The science of learning: a one-page cheat sheet for parents
The strategies with the strongest evidence (spacing, retrieval, interleaving, feedback, the 85% target), each with its key study and a two-minute home version.
Six habits have the strongest evidence for helping children remember what they learn: spreading practice out, recalling instead of re-reading, mixing problem types, getting quick feedback with a short reason, seeing a worked example before trying, and practising at a difficulty where most answers are right. This sheet gives each one its key study, the effect size with its sample and its catch, a two-minute version to do at home, and how Peabody (peabody.kids), a free AI-driven adaptive learning platform for kids, builds it in. Peabody is free through the end of 2026.
How to use this: print the first table and stick it somewhere. The "at home" column is the part that matters; the "how Peabody uses it" column is there so you can see the same rule in the app. AI writes the stories; code decides the rest.
The six that hold up
| Strategy | What it is | Key study, sample, effect | Catch | Two minutes at home | How Peabody uses it |
|---|---|---|---|---|---|
| Spacing | Practise a thing on several days with gaps, instead of all at once | Cepeda, Vul, Rohrer, Wixted and Pashler 2008, Psychological Science, 1,354 adults, gaps up to 3.5 months: the best gap is about 20 to 40% of a one-week horizon and 5 to 10% of a one-year horizon. For children, Sobel, Cepeda and Kapler 2011, 39 fifth-graders: one-week spacing beat same-day on a five-week test. In maths, Murray, Horner and Göbel 2025, 27 studies, g = 0.28. | Too-short gaps hurt more than too-long; most child studies are small | Ask tonight's new word or fact again tomorrow at breakfast, then at the weekend | A known item comes back after 1, 3, 7 days and then a 21-day gap; a miss resets it. Story words need three right cold reads on three different days, then keeper checks at 7, 30 and 90 days |
| Retrieval practice | Pull it out of memory (close the book and say it) instead of looking at it again | Roediger and Karpicke 2006, Psychological Science, 120 students: after one week, a tested group remembered 56% against 42% for a re-reading group, d = 0.83. Karpicke, Blunt and Smith 2016, 88 children of mean age 10: robust effects in three experiments without feedback. | Re-reading wins at five minutes and feels better, so children prefer it. For maths procedures the effect is not robust (Murray 2025: the interval crosses zero) | After a story, "tell me three things that happened" with the book shut. After a page of sums, cover it and ask one | Every question is a retrieval with no answer showing; a new item is taught, then asked a few questions later rather than right away |
| Interleaving | Mix kinds of problems so the child has to choose the method, not just run it | Rohrer, Dedrick, Hartwig and Cheung 2020, Journal of Educational Psychology, 787 seventh-graders in 54 classes, one month later: 61% interleaved against 38% blocked, d = 0.83 [0.68, 0.97]. Taylor and Rohrer 2010, fourth-graders: 77% against 38% the next day. | Practice looks worse while interleaving; the worksheets took longer and came after blocked teaching | Shuffle the flashcards or mix adding and taking away on one page; expect more mistakes and do not panic | A deck mixes due reviews, learning items and new items across a strand; about one round in four reviews a passed rung |
| Feedback and mastery | Say right or wrong straight away with a one-line why, and keep going until it is solid | Wisniewski, Zierer and Hattie 2020, 435 studies, d = 0.48, higher for feedback with more information. Shute 2008, Review of Educational Research: immediate feedback for hard tasks and weaker learners; on the task, never the learner. Kulik, Kulik and Bangert-Drowns 1990: mastery learning about 0.5 SD, mostly on locally made tests. | Big averages hide wide variation; "about 0.5" needs a check in the primary paper | "Not quite, it ends in -ed, try again" beats both silence and the whole answer | A miss shows the answer with a one-line why and comes back three questions later; a rung is passed only when 80% of what was seen is solid |
| Worked examples first | For younger children, show how before asking them to work it out | Sinha and Kapur 2021, Review of Educational Research, 53 studies: letting learners struggle first wins overall (g = 0.36), but for second to fifth graders instruction-first wins | The age split is from a moderator analysis, not a trial designed to test it | Do one together out loud, then hand over the next one | New items are taught first with a Teach moment, then rehearsed; nothing new is ever on a timer |
| About 85% right | Practise where most answers are right and a few are not | Wilson, Shenhav, Straccia and Cohen 2019, Nature Communications: for simple learners on two-choice tasks, learning is fastest at about 15.87% errors. Math Garden picks items at a 75% chance of success (Hofman et al. 2018) because failing half the time drives children off | The 85% result is from mathematical learners on yes-or-no tasks, by the authors' own caveat | Pick the book or page where your child gets most of it; a book with more than one hard word in ten is for reading together | A deck is about 85% known items, due reviews first, at most two new items a deck. Two misses in the last five and no more new items this round |
Primary links, one each: Cepeda 2008 (https://pubmed.ncbi.nlm.nih.gov/19076480/); Sobel 2011 (https://onlinelibrary.wiley.com/doi/abs/10.1002/acp.1747); Murray 2025 (https://link.springer.com/article/10.1007/s10648-025-10035-1); Roediger and Karpicke 2006 (https://psychology.ecu.edu/wp-content/pv-uploads/sites/216/2019/03/Roediger-Karpicke-2006.pdf); Karpicke, Blunt and Smith 2016 (https://www.frontiersin.org/articles/10.3389/fpsyg.2016.00350/full); Rohrer 2020 (https://ies.ed.gov/ncee/wwc/Study/88770); Taylor and Rohrer 2010 (https://digitalcommons.usf.edu/psy_facpub/1760/); Wisniewski 2020 (https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2019.03087/full); Shute 2008 (https://myweb.fsu.edu/vshute/pdf/shute%202008_b.pdf); Sinha and Kapur 2021 (https://doi.org/10.3102/00346543211019105); Wilson 2019 (https://pmc.ncbi.nlm.nih.gov/articles/PMC6831579); Hofman 2018 (https://www.mdpi.com/2079-3200/6/1/14/html).
Two more that are about you, not the app
| What | Key study | Number | Two minutes at home |
|---|---|---|---|
| Listening to your child read | Sénéchal and Young 2008, Review of Educational Research, 16 interventions, 1,340 families, K to grade 3 (https://carleton.ca/cllr/parents-and-educators/parent-involvement-and-learning-to-read-meta-analysis/) | Parents reading to children d = 0.18 (not significant); parents listening to children read d = 0.52; parents teaching a specific skill d = 1.15 | Hear one page. If they stop at a word, cover the ending and ask for the first sound |
| Systematic phonics early | Ehri, Nunes, Stahl and Willows 2001, Review of Educational Research, 66 comparisons (https://journals.sagepub.com/doi/10.3102/00346543071003393) | d = 0.41 overall; 0.55 when begun in kindergarten or first grade, 0.27 after; Camilli 2003 reanalysis 0.24 | Sound out one new word together, left to right, before guessing from the picture |
No app can do either of these. Peabody cannot hear your child read; it does not record sound, so it cannot measure fluency and does not try.
What does not hold up
| Claim | What the evidence says | What to do instead |
|---|---|---|
| Learning styles (visual, auditory, kinaesthetic) | Matching teaching to a child's stated style has not held up when tested; children learn the same material in the same ways. More in the learning styles myth | Adapt to what the child knows and misses, not to a label |
| Bloom's 2-sigma (one-to-one tutoring gives two standard deviations) | Bloom 1984 rested on two dissertations with about 20 students per cell and narrow tests. Nickow, Oreopoulos and Quan 2020, 96 randomised trials (https://www.nber.org/papers/w27476): pooled 0.37 SD, and none reached 2 sigma. The 2024 version reports 0.29 over 89 trials | Tutoring helps; expect a third of a standard deviation, more from teachers than volunteers, in school hours |
| Brain training | Getting better at the trained puzzle is reliable; carry-over to reading or maths is not | Practise the actual thing: the words, the facts, the books |
| Cramming | Roediger and Karpicke 2006: at five minutes re-reading wins 81% to 75%; at a week it loses 42% to 56%. Cepeda 2008: the shortest gaps are the worst. Bjork 1994: how well practice goes is a poor guide to what sticks | Little and often; two minutes on three days beats six minutes once |
| "Our users grew 0.36 SD" | Vendor numbers for the slice who "used it as recommended" (the 9% in Khan Academy's 2024 claim) are selection, not effect; the randomised trial of the same product found 0.22 overall and the AI chat layer added nothing measurable (Oreopoulos and Low 2026, https://edworkingpapers.com/sites/default/files/ai26-1551.pdf) | Ask for the whole enrolled sample and an independent trial. Peabody has no trial and says so |
How to read an effect size
An effect size (d or g) is the gap between two groups measured in standard deviations of the test. A d of 1.0 means the average child in one group scored where the top 16% of the other group did.
| d or g | Kraft 2020's scale for scaled interventions on standardised tests | Cohen's old labels, which education research should not use |
|---|---|---|
| Under 0.05 | Small | Nothing |
| 0.05 to 0.20 | Medium | Nothing |
| Over 0.20 | Large | "Small" starts at 0.2 |
Three checks before trusting a number: who made the test (narrow, author-made tests gave 0.84 for tutoring where standardised tests gave 0.27, per Cohen, Kulik and Kulik 1982 as read by von Hippel 2024, https://www.educationnext.org/two-sigma-tutoring-separating-science-fiction-from-science-fact/), how many children and classes (Rohrer 2020 is one district), and whether the test was delayed, since Taylor and Rohrer's fourth-graders looked worse during practice and better the next day.
Limits of this sheet
- Most of the lab studies are on adults or students over ten. The child studies (Sobel, Karpicke, Taylor and Rohrer) are small, and the developmental trajectory of these effects is poorly mapped (Fazio and Marsh 2019).
- The testing effect is solid for facts and words and weak for maths procedures (Murray 2025).
- Effect sizes from meta-analyses hide wide variation and depend on who made the test.
- Peabody has not been tested in a trial. The right-hand column says what it does, not what it achieves.
Try a story from the free library a level up and a level down and see which one your child reads more smoothly; that is a better reading check than any chart.
Learn more elsewhere
- The Learning Scientists: free posters and bookmarks on six study strategies, from cognitive psychologists.
- Deans for Impact: The Science of Learning: how students learn, in a short free summary (second edition, 2026).
- Dunlosky: strengthening the student toolbox: ten study habits ranked; practice tests and spacing win, highlighting and rereading don't.
- RetrievalPractice.org guides: short free guides to retrieval, spacing and interleaving by cognitive scientist Pooja Agarwal.
For about seventy more, checked and sorted by who they're for, see where to learn about AI.
Questions parents ask
How much practice is enough?
The studies say little and often. Sobel's fifth-graders learnt more from two short sessions a week apart than from one long one. Peabody calls about 80 first-try right answers "today's practice done", and nothing locks after it.
My child gets more wrong when I mix things up. Am I doing harm?
No. Taylor and Rohrer 2010 found fourth-graders made more mistakes during mixed practice and scored 77% against 38% the next day. The mistakes during practice are the work.
Should I correct every error straight away?
For a young child on something hard, yes, with a one-line reason and a second try (Shute 2008). Skip the lecture and the whole answer. Peabody shows the answer with a short why and asks again three questions later.
Does Peabody use these strategies?
The six rows map to fixed rules in its code: a 1, 3, 7, 21-day review ladder, retrieval on every question, mixed decks, a one-line why after a miss, Teach first for new items, and decks that are about 85% known. Those are design facts from the code, not outcome claims.
Is Peabody free?
Yes, free through the end of 2026, with no account needed and no ads.
Checked against how Peabody works on 18 October 2026. Written with help from AI. Peabody is a reading app, not a reading assessment: if you're worried about your child's reading, talk to their teacher.