Objective To investigate early-life biological and social predictors of educational outcomes, and compare the nature and magnitude of these effects across twentieth century Sweden.

Design Multi-generational data from a representative, population-based birth cohort, with linkage to routinely collected data.

Setting Sweden.

Participants 9829 Swedish male and females born 1915–1929 and 9465 of their grandchildren born 1973-1980, restricting participants to those who remained alive and in Sweden until age 20.

Characteristics measured at birth Sex, birthweight for gestational age, preterm birth, birth multiplicity, birth order, mother's age, mother's marital status and family social class.

Educational outcomes School achievement was measured using standardised schoolmarks in elementary school. Education continuation was measured as a) senior school attendance and b) entrance to higher education.

Results The predictors of both school achievement and education continuation were very similar in the two cohorts, and effect sizes were usually at least as large in the younger cohort. In both cohorts, the independent predictors of better schoolmarks were: female gender (adjusted effect size 0.35 standard deviations (SD) in 1915–1929, 0.41SD in 1979-1980); higher birthweight for gestational age (0.09SD in 1915–1929, 0.23SD in 1979–1980 for highest vs lowest quintile); lower birth order (eg, 0.33SD in 1915–1925, 0.65SD in 1973–1980 for birth order 1 vs 4–5); older mother (eg, 0.12SD in 1915–1929, 0.34SD in 1973–1980 for 35–39 years vs 20–24 years); married mother (0.14SD in 1915–1929, 0.15SD in 1973–1980 for married vs unmarried); and higher family social class (eg, 0.39SD in 1915–1929, 0.66SD in 1973–1980 for high/mediate non-manual vs semi/unskilled manual). There were no independent effects of preterm or twin status. The same characteristics predicted education continuation, except that for this outcome the older cohort now showed a marked male advantage and no birthweight effect. Even after adjusting for school achievement, education continuation was still predicted by lower birth order, older mother, married mother and higher social class.

Conclusions Multiple early-life characteristics predicted educational outcomes across the lifecourse. These included size at birth (foetal growth rate) and family composition effects which typically receive far less attention than socio-economic influences. A range of pathways including impaired cognitive development, are likely to mediate these effects. Most effects were remarkably stable across the half-century separating our cohorts, indicating their potential relevance for understanding educational inequalities in populations around the world. Greater understanding of educational inequalities would, in turn, shed light onto a major mechanism whereby health inequalities are created and recreated across generations.