What makes spelling practice actually work

The single most replicated result in learning science is that retrieving something from memory strengthens it far more than looking at it again. In the canonical experiment, students who practised recalling a passage retained about 61% of it a week later, against about 40% for students who spent the same time rereading, and meta-analytic estimates put the general advantage around half a standard deviation, growing as the delay lengthens (Roediger & Karpicke's 2006 experiment and the decade of work after it, reviewed in Karpicke, 2017). The implication for spelling is blunt: staring at a word list trains recognition, which you already have. Producing the spelling from memory, and finding out you can't, is where learning happens.

Across 317 experiments, spreading practice over time reliably beats massing the same amount of it together (Cepeda et al., 2006). The subtler result is that the best gap depends on how long you want to remember: as a rough orientation from the follow-up study, a useful gap is around a fifth of the delay until you need the word, shrinking as a proportion for very long horizons (Cepeda et al., 2008). Gaps can also be too long: performance follows an inverted U, not "more is better".

Two refinements matter for spelling practice specifically. First, successive relearning, which means recalling an item correctly in each of several spaced sessions, produced more than twice the retention of recalling it the same number of times within one session (Rawson & Dunlosky, 2022). Getting a word right three times today is not three units of learning; getting it right on three different days is. Second, the popular belief that review gaps must expand (one day, then three, then a week…) is ahead of its evidence: a meta-analysis found no reliable overall difference between expanding and evenly spaced schedules (Latimier, Peyre & Ramus, 2021). The spacing itself is what does the work.

The counterintuitive core of all this is Robert and Elizabeth Bjork's notion of desirable difficulties: conditions that make practice feel harder (spacing, mixing, testing instead of restudying) depress performance during practice while improving it afterwards (Bjork & Bjork, 2011). Partial forgetting before a retrieval attempt is precisely what makes the attempt productive. This carries a warning for anyone judging a practice method by how it feels: smooth, error-free practice sessions are often the least effective ones, and a method that feels effortful may be working exactly because it is.

Mostly the opposite, provided correction follows. Generating a wrong answer and then seeing the right one produces better memory than error-free study of the same material, apparently because the mismatch draws attention to what needs encoding (Metcalfe, 2017). Even failing before you had any chance to succeed helps: guessing at an answer you don't know, then being shown it, beats spending the same time studying it, an effect that survives excluding lucky guesses (Richland, Kornell & Kao, 2009). The requirement in both cases is corrective feedback that actually arrives; an uncorrected error teaches the error.

One documented exception: in populations with clinically impaired memory, errorless approaches win, because errors get encoded without the ability to suppress them later. That result comes from memory rehabilitation research and has not been shown to extend to, say, dyslexia or ordinary variation in working memory.

The instructional literature has produced a few named spelling routines with dozens of replications behind them. Cover-Copy-Compare (study the correct model, cover it, write it from memory, uncover and compare, correct if wrong) is effective across skills, students and settings, and needs no instructor in the loop (Skinner et al.). Self-correction goes further: having learners locate and fix their own errors against a model beat conventional review both during teaching and at follow-up (error self-correction study, 2007). And when an error occurs, an active response, meaning producing the correct form rather than merely looking at it, outperformed passive exposure across same-day, next-day and two-week tests, in a small but carefully measured comparison (Nulman & Gerber, 1984). The common thread: the learner does the work, the comparison against the correct form is explicit, and the error is repaired rather than just revealed.

Interleaving, which mixes different kinds of items rather than drilling one kind at a time, beats blocking on average (g ≈ 0.42 across 59 studies), with the benefit largest where the mixed items are similar enough to confuse, because mixing forces the learner to notice what distinguishes them (Brunmair & Richter, 2019). Spelling now has a direct test: in a preregistered classroom experiment, third graders who practised spelling rules interleaved made fewer errors than blocked practisers, immediately and at an eight-week follow-up, and explanations that pointed out the contrasts between rules were what helped the gains transfer to new words (interleaved spelling trial, 2025). That is one study, in German, which is a more rule-regular orthography than English, so treat it as a promising result rather than a settled one.

Practice that is too easy teaches little; practice that is too hard teaches little and feels terrible. Where is the useful middle? The most-quoted number is the "85% rule", a derivation that learning is fastest at roughly 85% success. Its fine print is routinely dropped: the result is proved for machine-learning systems on binary classification tasks, not for humans learning to spell (Wilson et al., 2019). More grounded is the shipping precedent: Math Garden, a large adaptive arithmetic-practice system used by thousands of schoolchildren, deliberately serves items at about 75% success, described as challenging but not discouraging (Klinkenberg, Straatemeier & van der Maas, 2011). The literature's honest summary: some challenge is essential, every source that names an exact number also declines to universalise it, and a learner who is succeeding on everything is probably not being stretched.

Yes: this is one of the better-supported claims in literacy education. A meta-analysis of 53 studies covering just over 6,000 K–12 students found formal spelling instruction beat no or unrelated instruction at an effect size of 0.54, that more instruction beat less, and (the striking part) that the benefit transferred to reading, including phonemic awareness and comprehension (Graham & Santangelo, 2014). A later independent synthesis reached the same overall conclusion while emphasising that implementation quality moderates the results (spelling instruction meta-analysis, 2024). Two honest caveats travel with those numbers: the studies are classroom programmes for children, not voluntary adult practice, and pooled effect sizes describe categories of teaching, not any particular product.

How people learn to spell covers the memory machinery all of this practice is feeding. Spelling and digital learning looks at what happens when these principles are implemented in software, and at the claims that outrun the evidence. Or go back to the Learn hub.

• Karpicke (2017). Retrieval-Based Learning: A Decade of Progress. A review of the decade following Roediger & Karpicke's 2006 test-enhanced-learning experiment; the 61%/40% figures below are reported through this review, not by it
• Cepeda et al. (2006). Distributed practice in verbal recall tasks: A review and quantitative synthesis. Psychological Bulletin
• Cepeda et al. (2008). Spacing effects in learning: A temporal ridgeline of optimal retention. Psychological Science
• Latimier, Peyre & Ramus (2021). A meta-analytic review of the benefit of spacing out retrieval practice episodes on retention. Educational Psychology Review
• Rawson & Dunlosky (2022). Successive Relearning: An Underexplored but Potent Technique. Current Directions in Psychological Science
• Bjork & Bjork (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning
• Richland, Kornell & Kao (2009). The pretesting effect: Do unsuccessful retrieval attempts enhance learning?. Journal of Experimental Psychology: Applied
• Metcalfe (2017). Learning from Errors. Annual Review of Psychology
• Skinner et al.. Cover, Copy, and Compare: A Self-Managed Academic Intervention Effective Across Skills, Students, and Settings
• Error self-correction study (2007). Error self-correction and spelling: Improving the spelling accuracy of secondary students with disabilities. Journal of Behavioral Education
• Nulman & Gerber (1984). Improving spelling performance by imitating a child's errors. Journal of Learning Disabilities. Single-case design; six students
• Brunmair & Richter (2019). Similarity matters: A meta-analysis of interleaved learning and its moderators. Psychological Bulletin
• Interleaved spelling trial (2025). Spelling acquisition in children through interleaved practice: the role of instructional guidance. Cognitive Research: Principles and Implications. Preregistered classroom experiment, N = 147, German orthography
• Wilson et al. (2019). The Eighty Five Percent Rule for optimal learning. Nature Communications. Derived for machine-learning systems on binary classification; transfer to human learning is not established
• Klinkenberg, Straatemeier & van der Maas (2011). Computer adaptive practice of Maths ability using a new item response model for on the fly ability and difficulty estimation. Computers & Education
• Graham & Santangelo (2014). Does spelling instruction make students better spellers, readers, and writers? A meta-analytic review. Reading and Writing
• Spelling instruction meta-analysis (2024). Systematic review and meta-analysis of the implementation and effectiveness of spelling instruction and intervention. Psychology in the Schools