Getting good at a skill: deliberate practice and feedback

Think of something you've done almost every day for years: typing, driving, cooking, writing emails. If you type, you've typed for thousands of hours. Are you much faster or more accurate than you were after the first year? For most people, on most skills, the honest answer is "not really". You reached a level that was good enough, then stopped improving while the hours kept piling up.

That plateau is the problem this lesson solves. There's a specific kind of practice that keeps improvement going. It's quite different from doing the thing a lot, and you can build a session of it for any skill in about ten minutes. There's also an unresolved argument among researchers about how far it can take you. You'll get both: the method, and a fair account of what the evidence does and doesn't show.

What deliberate practice actually is

The term comes from a 1993 paper by Anders Ericsson, Ralf Krampe and Clemens Tesch-Römer. ¹ Their main study took 30 violin students at a Berlin music academy and sorted them into three groups. The best had been nominated by their professors as having the potential for careers as international soloists. A "good" group was matched to them by age and sex. The third group came from the academy's music-education department and were headed for careers as teachers. Each student estimated how many hours a week they had practised alone at each age since they started, and kept a diary for one week.

But the hours weren't the paper's main idea. Its main idea was that the hours only counted when the activity had a particular structure. Pull the paper's definition into five parts, and deliberate practice is activity that is:

• Designed to improve, not to perform. It works on a weakness. Not "practise the concerto" but "get the shift in bar 34 clean at tempo".
• At the edge of current ability. The teacher sets tasks the student can't yet do reliably, and moves on to harder ones once the current ones are mastered.
• Fed back immediately. You know, right away, whether the attempt worked and what to change.
• Designed by someone who knows the domain. A teacher who can diagnose the error and prescribe the drill.
• Effortful, and not enjoyable in itself. You can only keep it up for a limited time each day. The two best groups' diaries showed about three and a half hours a day of solitary practice, and the paper found no sign that more than about four hours a day helped. ¹

Ericsson kept this separate from two other things that also fill a musician's day. Work is doing it for an external reward: public performance, competitions, paid services. Play is doing it because it's enjoyable, with no explicit goal. Both add hours. Neither is designed to change what you can do. ¹

In Peak (2016), Ericsson and Robert Pool drew a three-step ladder. ² Naive practice is repetition: doing the thing and hoping. Purposeful practice adds specific goals, focus, feedback, and working outside your comfort zone, and you can do all of that on your own. Deliberate practice adds the last ingredient: a field where expert performance is well understood and a teacher can prescribe drills known to work. Music, chess, gymnastics and ballet qualify. Most skills you want don't have that body of proven methods, so purposeful practice is what you'll actually build. ² Same ingredients; you supply the design.

One caution. The controlled trials elsewhere in this course are on retrieval, spacing and interleaving. The case for designing your own purposeful practice is Ericsson's argument plus the general finding that goal-directed practice with feedback works. Nobody has run a trial of adults coaching themselves with a metronome and a phone. It's the best-supported method available, not a proven one.

Why it works: closing the loop

Ambrose and colleagues, in How Learning Works (2010), state the same idea as one of their seven principles: goal-directed practice coupled with targeted feedback is what moves a skill. ³ Their principle on mastery says why the practice has to be broken down. Mastery is component skills, plus integrating them, plus knowing when to use them, and each of those fails separately. ³ A guitarist who can play a C and an F but can't switch between them at tempo has the components and not the integration. That's why the drill below isolates a single transition, and why the session ends by putting it back into a song.

The mechanism goes like this. Every attempt at a skill is an experiment: you intended one thing and something happened. Feedback tells you the gap. Without it, you keep the gap. And because repetition strengthens whatever you repeat, practising without feedback is worse than nothing. It entrenches the version you already have, errors included. Twenty years of typing with two fingers produces a very fast two-finger typist.

Ericsson's answer to what is being built is mental representations: rich, domain-specific patterns in long-term memory that let an expert see a situation as a whole, predict what comes next, and notice their own mistakes. ² This is lesson 2's chunking, applied to skill. A chess master sees a position as a handful of familiar structures where a novice sees thirty-two pieces. Chase and Simon showed in 1973 that this, not a better memory in general, is what separates them. ¹⁵ Deliberate practice builds these representations because it forces your attention onto exactly the pattern you can't yet see, and feedback tells you whether you saw it right. Once you have the representation you can monitor yourself, which is why experts eventually need a teacher less. Until then, someone else has to be the feedback.

"At the edge" matters for the same reason lesson 1 gave for desirable difficulties. Bjork and Bjork (2011) were explicit that a difficulty is only desirable if the learner can meet it. ⁴ Working at the edge means slightly beyond current ability: hard enough that you fail sometimes, close enough that feedback can tell you why. Far beyond it, you fail everywhere and the feedback has nothing to attach to.

Worked example: two guitarists

Two people each pick up a guitar and play an hour a day for a year. That's roughly 365 hours each.

The first plays songs he knows. He warms up with the same three, tries a new one when he's bored, and stops when it sounds rough. He enjoys it, and can play about twenty songs by the end of the year. Run the list. Designed to improve? No: he plays what already works. At the edge? Rarely. Feedback? His own ear, in the moment, no record. Effortful? Mildly. This is play with a bit of naive practice. It produces a competent, stalled amateur, which is fine if that's what he wanted.

The second starts each session by naming one transition she can't yet make cleanly, say the change from a C to an F barre chord. She sets a metronome at a speed where she can make it about seven times in ten, and drills that change, and only that change, for ten minutes. She records the last minute on her phone and listens back: is the F buzzing? Which string? She moves the metronome up 4 beats per minute when she's clean nine times in ten. Then she does the same for one more weak spot. The final fifteen minutes she plays songs, partly for pleasure and partly to check whether the drilled change survives inside real music.

Same instrument, same hours. The difference is entirely in the structure.

A harder example: twenty years of experience

Take one professional: a commercial lawyer, twenty years in, who drafts leases. Is she better at it than she was at year five? Not more senior, not more confident: does a lease she drafts today contain fewer clauses that later cause a dispute? Almost nobody has that number, and that's the first clue. Her feedback loop is years long. A lease drafted in 2012 may first be tested in court in 2021, and by then nobody will connect the outcome to the drafting decision. The firm's review scores measure billing and client retention, not drafting accuracy.

Treat that number gently. What it means is disputed. Ericsson's side argues that none of the professions studies measured deliberate practice as defined; when Ericsson and Harwell reanalysed the meta-analysis with the original criteria, no professions study qualified at all. ¹⁰ Macnamara's side replies that they used the best measures the studies contained, and that a definition which excludes the available evidence makes the theory hard to test. ⁸ What both sides agree on is the point that matters for you: ordinary working hours are not deliberate practice. Run the checklist on the lawyer:

• Designed to improve? No. The work arrives as it arrives; nobody is prescribing drills.
• Edge of ability? No. Most of a working day is what you can already do. That's what being competent means.
• Immediate feedback? Absent or delayed by years, as above. The same holds elsewhere: a clinician's diagnosis may never be checked against the outcome; a salesperson hears "no" without learning why.
• Designed by someone who knows the domain? Nobody is designing it.

So she has accumulated something like 40,000 hours, almost none of which meet the definition, and they've done what work does: entrenched what she was already doing at year five. Notice what this doesn't show. It doesn't show that professionals can't improve; it shows that experience is not practice. A lawyer who keeps a file of her disputed clauses, reads the judgments, and asks a better drafter where she went wrong is doing something the meta-analysis mostly couldn't measure.

How much does practice explain? The debate

Here the answer is "it isn't settled", and you should know both sides well enough to explain each to someone who holds the other.

One term needs a sentence before the numbers. "Variance explained" is not the share of your skill that comes from practice. It's the share of the differences between people in a study that lines up with differences in how much they practised. If everyone in a sample has practised roughly the same amount, practice can't explain their differences even if it built all of their skill. That's one reason a figure from a narrow, elite sample is hard to read.

The Macnamara–Hambrick position. In 2014, Brooke Macnamara, David Hambrick and Frederick Oswald pooled 88 studies (157 effect sizes, 11,135 participants) that had measured both accumulated practice and performance. ⁵ Practice explained 26% of the variance in performance in games, 21% in music, 18% in sports, 4% in education, and under 1% in professions.

Hambrick and colleagues, in a separate 2014 paper correcting for measurement error, put the figures at 34% of the reliable variance in chess and about 30% in music. ⁶ Two further findings anchor their view. In one chess data set, the 16 masters had accumulated anywhere from 832 to 24,284 hours of solitary study, and five of them had less than the average of the expert group one level below. ⁶ And a study of 10,500 Swedish twins found that how much people practise music is itself substantially heritable (40–70%), and that within identical-twin pairs, the twin who practised more scored no better on tests of pitch, melody and rhythm discrimination. ¹¹ (Note what it measured: discrimination tests in a general population, not expert performance. Whether it bears on the violinists is itself disputed.) Their conclusion: practice is important, "but not as important as has been argued". It leaves most of the difference between people unexplained, and the rest is likely to include starting age, working-memory capacity, and genetic differences. A fifth to a third of the variance is a large effect for a single modifiable factor; their claim is that the 1993 framing went further than its data allowed.

The Ericsson position. The 1993 paper had argued that "many characteristics once believed to reflect innate talent are actually the result of intense practice extended for a minimum of 10 years", and that the search for heritable characteristics predicting expert performance had been "surprisingly unsuccessful", with height in some sports the clearest exception. ¹ Ericsson's reply to the meta-analyses, in 2016 and again with Kyle Harwell in 2019, was that they had counted the wrong thing. ^{7 10} The pooled studies had summed up virtually any activity in the domain: team practice, group activities, hours in lectures and seminars, play and competition, even watching the sport on television. By the 1993 definition most of those hours aren't deliberate practice, so the meta-analysis measured a mixture and reported that the mixture explains little. On this view, 20–30% is a floor produced by counting activities that don't meet the definition, not an estimate of what deliberate practice can do; and the studies that could show more are rare because practice of that kind is rare.

The rejoinder. Macnamara, Hambrick and Moreau (2016) answered that Ericsson had himself defined deliberate practice as activities designed "by a teacher or the performers themselves", that his own studies hadn't asked participants to count only teacher-designed practice, and that restricting their sports meta-analysis to solitary practice or to objective performance measures moved the figure only from 18% to about 20–22%. ⁸ For a theory to be falsifiable, they said, its definitions must be used consistently; and while longitudinal studies would help, the task was to weigh the evidence that exists.

The replication. In 2019 Macnamara and Megha Maitra ran a preregistered replication of the violin study with 39 violinists in Cleveland: 13 best and 13 good from the Cleveland Institute of Music, and 13 less accomplished from the music department of Case Western Reserve University. ⁹

Both groups had passed 10,000 hours by 20. Practice explained 26% of the variance in which skill group a violinist belonged to, against the 48% that Ericsson's own comparison had explained in 1993. The authors say the replication doesn't show practice is unimportant; 26% "is not an inconsequential amount". They read it as showing that among elite performers, the amount of practice can't account for who rises highest. Ericsson died in 2020 and, as far as we can find, published no reply. His standing objection, that counting remembered hours doesn't measure the quality of practice, applies to Cleveland as much as to Berlin; and the replication can't itself say whether its groups, from two schools in one city, were as far apart in ability as the Berlin groups. Neither reading has been tested.

Where that leaves you. Three things nobody in this dispute denies. Deliberate practice, structured as above, is the best-supported method anyone has for improving a skill. Accumulated practice, however counted, is a substantial predictor of performance in music, games and sport. And the replication didn't reproduce the size of the original violin effect. What's contested is everything else: whether the 20–30% figures fairly estimate deliberate practice or reflect counting the wrong hours; what the professions figure means; whether the Cleveland and Berlin groups were comparable; and what fills the unexplained share. What would settle it is a preregistered longitudinal study measuring practice quality prospectively, as it happens, rather than asking adults to remember childhood hours. Both sides say so. That study hasn't been done. Until it is, anyone who tells you the answer is known, in either direction, is telling you their side.

What people get wrong

"Experience equals expertise." True only in the easy sense that you can't be expert without experience. The lawyer in the second example had 40,000 hours; the checklist showed almost none of them met the definition. When you assess someone's skill, including your own, ask what their practice looked like, not how long they've been at it.

"Talent doesn't matter." This is close to what the 1993 paper claimed: that many characteristics thought to be innate are products of practice, and that evidence for heritable predictors of expert performance was thin outside body size. ¹ But even there, Ericsson didn't claim that everyone can reach the top, only that the evidence for innate limits was weak. Macnamara and Hambrick read the replication and the twin study as evidence for other factors; Ericsson's standing reply is that those studies counted hours, or tested general abilities, rather than measuring expert practice.

"Only talent matters." This overstates Macnamara and Hambrick, who report practice explaining a fifth to a third of variance in several domains and call it important. Nobody in the debate thinks practice is decorative, and Hambrick's side doesn't talk about "talent" at all; it names measurable factors such as starting age, working memory, and general cognitive ability.

"Harder is always better." The edge is a narrow band. Bjork and Bjork's condition applies: the difficulty must be one you can meet, and it must engage the thing you're trying to learn. ⁴ Attempting things far beyond your level, without feedback, isn't deliberate practice. It's failing, repeatedly, in a way that teaches you nothing specific.

"I'll know when I'm improving." Lesson 1 again: your felt sense tracks fluency, not learning. Soderstrom and Bjork's review is blunt about it: what you can do during practice is an unreliable guide to what you've learned. ¹³ Dunlosky and Rawson found that students who trusted their own judgement of mastery stopped practising early and retained less. ¹⁴ The guitarist's rule (move on at nine clean in ten) exists because her ear alone would move on too early. Build the measure in.

Practice

Before the exercise, close the page and answer on a blank sheet: the five parts of Ericsson's definition; the rough 2014 figures for games, music and professions; and, in one sentence each, what the 2019 replication found and what it didn't show. Then check.

Connections

This lesson is lesson 1's desirable difficulty applied to skills: the edge of ability is where the difficulty is high enough to matter and low enough to be met, and feedback is what makes it surmountable rather than just hard. Ericsson's mental representations are lesson 2's chunks, built by doing rather than reading. The feedback loop is lesson 3's retrieval-with-feedback in another form: attempt, find out, adjust. And the metronome rule defends against lesson 1's fluency illusion by replacing "that felt clean" with a count.

Lesson 8 takes the session you just designed and fits it into a week alongside spacing, interleaving and retrieval, in the hours you actually have.

Go deeper

• Ericsson & Pool, Peak (2016), chapters 1–4. The clearest statement of naive, purposeful and deliberate practice, and of mental representations as the thing being built. The author is one side of the debate above; read it as his case.
• Macnamara, Hambrick & Oswald, "Deliberate Practice and Performance in Music, Games, Sports, Education, and Professions", Psychological Science 25 (2014), read as a pair with Ericsson & Harwell, "Deliberate Practice and Proposed Limits on the Effects of Practice", Frontiers in Psychology 10 (2019, open access). The meta-analysis and the fullest reply to it. The authors are the two sides; the abstract of the first carries the figures and the second explains why Ericsson thought they measured the wrong thing.
• Macnamara & Maitra, "The role of deliberate practice in expert performance: revisiting Ericsson, Krampe & Tesch-Römer (1993)", Royal Society Open Science 6 (2019). Open access, preregistered, and a model of what a replication report looks like, written by one side of the debate. Worth reading for the method as much as the result.
• "The Expert on Expertise" (International Center for Clinical Excellence, 2020): a 27-minute interview with Ericsson, recorded shortly before his death, reviewing the research in his own words. Too long to embed in the lesson; worth the time if the debate above caught you.
• Ambrose et al., How Learning Works (2010), chapters 4–5. Mastery as components plus integration, and the goal-directed-practice-plus-feedback principle with the classroom evidence behind it; the most practical of the four.

Sources

1. Ericsson, K. A., Krampe, R. T. & Tesch-Römer, C., "The role of deliberate practice in the acquisition of expert performance", Psychological Review 100, 363–406 (1993). Study 1: 30 Berlin violin students in three groups (best, nominated by professors as potential international soloists; good; music-education students) plus 10 professional orchestral violinists; Study 2: 12 expert and 12 amateur pianists. Retrospective weekly estimates plus a one-week diary. About 7,410 hours of solitary practice by 18 for the best group, about 5,300 for the good, about 3,420 for future teachers; over 10,000 by 20 for the best. Diaries: about 3.5 hours a day for the two best groups; no benefit found beyond about 4 hours a day. The definition of deliberate practice and its distinction from work and play. Innate talent: "many characteristics once believed to reflect innate talent are actually the result of intense practice"; the search for heritable predictors "surprisingly unsuccessful", height in sport the exception. Endorses Simon and Chase's "10-year rule".
2. Ericsson, K. A. & Pool, R., Peak: Secrets from the New Science of Expertise (Houghton Mifflin Harcourt, 2016), chs. 1–4. Naive, purposeful and deliberate practice; mental representations as mechanism; rejection of the "10,000-hour rule": an average, half the best group below it at 20, and the students still "had a long way to go".
3. Ambrose, S. A., Bridges, M. W., DiPietro, M., Lovett, M. C. & Norman, M. K., How Learning Works: Seven Research-Based Principles for Smart Teaching (Jossey-Bass, 2010), chs. 4–5. Mastery as component skills plus integration plus knowing when to apply; goal-directed practice coupled with targeted feedback.
4. Bjork, E. L. & Bjork, R. A., "Making things hard on yourself, but in a good way", in Gernsbacher et al. (eds), Psychology and the Real World (Worth, 2011), 56–64. A difficulty is desirable only if the learner has the background to meet it.
5. Macnamara, B. N., Hambrick, D. Z. & Oswald, F. L., "Deliberate Practice and Performance in Music, Games, Sports, Education, and Professions: A Meta-Analysis", Psychological Science 25, 1608–1618 (2014). 88 studies, 157 effect sizes, 11,135 participants. Variance explained: 26% games, 21% music, 18% sports, 4% education, under 1% professions (7 effect sizes, 321 participants, in programming, military piloting, soccer refereeing and insurance sales; not significant). "Important, but not as important as has been argued."
6. Hambrick, D. Z., Oswald, F. L., Altmann, E. M., Meinz, E. J., Gobet, F. & Campitelli, G., "Deliberate practice: Is that all it takes to become an expert?", Intelligence 45, 34–45 (2014). 34% of reliable variance in chess, about 30% in music. Range of solitary study among 16 chess masters (Gobet & Campitelli 2007 data): 832 to 24,284 hours; 5 of 16 below the expert-group mean.
7. Ericsson, K. A., "Summing Up Hours of Any Type of Practice Versus Identifying Optimal Practice Activities: Commentary on Macnamara, Moreau, & Hambrick (2016)", Perspectives on Psychological Science 11(3), 351–354 (2016). The reply to the sports meta-analysis (Macnamara, Moreau & Hambrick, same journal, 333–350): the studies counted virtually any sport-specific activity, including group activities, watching games on television, play and competitions.
8. Macnamara, B. N., Hambrick, D. Z. & Moreau, D., "How Important Is Deliberate Practice? Reply to Ericsson (2016)", Perspectives on Psychological Science 11(3), 355–358 (2016). Ericsson's own definition includes practice designed "by a teacher or the performers themselves"; his own studies did not restrict estimates to teacher-designed practice; restricting to solitary practice or objective measures moves the sports figure from 18% to 20–22%; "for a theory to be falsifiable, definitions must be used consistently".
9. Macnamara, B. N. & Maitra, M., "The role of deliberate practice in expert performance: revisiting Ericsson, Krampe & Tesch-Römer (1993)", Royal Society Open Science 6, 190327 (2019). Preregistered replication; 39 violinists (13 per group; best and good from the Cleveland Institute of Music, less accomplished from Case Western Reserve University). Solitary practice by 18: best 8,224, good 9,844, less accomplished 4,558; best-versus-good difference not significant (p = 0.36). Practice explained 26% of the variance in skill-group membership against the 48% explained by the 1993 study's own best-plus-good versus less-accomplished comparison; "not an inconsequential amount".
10. Ericsson, K. A. & Harwell, K. W., "Deliberate Practice and Proposed Limits on the Effects of Practice on the Acquisition of Expert Performance: Why the Original Definition Matters and Recommendations for Future Research", Frontiers in Psychology 10, 2396 (2019). The reply to the 2014 meta-analysis: it aggregated lectures and seminars, team practice, television watching and unspecified practice; reanalysed with the 1993 criteria, no professions study qualified (k = 0). Does not discuss the 2019 replication.
11. Mosing, M. A., Madison, G., Pedersen, N. L., Kuja-Halkola, R. & Ullén, F., "Practice Does Not Make Perfect: No Causal Effect of Music Practice on Music Ability", Psychological Science 25, 1795–1803 (2014). 10,500 Swedish twins; practice heritability 40–70%; no within-pair association between practice and pitch, melody or rhythm discrimination in identical twins.
12. Gladwell, M., Outliers: The Story of Success (Little, Brown, 2008), ch. 2. Origin of the "10,000-hour rule" as a popular reading of source 1; the chapter's own argument is about opportunity to accumulate practice.
13. Soderstrom, N. C. & Bjork, R. A., "Learning Versus Performance: An Integrative Review", Perspectives on Psychological Science 10, 176–199 (2015). Performance during practice is an unreliable index of learning.
14. Dunlosky, J. & Rawson, K. A., "Overconfidence produces underachievement: Inaccurate self evaluations undermine students' learning and retention", Learning and Instruction 22, 271–280 (2012). Overconfident self-judgements led students to stop practising early and retain less.
15. Chase, W. G. & Simon, H. A., "Perception in chess", Cognitive Psychology 4, 55–81 (1973). Masters recall real positions in chunks; their advantage disappears for random positions.