
The strongest long-run pro-class-size result — and the database's only credible instance of an instructional lever durably moving IQ-TYPE general-ability measures (small, at 18, for low-income sons). Its two-ledger heterogeneity (score gains neutral; wage gains to the advantaged) complicates every simple ATI story.
Genetic confound
Rule-predicted class size orthogonal to family background.
Replication notes
Verified incl. the heterogeneity correction (skills income-neutral at 13; wages concentrate among the advantaged). The direct wage effect is 3-6x what imputing from scores predicts — score-per-dollar comparisons may understate class-size returns.

Effects

OutcomeMetricValueMeasureTimingVsHorizonClass
Cognitive skills at 13 (IQ-type tests)SD+0.020/pupil removed — SIMILAR across parental incomestandardizedage 13business-as-usualend-of-treatmentg
General ability at 18 (AFQT-like enlistment battery)effectsignificant for sons of LOW-income parents — a durable general-ability result, small and subgroup-fragilestandardizedage 18business-as-usualover-2yrg
Wages ages 27-42%+0.7% per one-pupil reduction; concentrated among men/HIGH-income families; IRR 18.6% (9.4% with costs doubled/benefits halved)standardizedages 27-42business-as-usualadulthoodattainment

Cited by