School starting age, relative age, and academic redshirting

Verdict

This literature splits cleanly into a large artifact and a small real residue, and almost all popular discussion of "redshirting" confuses the two.

The artifact. Older children in a cohort score higher — but mostly because they are older when tested, not because starting later helped them. The decisive study is the Norwegian decomposition: because military-conscription test dates vary independently of the school-entry mapping, it can separate the two. The naive starting-age coefficient on adult IQ is −0.754; once age-at-test is separately identified, the genuine starting-age effect collapses to −0.06 to −0.09 while age-at-test carries a coefficient 3-4× larger. Ultimate educational attainment: ~0. Earnings gains: present early, gone by age 30 — pure life-cycle timing. Consistent with this, the international relative-age gap roughly halves between grade 4 and grade 8.

The residue is real but narrow. Starting older genuinely reduces inattention/hyperactivity (−0.73 SD at age 7, still −0.69 at 11, concentrated in girls, and specific to that one subscale — which is why it isn't just reference bias). It genuinely reduces teen pregnancy. And in Sweden it genuinely raises adult earnings — but only for children of low-educated parents.

Redshirting is not what it's sold as. It is rare (4-5.5%), practised disproportionately by White, male, high-SES families, and there is no evidence that low-ability four-year-olds are redshirted more. It is a positional strategy, and its apparent benefits are confounded with the advantage that produces it.

What the evidence shows

The ADHD finding deserves special attention from anyone running a school. The diagnosis gradient is driven by teacher ratings (0.59 SD discontinuity) far more than parent ratings (0.14 SD) — teachers judge against classroom peers, parents don't. The youngest children in a class are roughly twice as likely to be on stimulant medication by grades 5-8. That is a labelling artifact with serious, persistent consequences, and it is one of the most robust findings in education research.

Norway and Sweden disagree about whether starting age affects ultimate attainment (Norway: no; Sweden: yes). The leading explanation is institutional — Sweden shows that postponing ability tracking to 16 shrinks the starting-age effect, implying it operates through early institutional sorting rather than through learning. Unresolved.

Hereditarian-lens assessment

Risk: low for the causal claims — these are large population-register natural experiments with cutoff-based identification, so family selection is not the identifying variation. Two notes:

• No grade-A evidence exists here. There are no RCTs or lotteries; everything rests on cutoff designs, so residual confounding (parental sorting around cutoffs, birth-timing manipulation) cannot be entirely excluded. Confidence is high because six independent B-grade designs converge, not because any single study is decisive.
• The redshirting-selection finding is the confounded one. That advantaged families redshirt more is descriptive; the apparent benefits of being older-for-grade are entangled with the SES — and the heritable parental traits — that drive the choice. This is exactly the pattern the database's premise predicts, and it is why "older-for-grade children do better" should not be read as advice.
• Nothing here durably moves g. The one place ability appears to move is age-at-test, which is a measurement artifact, not a gain.

Practical guidance

• Don't redshirt for academic advantage. The test-score benefit is mostly measurement, it halves by grade 8, and Norway finds no effect on ultimate attainment.
• Do consider a later start for self-regulation reasons if a specific child is struggling with attention — that residue is real and persists to at least age 11.
• Adjust for relative age when assessing young children. Compare a child to same-age peers, not same-grade peers, before referring for ADHD assessment or special-education placement. The teacher-vs-parent rating gap is the tell.
• Place on measured readiness, not birth date — the same conclusion the database reached for algebra timing, from an entirely separate literature.
• If you serve disadvantaged families specifically, the Swedish result suggests a later start may carry genuine long-run benefit for them — the one subgroup where it survives.

Open questions

• Norway vs Sweden on attainment — institutional differences are the leading explanation but unresolved.
• The mental-health findings have never had the Black/Devereux/Salvanes age-at-test decomposition applied to them, so some of the ADHD/self-regulation result could still be relative-age artifact.
• No grade-A design exists anywhere in this literature.