Explore the evidence
41 of 88 decisions · clear everything
Early gains fade by default — halving every 12–18 months, faster when bigger. What persists is trajectory (placement, graduation), not ability.
strong supportconf: highgc: lowFadeout — why early gains disappear, and what actually persists · early-childhood · ages 4–18 · structure
Testing yourself beats rereading — robust in real classrooms; honest durable size ~0.1–0.3 SD, biggest after delay, thinnest on transfer.
strong supportconf: highgc: lowRetrieval practice (the testing effect) · practice · ages 8–18 · method
Spacing beats massing at equal total time — the most robust finding in learning science. Space repetitions at ~10–20% of how long you need to remember.
strong supportconf: highgc: lowSpaced (distributed) practice · practice · ages 6–18 · method
Physical capacity is about as heritable as cognitive ability (~60%, no shared environment). Whether trainability is a stable trait remains unproven — the skeptics are winning.
strong supportconf: highgc: lowTalent and trainability — what is heritable, and what that does not license · talent · ages 4–18 · input
Teacher quality is the largest within-school lever — 1 SD of teacher ≈ 0.10–0.15 SD/yr, worth more than ten fewer students — and credentials predict none of it. Select; don't workshop.
strong supportconf: highgc: lowTeacher quality — selection over credentials and workshops · teachers · ages 4–18 · structure
High-dosage tutoring is education's most reliable lever: ~0.29 SD in trials, ~0.2 well-scaled — not Bloom's 2σ. Groups of 3–4 work; 1:1 is unnecessary.
strong supportconf: highgc: lowTutoring — the honest effect, the Bloom 2-sigma myth, and what survives scale · tutoring · ages 5–18 · method
Group within classes or across grades by subject: modest, nearly free wins. Whole-school streaming does nothing, and early between-school tracking harms the bottom.
moderate supportconf: highgc: lowAbility grouping and tracking — four practices, four verdicts · grouping · ages 5–18 · structure
Accelerate ready kids: they keep pace with older classmates, bank a year, and show no social-emotional harm at 50. The gifted label itself does nothing; the content does.
moderate supportconf: highgc: mediumAcceleration and gifted programs — the label vs the content · grouping · ages 5–18 · structure
Smaller classes buy real K-1 gains that fade on tests yet persist in attainment — at roughly triple tutoring's cost, and diluted to nothing when scaled fast.
mixedconf: highgc: lowClass-size reduction · class-size · ages 4–18 · structure
Curriculum is nearly free, so choosing beats not choosing — but the payoff is avoiding a demonstrated loser, not finding a magic winner. Content is the high-upside bet.
mixedconf: highgc: lowCurriculum choice as a school-level lever · curriculum · ages 4–14 · structure
The 10,000-hour rule is dead: practice explains ~14% of performance variance, is itself heritable, and identical twins 20,000 hours apart didn't differ. Necessary, insufficient.
mixedconf: highgc: mediumDeliberate practice and the 10,000-hour rule · deliberate-practice · ages 6–18 · method
Feedback on the task helps modestly; feedback on the person backfires — a stable third of studies reverse. The famous 0.4–0.7 number has no computed source.
mixedconf: highgc: lowFeedback and formative assessment · feedback · ages 5–18 · method
Mastery learning moves tests of what it taught (~0.25) and barely moves independent measures (~0.05) — and time-to-mastery gaps widen, converting ability differences into time.
mixedconf: highgc: lowMastery learning (teach → test → reteach to criterion → advance) · mastery · ages 6–18 · method
At scale, pre-K doesn't durably raise test scores — Tennessee went negative — yet Boston shows real attainment gains beside a test-score zero. It buys trajectory, not ability.
mixedconf: highgc: lowPreschool at scale — Head Start, state pre-K, and what universal provision delivers · early-childhood · ages 4–5 · structure
Unallocated money is the worst buy in the database; the same dollars as No-Excuses charters buy +0.3–0.4 SD/yr — and the average charter is a precise null.
mixedconf: highgc: lowSchool spending, charters, and what school-level choices actually move outcomes · school-structure · ages 4–18 · structure
Starting school older mostly manufactures an age-at-test artifact: the IQ effect collapses to ~−0.07 once identified. Real residues: less hyperactivity, unchanged attainment.
mixedconf: highgc: lowSchool starting age, relative age, and academic redshirting · early-childhood · ages 4–7 · structure
Most measured 'home environment' effects are parents' genes: three-quarters of parent-child transmission isn't rearing, and a whole better childhood buys ~4 IQ points.
mixedconf: highgc: lowThe early home environment — what parents can and cannot causally move · early-childhood · ages 4–10 · input
Brain training improves the trained task and nothing else: far transfer to intelligence or achievement is 0.001 against active controls.
no effectconf: highgc: lowDoes working-memory or brain training transfer to intelligence and achievement? · transfer · ages 4–18 · debunked
Matching instruction to 'learning styles' does nothing: real matching experiments return d=.04. The correlational literature IS the myth.
debunkedconf: highgc: lowDoes matching instruction to a child's "learning style" improve learning? · debunked · ages 4–18 · debunked
Later bells buy adolescents 40+ measured minutes of sleep — the best-identified positive effect in the archive; the achievement payoff is real but far smaller.
strong supportconf: mediumgc: lowDo later school start times increase adolescent sleep, and does that raise achievement? · health · ages 11–18 · input
Unassisted discovery loses to explicit teaching; well-scaffolded guided discovery beats both. The operative variable is guidance, not ideology.
moderate supportconf: mediumgc: lowExplicit instruction vs discovery/inquiry — how much guidance? · instruction-style · ages 4–18 · method
Mixing problem types helps where confusion is the enemy — discriminating similar categories, mixed math practice — and is useless or worse for facts and prose.
moderate supportconf: mediumgc: lowInterleaving (mixing problem types vs blocking) · practice · ages 10–18 · method
Diagnose what a child already knows: teachers cut 40–50% of curriculum for high-ability children and achievement ROSE. The payoff is skipping, not monitoring.
moderate supportconf: mediumgc: mediumPlacement and mastery diagnosis — deciding what to teach next from evidence of current skill · assessment · ages 5–18 · structure
Outdoor time prevents myopia from starting (not progressing); screening plus free glasses raises test scores in children who need them. Two claims, both real.
moderate supportconf: mediumgc: lowVision — outdoor time against myopia, and screening and correction for achievement · health · ages 4–18 · input
Novices learn faster studying solutions than solving problems — then the effect reverses with expertise. Use worked examples early; fade them.
moderate supportconf: mediumgc: lowWorked examples (studying solutions vs solving problems) · instruction-style · ages 10–18 · method
Correcting a real deficiency moves cognition (iron in anaemic children: 0.79 SD); supplementing already-fed children moves nothing (35 RCTs, 19,343 children).
mixedconf: mediumgc: lowBreakfast, school meals, and micronutrients — what feeding children actually buys · health · ages 4–18 · input
A missed day costs little (~0.005 SD); interventions reliably buy days back cheaply, but nobody has shown the recovered days move achievement.
mixedconf: mediumgc: mediumChronic absenteeism — does raising attendance raise achievement? · time · ages 4–18 · input
Sort by what the software replaces: adaptive drill inside the school day buys +0.05-0.20 SD; a device, a connection or the teacher buys zero to negative; LLM tutors have no usable evidence at all.
mixedconf: mediumgc: lowDoes educational technology raise learning — CAI, adaptive software, devices, screens, and AI tutors? · edtech · ages 5–18 · structure
The received wisdom that retention is harmful was two measurement errors, not a finding. Fix them and the achievement effect is zero — but the grade at which you hold a child back decides everything.
mixedconf: mediumgc: lowGrade retention — holding a child back versus promoting, and test-based promotion policies · retention · ages 5–18 · structure
Near-zero in primary and almost entirely correlational in secondary. The causal base is three small trials; what the homework IS beats how much of it there is.
mixedconf: mediumgc: mediumHomework — effects by age and by dosage · homework · ages 5–18 · structure
Classroom order is causally worth a great deal — one disruptive peer costs classmates 3% of adult earnings — yet no branded behaviour programme reliably delivers it and exclusion makes things worse.
mixedconf: mediumgc: lowHow should a school run its classrooms and its discipline system? · behavior · ages 4–18 · structure
Taking time away hurts measurably; adding it back buys almost nothing. The return per hour is ~0.02-0.03 SD, concave, and near zero in disorderly classrooms.
mixedconf: mediumgc: lowLength of the school day and year, time-on-task, extended time, and summer · time · ages 4–18 · structure
Perry and Abecedarian: n=123 and n=111, IQ gains gone by adolescence, attainment effects real but multiplicity-fragile. Too thin to carry the policy built on them.
mixedconf: mediumgc: lowPerry Preschool and Abecedarian — what the famous studies actually establish · early-childhood · ages 4–5 · structure
Coaching buys ~0.25 SD of score and has for 42 years — a fraction of what's advertised — and none of it is ability. The score moves; the construct doesn't.
mixedconf: mediumgc: mediumTest preparation — does coaching raise scores, and does a raised score mean raised ability? · assessment · ages 5–18 · structure
What parents are asked to DO is the whole effect: tutoring a skill d≈1.15, listening to reading 0.51, reading aloud ~0.18 — same parent, same child. All deflate under better designs.
mixedconf: mediumgc: mediumThe parent as instructor — what survives when the person teaching is the person who raised the child · homeschool · ages 4–18 · method
Trust the composite; distrust the breakdown. Subtest strengths-and-weaknesses replicate at chance on retest, and most of what people read off a score report is noise.
mixedconf: mediumgc: lowWhat a standardized achievement score does and does not license · assessment · ages 5–18 · structure
School exercise programs don't raise achievement: four of the five largest RCTs are null, and more dose doesn't rescue it. Exercise for health, not for grades.
no effectconf: mediumgc: lowPhysical activity as an input — dose, fitness, and school outcomes · health · ages 4–18 · input
Never operationalised well enough to test: no MI-based trial meets even a loose evidential bar. Four decades in, a valid evaluation is still impossible.
insufficientconf: mediumgc: mediumShould Gardner's multiple intelligences be used as a basis for instruction? · debunked · ages 4–18 · debunked
Brain Gym, Fast ForWord, coloured overlays, the Mozart effect: tested and failed, or never evidence-based at all. Mozart works exactly as well as any other music.
debunkedconf: mediumgc: lowBrain Gym, Whole Brain Teaching, and the smaller classroom brain fads · debunked · ages 4–18 · debunked
No study of homeschooling has a counterfactual — the famous percentile claims are selection all the way down. The honest answer: nobody knows.
insufficientconf: lowgc: highDoes homeschooling raise achievement and attainment, or does it select the children who would have done well anyway? · homeschool · ages 5–18 · structure
Homeschoolers get plenty of social contact but measurably fewer, closer peer ties; whether that helps or harms has never been causally tested.
insufficientconf: lowgc: highHomeschool socialisation — the most-asked question, and the least answerable one · homeschool · ages 6–18 · structure
The URL is the state — any view you build here is a permalink you can hand to someone mid-argument.