Vision — outdoor time against myopia, and screening and correction for achievement

Verdict

This topic holds two claims that get blurred together constantly, and the blurring is the error worth preventing.

(A) Outdoor time reduces myopia incidence. This is a health finding, not an education finding. It appears in an archive about educating children because it is decided by a school schedule — recess policy, whether a class is held outside — and for no other reason. Not one of the outdoor trials measured achievement, cognition, or attainment. Four cluster RCTs and a Cochrane review measured cycloplegic refraction, axial length, and myopia onset; none of them measured a test score. That absence is itself the finding, and it should be stated to any founder who has been sold outdoor time as a learning intervention.

(B) Screening for and correcting refractive error improves achievement. This is a genuine education claim with genuine RCTs behind it, and the verdict field describes this claim. The effect is real, small, concentrated in the children who actually could not see, and it fades. Both US trials — the ones using independent standardized tests rather than researcher-designed ones — show a first-year gain that does not survive to year two.

moderate-support rather than strong-support because five randomized trials agree on direction while disagreeing on magnitude by a factor of two; because the two largest point estimates come from researcher-designed maths tests and one of them rests on a one-sided test; and because the durable effect, which this archive scores separately and preferentially, is approximately zero. moderate-support rather than mixed because the fadeout here is not the usual fadeout of a taught skill — glasses are a prosthesis, and when provision stops and wear collapses the child is simply uncorrected again. That is a programme-design failure, not a refutation of the mechanism.

What the evidence shows

Claim B — vision screening and correction → achievement.

Source	Design	Grade	Key effect
Glewwe 2018	multi-arm RCT, Title I schools, 3 Florida districts	B	Screening alone: null. Screening + exams + glasses +2.0 pp FCAT pass rate (independent standardized), averaged over all students; impact fades out
Neitzel 2021 (Vision for Baltimore)	cluster RCT, 120 schools / 2,304 students, grades 3–7	B	i-Ready reading ES 0.09 at 1 year (0.28 lowest baseline quartile, 0.25 special education); no sustained impact at 2 years; no impact on PARCC at all
Ma 2014	cluster RCT, 252 schools / 3,177 children	B	Maths +0.11 SD (95% CI 0.01–0.21) on a specially designed test — below the 0.20 SD the trial was powered for; observed wear 41% vs 26%
Glewwe 2016	township-randomized RCT, rural western China, grades 4–6	B	+0.16 to 0.22 SD in children with poor vision; preferred estimates drop contaminated township pairs
Ma 2018	cluster RCT, 31 schools / 1,200 eligible, early vs late referral	C	+0.25 SD maths — but one-sided P = .04, CI lower bound 0.01, researcher-designed test
Yi 2015	cluster RCT, 94 schools / 728 children	B	Observed wear 68.3% vs 23.9% with a teacher incentive (OR 11.5); 41% with free glasses and no incentive

Measure type explains most of the spread. The two Chinese trials reporting 0.11–0.25 SD used tests the investigators built. The two US trials, using an independent state accountability assessment and a commercial normed assessment, report ES 0.09 and roughly 2 percentage points on a pass rate. That is the 2× inflation this archive expects from treatment-aligned measures, showing up cleanly within a single literature. The honest headline number is ~0.1 SD on an independent standardized measure, in the children who needed glasses.

Screening alone does nothing. Glewwe's multi-arm design is the most policy-relevant result here, because screening is the half that schools already do and the half that is cheap. Enhanced screening on its own was insufficient to move maths or reading. Identifying the problem and not fixing it buys nothing.

Claim A — outdoor time → myopia.

Source	Design	Grade	Key effect
He 2022 (STORM)	3-arm cluster RCT, 24 schools / 6,295 students, wearable-monitored	A	2-yr incidence IRR 0.84 (+40 min) and 0.89 (+80 min); no dose-response — objective monitoring showed both arms hit the same 127 min/day
He 2015 (Guangzhou)	cluster RCT, 12 schools / 1,903 children, 3 years	B	Incidence 30.4% vs 39.5% (−9.1 pp, 95% CI −14.1 to −4.1); refraction +0.17 D; axial length null (P = .07)
Kido 2024 (Cochrane)	meta of 5 RCTs only / 10,733 participants	B	Incidence RR 0.84 at 2 yrs (moderate certainty); CI crosses 1 at both 1 and 3 years
Xiong 2017	meta of 25 studies, mixed designs	C	Incident myopia RR 0.536 in trials; progression in already-myopic eyes: R² = 0.00064
Wu 2018	cluster RCT, 8 of 24 schools withdrew pre-enrollment	C	Myopic shift 0.35 vs 0.47 D; ≥1000 lux is sufficient — hallways and shade count
Dhakal 2022	overview of 7 systematic reviews	C	Progression effect 0.13–0.17 D/yr judged clinically insignificant; 7/7 reviews had ≥1 critical AMSTAR-2 flaw
Chen 2026	post-hoc dose-response inside STORM	C	Premyopic children: no significant benefit at any dose; hyperopes benefit, plateauing ~120 min/day

Prevention, not treatment. Every credible synthesis lands on the same dissociation: outdoor time lowers the rate at which children become myopic and does approximately nothing for children who already are. Xiong's dose-response in myopes returns an R² of 0.0006 — not a small effect, an absent one.

The Cochrane numbers are weaker than the flagship trial. Restricted to RCTs, pooled incidence is RR 0.84 at two years with moderate certainty, and the confidence interval includes no benefit at both one and three years. Guangzhou's headline −9.1 percentage points is the single largest estimate in the pool, and even within that trial the axial-length outcome — the harder biometric endpoint — was not significant.

Hereditarian-lens assessment

Risk: low, because both verdicts rest on cluster-randomized allocation. But this topic is where the hereditarian premise produces its most interesting result in the whole archive, and it runs in an unfamiliar direction.

Refractive error is among the most heritable common human traits. Lopes 2009 puts h² at 77% (95% CI 68–84%) across 2,301 twin pairs, with shared household environment at 7%, falling to about 2% once age effects are modelled — the same near-zero c² that shows up for cognitive ability in adulthood. Tedja 2018 adds 161 independent loci and a cross-ancestry genetic correlation above 0.78, meaning the East Asian myopia epidemic is an environmental shift acting on a genetic architecture Europeans share, not a different disease.

Then Mendelian randomization inverts the usual story. The normal hereditarian move is to say a correlation between an educational input and a child outcome is confounded by heredity. Here, genotype is used as the instrument instead of being the confounder, and the answer is that the education → myopia arrow is real and larger than the observational estimate: Mountjoy 2018 finds −0.27 D per year of education causally versus −0.18 D observationally, while the reverse direction — myopia affecting educational attainment — is null (−0.008 years per dioptre, P = 0.6). A UK graduate with 17 years of schooling ends up at least a dioptre more myopic than someone who left at 16. Schooling damages eyesight; bad eyesight does not measurably shorten schooling.

And the two halves of this topic then join up. Clark 2023 runs multivariable MR and finds that accounting for time outdoors shrinks the education → myopia effect by 40% — about half the damage schooling does to children's eyes is not near work at all, it is the hours indoors. The genetically-instrumented mechanism and the randomized-trial intervention are the same variable. Tedja's molecular finding closes the loop: the pathway the risk alleles act through is light-induced retina-to-sclera signalling. A highly heritable trait with a cheap environmental lever, because the heritable pathway and the lever are the same pathway.

The contrast with the rest of this archive is the point. Uncorrected refractive error is one of the very few genuinely environmental, cheaply removable barriers to learning in the whole database. A child who cannot see the board is not exhibiting low ability, and the largest effects in Baltimore were in the lowest baseline quartile (ES 0.28) and in special education (ES 0.25) — precisely the children whose sensory deficit was most likely being read as a cognitive one.

Boundaries & what critics say

• The outdoor literature's own synthesists are unimpressed with it. Dhakal's overview found every one of seven systematic reviews carried at least one critical AMSTAR-2 flaw and only two had low risk of bias on ROBIS. Cochrane rates most timepoints low certainty.
• The dose is unknown, because it was never delivered. STORM randomized 40 versus 80 extra minutes and objective wearables showed both arms achieved 127 min/day. The trial that was designed to find the dose-response found that compliance ate the contrast.
• Timing matters and the window may close early. Chen's post-hoc analysis inside STORM finds no significant protection at any outdoor dose for children who are already premyopic; the benefit sits with hyperopic children who still have refractive reserve.
• Only 12 clusters in Guangzhou, and 8 of 24 schools withdrew from the Taiwan trial before enrollment. These are not enormous cluster counts, and the Taiwanese withdrawal is a randomization-integrity problem, not a footnote.
• The Baltimore null on PARCC is the sharpest available challenge to claim B. The gain appeared on the vendor progress-monitoring assessment and not on the state accountability test given to the same children in the same year.
• The evidence window is narrow. Outdoor trials ran on 6-to-9-year-olds; the correction trials on grades 3–7 and 4–6. Nothing here speaks directly to 14-to-18-year-olds, and preschool hyperopia evidence is almost entirely observational.
• Every reported effect is small and none of it touches g. Correcting vision removes an obstacle to accessing instruction. It is not an ability intervention and should never be described as one.

Practical guidance

• Screen, then correct — screening alone is a null result, not a half-measure. Budget for exams and glasses or do not bother.
• Treat wear rate as the intervention. Free glasses alone gets ~41% observed wear; adding a teacher accountability incentive gets ~68%. Everything downstream scales with this number, and the published effect sizes were achieved at low compliance — the ceiling is higher than 0.1 SD if the glasses actually stay on.
• Re-screen and re-supply every year. Both US trials fade because provision stopped. Glewwe's own conclusion is that follow-up actions are necessary to sustain the gains.
• Expect roughly 0.1 SD, and expect it where you would predict: the lowest-achieving quartile and children already flagged for special education. Do not promise a schoolwide average.
• Move instruction outdoors for eyesight, and say so. Forty extra minutes a day is a defensible myopia-prevention policy with randomized backing. It is not an achievement policy, and staking it on achievement will get it cancelled when the test scores do not move.
• You do not need full sun. Wu's light-meter data show ≥1000 lux suffices — covered walkways, shade under trees. This makes the policy feasible in climates where "outdoor class" sounds absurd.
• Do not sell outdoor time to parents of already-myopic children as treatment. It prevents onset; it does not slow progression once myopia is established. That is an atropine/optical-intervention question and out of scope here.

Open questions

• No outdoor-time trial has ever measured an academic outcome. Adding a standardized test to one of these trials would cost almost nothing and would settle whether the recess-outdoors decision has any achievement consequence at all.
• Does sustained correction produce a durable effect? Every trial stopped provision and watched the effect decay. Nobody has run multi-year continuous re-supply with annual re-screening, which is the design the fadeout results actually call for.
• Attainment outcomes are unmeasured. Test scores fade; this archive tracks graduation and earnings separately, and no vision trial has followed children that far.
• The true dose-response for outdoor time is unestablished — STORM's compliance failure means the ceiling of the effect is unknown.
• Does the education → myopia causal effect mean anything for schools that raise achievement most? If schooling causes myopia at −0.27 D/year, an intensive academic programme is buying test scores partly with eyesight. Nobody has priced that trade-off.