School types in adolescence and subsequent health and well-being in young adulthood: An outcome-wide analysis

12,288 adolescents (mean age 14.56 in 1999) followed to mean age 25.10 in 2010; n=187 home-schooled

Growing Up Today Study (GUTS), the children of participants in Nurses' Health Study II — overwhelmingly non-Hispanic White, all children of well-educated mothers

Outcome-wide design: one exposure (school type at ~14) against ~40 pre-registered adulthood outcomes, each model adjusted for demographics, maternal subjective SES, household income, census-tract characteristics, family structure, family dinners, parental religious attendance, parental relationship satisfaction, maternal depression and smoking, and BASELINE values of the health and behaviour outcomes. Bonferroni correction applied across outcomes; E-values reported for sensitivity to unmeasured confounding. The home-schooled cell is small (n=187) and cannot be split into structured vs unstructured.

Relative to public school, home-schooling survived multiple-testing correction on four outcomes: more frequent volunteering (beta 0.33, 95% CI 0.15-0.52), greater forgiveness (beta 0.31, 95% CI 0.16-0.46), more frequent religious service attendance (RR 1.51, 95% CI 1.27-1.80) — and LOWER college degree attainment (RR 0.77, 95% CI 0.67-0.88, i.e. 23% less likely). Suggestive-only: greater purpose in life, less marijuana use, fewer lifetime sexual partners, higher PTSD risk. E-values for the home-schooling associations were 1.54-2.04, which the authors call at least moderately robust; by contrast the private- and religious-school comparisons had confidence-interval E-values of at most 1.24 and are explainable by modest confounding.

High but partially bounded. Every participant is the child of a nurse, so parental education is nearly constant by construction — an unusual and useful restriction. Prior values of the behavioural outcomes are adjusted. Nothing adjusts for child genotype or for the reasons a family chose home-schooling, and reverse causation (a struggling child is withdrawn from school) is not addressable here and would produce exactly the attainment result observed.

Single cohort, never replicated. The attainment result is the only well-adjusted longitudinal estimate of home-schooling on a hard outcome in the literature and it points the opposite way from the advocacy consensus, which is a reason to want a replication rather than a reason to discount it.