The Effect of Schooling on Cognitive Skills

128,617 native Swedish males attending upper-secondary school at the time of military enlistment

Swedish 18-year-old males in the academic track, tested at compulsory military enlistment before the late-1994 test-battery change

Review of Economics and Statistics 97(3), 533-547. Full text read from NBER working paper 18484 (October 2012); the published version adds Öckert as an author. This is the cleanest measurement of the return to a DAY of schooling that exists. Sweden required almost every 18-year-old male to sit a battery of cognitive tests at enlistment, and the military — not the enlistee — assigned the date, using only birthdate, parish and (at some offices) expected graduation date, subject to travel logistics and office capacity. Conditional on those three variables, both age at test and number of school days completed vary randomly, which lets the paper separate four things that are normally hopelessly collinear: days of instruction, age at test date, relative age in class, and cohort. Enlistment dates were strictly enforced, so compliance is near-total. The authors report the design's own fragility honestly: the estimates are NOT robust to dropping the conditioning variables, in which case the synonyms coefficient halves and the technical-comprehension coefficient falls to almost zero. Scope limits: males only, academic track only, one country, and the outcome is a cognitive battery rather than curricular achievement.

Ten extra days of school instruction raise crystallized-intelligence scores by 1.1% of a standard deviation on the synonyms test and 0.8% on technical comprehension, while ten extra NON-school days have virtually no effect — the day only counts if it is a school day. Scaled to a Swedish school year of 180 days this is +0.21 SD on synonyms and +0.14 SD on technical comprehension, which is squarely inside the archive's benchmark of 0.2-0.4 SD per year of schooling. The fluid-intelligence tests behave completely differently: spatial and logic scores rise modestly with AGE (0.3-0.5% of an SD per 10 days older) but the additional effect of a school day is statistically indistinguishable from zero and if anything negative, which the authors read as schooling not contributing to fluid ability. The return to school days is HOMOGENEOUS — similar effect sizes by ninth-grade math and Swedish grades, by mother's and father's education, and by father's earnings — so there is no well-identified subgroup for whom extra days are worth much more, which is bad news for targeting arguments.

Low, and unusually so. Test-date assignment is made by a military bureaucracy on birthdate, parish and expected graduation date, none of which is chosen by the family, and the paper demonstrates conditional randomness directly by showing that days of schooling are uncorrelated with ninth-grade grades, parental education and paternal income once the conditioning variables are included. The comparison is between two boys of the same birthdate and parish who happened to be called up a few weeks apart, so heritable ability is balanced by construction. This is also why the fluid/crystallized split is credible rather than an artefact of who gets more schooling.

No direct replication of the Swedish enlistment design exists; the identifying institution is unusual and the data are administrative. The finding is nonetheless corroborated structurally from two directions already in this archive: Ritchie and Tucker-Drob's meta-analysis of natural experiments finds schooling durably raises measured IQ, and Ritchie, Bates and Deary find the gain is on specific taught skills rather than on the general factor — which is exactly the crystallized-yes / fluid-no pattern reported here, arrived at by a different design.