TL;DR. The 10,000-hour rule does not exist in the research it came from. In the 1993 violin study by K. Anders Ericsson, Ralf Krampe and Clemens Tesch-Romer, the best students had accumulated an average of 7,410 hours of practice alone by age 18, against 5,301 for good students and 3,420 for future music teachers. “10,000” was roughly where the best group’s average sat around age 20: a group average, not a threshold. Ericsson and a co-author later wrote that the research found “no evidence for a magical number.” The larger studies since then are more useful. A meta-analysis of 88 studies found that deliberate practice explains about 14% of performance differences overall, a figure from a 2018 corrigendum; the “26% in games, less than 1% in professions” set most articles still repeat is out of date. Practice explains most where the task is predictable (23% of variance) and least where it is not (6%). The lesson is not “practice less.” It is that hours are a weak proxy and the design of each hour is what counts. Below: what the original study measured, the corrected numbers, when hours predict skill, and an audit for the quality of your own practice.
Where did the 10,000-hour rule come from?
The rule has two parents, and only one of them was a scientist.
The scientific parent is a 1993 paper in Psychological Review by Ericsson and colleagues at the Max Planck Institute for Human Development in Berlin. Professors at the Music Academy of West Berlin nominated violin students with the potential for careers as international soloists; 10 of the 14 nominated took part as “the best violinists.” Ten “good violinists” from the same department were matched for sex and age, and 10 more came from the music education department, which had lower admission standards. Ten middle-aged violinists from the Berlin Philharmonic and the Radio Symphony Orchestra were also interviewed. Each violinist estimated, year by year, how many hours a week they had practised alone, and the students kept a diary for one week.
The popular parent is Malcolm Gladwell’s 2008 book Outliers, which devoted a chapter to the “10,000-hour rule” and cited the Ericsson paper as its main evidence. In a 2019 article in Frontiers in Psychology, Ericsson and Kyle Harwell quote Gladwell calling ten thousand hours “the magic number for true expertise” and then reject the idea directly. Their research showed that an extended period of training was required for international-level performance, they wrote, but “there was no evidence for a magical number.” They added two points that rarely make it into the retellings. First, Gladwell never used the term deliberate practice; his examples counted public performances and work as practice. Second, Ericsson’s own estimate for winning international piano competitions was around 25,000 hours, not 10,000.
So the rule took a group average from one small study, removed the word “deliberate,” and turned the average into a finish line.
What did the violin study actually measure?
The table below sets out the numbers as the 1993 paper reports them, next to a 2019 replication that is discussed later in this article.
Table 1. Accumulated practice alone by age 18 in the violin studies (verified data)
| Group | Ericsson, Krampe and Tesch-Romer (1993), Berlin | Macnamara and Maitra (2019), Cleveland |
|---|---|---|
| Best violinists (potential international soloists) | 7,410 hours (n = 10) | 8,224 hours (n = 13) |
| Good violinists (same department) | 5,301 hours (n = 10) | 9,844 hours (n = 13) |
| Less accomplished (music education or music department) | 3,420 hours (n = 10) | 4,558 hours (n = 13) |
| Middle-aged professional orchestra violinists | 7,336 hours (n = 10) | Not studied |
| Is “best” above “good”? | Yes, reported as reliably different (p < .05) | No; the good group was higher, difference not significant (p = 0.364) |
Sources: Ericsson, Krampe and Tesch-Romer, Psychological Review (1993), Study 1; Macnamara and Maitra, Royal Society Open Science (2019), section 3.5.3. Gaps between groups are a CEOtudent calculation: in 1993 the best group had 2,109 more hours than the good group and about 2.2 times the hours of the music education group.
Three details in the original study matter more than the totals.
First, the number Ericsson’s team chose to test was hours by age 18, not 10,000. They picked 18 to avoid any confounding influence from activities at the music academy. The “more than 10,000 hours by age 20” figure for the best group comes from reading the curve in the paper’s Figure 9; a 2014 meta-analysis summarised it as more than 10,000 hours for the best violinists at age 20, against about 7,800 for the good group and about 4,600 for the least accomplished.
Second, what the best performers did each day was moderate and structured. In the diary week, the two top groups practised alone for an average of 24.3 hours a week, about 3.5 hours a day, every day including weekends, while the future teachers practised 9.3 hours a week. The top groups also napped more in the afternoon and rated sleep as highly relevant to improving their playing. The same paper notes that training studies show essentially no benefit from practice durations beyond 4 hours a day. At the top groups’ diary pace, reaching 10,000 hours would take about 7.9 years (CEOtudent calculation: 10,000 / 24.3 hours a week). The best violinists kept a sustainable daily dose for a decade rather than grinding beyond recovery.
Third, the definition. Ericsson and colleagues defined deliberate practice as activity “specially designed to improve the current level of performance,” effortful and “not inherently enjoyable.” They separated it from work (performing for pay or reward) and play (activity done for enjoyment). Of the 30 student violinists, 27 rated practice alone as the single most relevant activity for improving. Playing alone for fun, which looks identical to an observer, was rated much less relevant. The paper also states the boundary condition that the popular rule drops entirely: “In the absence of adequate feedback, efficient learning is impossible.”
What did the large studies find when they pooled the evidence?
The violin study had 40 participants. In 2014 Brooke Macnamara, David Hambrick and Frederick Oswald published a meta-analysis in Psychological Science that pooled every study they could find that measured accumulated deliberate practice and performance in the same people: 88 studies, 111 independent samples, 157 effect sizes and 11,135 participants across music, games, sports, education and professions.
Their headline numbers have been quoted thousands of times: deliberate practice explained 26% of the variance in performance for games, 21% for music, 18% for sports, 4% for education and less than 1% for professions, and 12% overall.
Those numbers are out of date. In 2018 Psychological Science published a corrigendum (volume 29, issue 7, pages 1202-1204). The authors explained that they had applied a sample-size adjustment for dependent effect sizes, a method from Cheung and Chan, incorrectly: to each individual effect size rather than to the average. They stated that the correction did not change the substance of their conclusions. But it did change the figures, and the corrected set is the one that should be quoted.
Table 2. How much of performance deliberate practice explains, original vs corrected figures (verified data)
| Result | Originally reported (2014) | Corrected (2018 corrigendum) |
|---|---|---|
| Average correlation, practice and performance | r = .35 | r = .38 |
| Overall variance explained | 12% (88% unexplained) | 14% (86% unexplained), 95% CI 11% to 18% |
| Games | 26% | 24% |
| Music | 21% | 23% |
| Sports | 18% | 20% |
| Education | 4% | 5% |
| Professions | Less than 1% (r = .05, p = .62) | 1% (r = .09, p = .377, not statistically significant) |
| High-predictability activities (example: running) | 24% | 23% |
| Moderate-predictability activities (example: fencing) | 12% | 14% |
| Low-predictability activities (example: handling an aviation emergency) | 4% | 6% |
| Studies that measured practice with a log rather than recall | 5% | 4% |
| Studies that measured practice with a questionnaire | 12% | 15% |
Source: Macnamara, Hambrick and Oswald, Psychological Science 25(8), 1608-1618 (2014); Corrigendum, Psychological Science 29(7), 1202-1204 (2018), Table 1.
Four readings of this table matter for anyone planning their own development.
Practice matters. Explaining 23% of the differences between musicians is a large effect by behavioural-science standards.
Practice is not the whole story. Even in music and games, three quarters of the variance comes from somewhere else: starting age, quality of teaching, working memory, personality, the specific type of practice, and measurement error. A 2014 reanalysis in the journal Intelligence by Hambrick, Gobet, Campitelli and colleagues reached the same conclusion with different methods. After correcting for measurement error, deliberate practice explained 34% of the reliable variance in chess performance and about 29% in music, leaving most of it unexplained.
Predictability is the strongest signal in the table. Practice explains about 3.8 times as much variance in high-predictability activities as in low-predictability ones (CEOtudent calculation: 23% / 6%). The meta-analysis defined predictability as the degree to which the task environment can change while the performer is planning and executing an action, and the range of possible actions. That is the variable that matters most for knowledge work, and the next section builds on it.
How you count the hours changes the answer. Studies that tracked practice in a log as it happened found a weaker link (4%) than studies that relied on people’s memory of years of practice (15%). Remembered hours are generous hours.
Did the original violin study replicate?
Not cleanly. In 2019 Macnamara and Megha Maitra repeated the study in Royal Society Open Science with three changes: a double-blind procedure (interviewers did not know each violinist’s group, and violinists were not told there were groups), analyses that included all three groups at once, and a separate measure of practice designed by a teacher.
Faculty at the Cleveland Institute of Music nominated 24 students with the potential for careers as international soloists; 13 took part. Good violinists came from the same department and less accomplished players from the music department of Case Western Reserve University, 39 violinists in all.
The results, shown in Table 1, broke the neat staircase. Accumulated practice still differed by group overall. But the best violinists had not practised more than the good ones; the good group’s average was 1,620 hours higher, though the difference was not significant (CEOtudent calculation: 9,844 – 8,224). By age 20, both the best and good groups had passed 10,000 hours on average. Practice explained 26% of performance variance in the replication, against 48% in the original. The authors note that this is close to the corrected meta-analytic estimate for music (23%) and that 26% “is not an inconsequential amount,” but it does not support the original claim that performance can “largely” be accounted for by practice.
They also point to a detail of the original data: according to their account, Ericsson reported in 2014 that the 95% confidence interval around the original best group’s practice to age 18 ran from 2,894 to 11,926 hours. The good group’s average of 5,301 sat inside it.
What do the defenders of deliberate practice say?
Ericsson did not accept these conclusions, and his argument deserves a fair hearing because it contains the most useful idea in the whole debate.
In a 2016 commentary in Perspectives on Psychological Science, he argued that the sports meta-analysis by Macnamara, David Moreau and Hambrick used “deliberate practice” for almost any sport-specific activity, including group sessions, watching games on television, play and competition. Summing every hour of every type of practice, he wrote, assumes that all practice activities have an equal effect, which is inconsistent with the evidence.
In 2019 Ericsson and Harwell went further. They applied three criteria for reproducible performance and for purposeful or deliberate practice, including individualised practice under the supervision of a teacher or coach, to the effect sizes in the 2014 meta-analysis and kept only those that met them. Fourteen effects survived: five from games, three from music, five from sports and one from education (a spelling bee study). None from professions met the criteria. In that subset, the correlation was r = .54, about 29% of variance. After adjusting for the imperfect reliability of practice estimates and performance measures, they estimated 61%.
Both camps report real analyses, and the disagreement narrows to one question: what counts as practice. The critics count the hours people log and find a modest link. Ericsson counts a narrow kind of practice, designed by a teacher and aimed at specific weaknesses with immediate feedback, and finds a much stronger link in a much smaller set of studies. For a working adult the conclusion is the same from both sides: the total matters less than the kind of hour.
There is one more finding that should worry anyone who builds a career in knowledge work. In the corrected meta-analysis, practice in professions explained 1% of performance, a result that was not statistically significant. In Ericsson and Harwell’s stricter reanalysis, no study of professions qualified at all. The research does not show that practice is useless at work. It shows that almost no one has measured well-designed practice in the professions, probably because most jobs offer so little of it.
How variable are the hours to expertise?
Very, which is the other reason a single number misleads. Fernand Gobet and Guillermo Campitelli surveyed 104 Argentinian chess players, from weak amateurs to grandmasters, and published the results in Developmental Psychology in 2007. For the 34 players whose path to master level (2200 Elo) could be dated, the mean total practice at that point was 11,053 hours, close to the old “10,000 hours” estimate. The spread is the finding. One player reached master level with 3,016 hours; another needed 23,608, nearly eight times as many (CEOtudent calculation: 23,608 / 3,016 = 7.8). Some players in the sample had passed 25,000 hours of practice without reaching master level.
Sport tells a similar story from a different angle. Macnamara, Moreau and Hambrick’s 2016 meta-analysis covered 34 studies, 52 independent samples and 2,765 athletes. Deliberate practice explained 18% of the variance in sports performance overall, but only 1% among elite athletes competing at national level or above. Higher-skill athletes also did not start their sport earlier in childhood than lower-skill athletes.
When do practice hours predict skill, and when do they not?
The corrected meta-analysis points to predictability. A separate line of research on intuition points the same way. In a 2009 paper in American Psychologist, Daniel Kahneman and Gary Klein, who came from opposing schools of thought on expert judgment, agreed that the quality of an expert’s intuition depends on two things: how predictable the environment is, and whether the person has had an adequate opportunity to learn its regularities. They also concluded that subjective confidence is not a reliable indicator of accuracy.
Combine the two findings and you get a practical rule: hours turn into skill when the task is predictable and the feedback is fast, clear and tied to your own actions. When either condition fails, hours turn into experience, which is not the same thing.
The table maps common knowledge-work skills onto those two conditions. It is an editorial framework built from the research above, not measured data; no study has estimated variance explained for these specific skills.
Table 3. When practice hours predict skill in knowledge work (CEOtudent editorial framework)
| Skill | Task predictability | Default feedback at work | Do raw hours predict skill? | What makes an hour count |
|---|---|---|---|---|
| Touch typing, keyboard shortcuts, tool fluency | High | Immediate and objective (speed, errors) | Strongly, up to a plateau | Timed drills on your slowest movements, not more typing |
| Spreadsheet modelling, SQL, writing code to a specification | High | Fast and objective (it runs, the numbers reconcile) | Fairly well | Problems slightly beyond your level, solved before checking the answer |
| Presenting and public speaking | Moderate | Delayed and polite | Weakly | Record, watch, fix one named habit per session |
| Writing for a target reader | Moderate | Delayed, filtered by editors or AI tools | Weakly | Draft first, then compare against an expert edit and log the patterns you missed |
| Sales and negotiation conversations | Moderate to low | Outcome arrives weeks later, mixed with luck | Weakly | Rehearse specific moments with a partner, review recordings against a rubric |
| Hiring, strategy, forecasting markets | Low | Slow, noisy, rarely traced back to the decision | Barely | Write predictions down with reasons, score them later, track calibration |
How to read it: the higher a skill sits in the table, the more your logged hours mean. The lower it sits, the more the design of the hour matters and the less you should trust your own sense of how good you have become. That is the same gradient the corrected meta-analysis found between running and handling an aviation emergency, applied to a desk job.
This is also where AI changes the economics. In the bottom half of the table, the scarce input has always been feedback. An AI tool can now generate a practice negotiation, critique a draft against a stated rubric, or score last quarter’s forecasts in minutes. Used that way, it moves a skill up the table. Used to produce the output for you, it removes the practice altogether. Our cornerstone on deliberate practice in the age of AI sets out how to decide which repetitions to keep by hand and which to offload.
How do you audit the quality of your practice hours?
If quality matters more than quantity, you need a way to tell the two apart. The audit below turns the defining features of deliberate practice from the 1993 paper (a specific goal, effortful work at the edge of current ability, immediate feedback, repetition with refinement, and limited daily duration) into six questions you can ask about any block of time you count as learning.
Table 4. The hour-quality audit (CEOtudent editorial framework)
| Question about this hour | Score 0 | Score 1 | Score 2 |
|---|---|---|---|
| 1. Was there a specific target before I started? | “Get better at X” | A topic | One named weakness with a pass mark |
| 2. Was it at the edge of my ability? | Comfortable, mostly things I can already do | Some hard parts | I failed often and had to slow down |
| 3. How fast did I learn whether I was right? | Weeks later or never | End of the session | Within seconds or minutes |
| 4. Was the feedback independent of my own opinion? | Only my impression | A tool or AI with no rubric | A teacher, an objective score, or an AI checked against a rubric I set before starting |
| 5. Did I repeat the weak part after correcting it? | No | Once | Several times, until it held |
| 6. Did I do the thinking, or did a tool? | The tool produced the answer | Shared | I produced it first, then compared |
How to use it. Score a typical week of learning time. Count an hour at full value only if it scores 9 or more out of 12; count it as half an hour at 6 to 8; and do not count it at all below 6. The thresholds are an editorial convention for planning, not research findings, but the exercise usually produces a sobering number. Many people who say they have “practised for years” find that most of their hours were work or play in Ericsson’s terms, valuable for other reasons but not practice.
Three rules from the evidence keep the audit honest:
- Log, do not remember. In the meta-analysis, remembered practice showed a stronger link to performance than logged practice, which suggests memory flatters. Record time and score at the end of each session.
- Cap the daily dose. The violinists who practised most did about 3.5 hours a day and protected rest and sleep. Above roughly 4 hours a day, the training studies Ericsson reviewed showed essentially no further benefit.
- Re-score the bottom half of Table 3 more harshly. In low-predictability skills, confidence grows faster than accuracy. Kahneman and Klein’s warning that subjective experience is not a reliable guide applies to judging your own progress too.
What should a CEO-minded learner take from this?
Think about practice the way a CEO thinks about capital. A budget measured only in hours tells you what you spent, not what you built. The useful question is return per hour, and the research says that return depends on the environment and the feedback more than on effort. The student half of the lens supplies the discipline: deliberate practice is not enjoyable by design. It means working on what you are bad at, in front of honest feedback, in limited daily doses, for a long time. The 10,000-hour rule made that sound like a waiting game. The research makes it a design problem.
Four practical moves follow:
- Replace the target number with a target performance. “10,000 hours” is an input. “Close a negotiation role-play scored 8 of 10 by a skeptical partner” is an output you can practise toward. For realistic time ranges by skill and level, see how long it takes to learn a skill.
- Spend effort on feedback before you spend it on hours. If a skill sits in the bottom half of Table 3, build the feedback loop first: recordings, scored predictions, a rubric, a coach, or an AI critic.
- Run the hour-quality audit monthly. It takes ten minutes and shows where your time is going.
- Compare yourself with your own log, not an average. The fastest and slowest chess masters in Gobet and Campitelli’s sample differed by nearly eight times.
Next steps: our guide to meta-skills that make every skill faster and our analysis of how expert intuition develops.
Frequently asked questions
Is the 10,000-hour rule true?
Not as a rule. It comes from Ericsson’s 1993 violin study, where the best students had accumulated an average of 7,410 hours of practice alone by age 18 and passed 10,000 hours on average around age 20. It was a group average in one domain, not a threshold, and Ericsson later rejected the idea of a magical number.
How much does practice really explain?
In the largest meta-analysis, as corrected in 2018, deliberate practice explained about 14% of the differences in performance overall: 24% in games, 23% in music, 20% in sports, 5% in education and 1% in professions. Ericsson argued that with a stricter definition of deliberate practice the figure is much higher, around 29%, or 61% after correcting for measurement error.
Why do articles say 26% for games and less than 1% for professions?
Those are the original 2014 figures. A corrigendum published in Psychological Science in 2018 corrected an error in how the authors adjusted for dependent effect sizes. The corrected figures are 24% for games and 1% for professions, with 14% overall instead of 12%. Many articles have not caught up.
Did the famous violin study replicate?
Partly. A 2019 double-blind replication with 39 violinists found that practice still related to skill and explained 26% of performance variance, but the best violinists had not practised more than the good ones. The original study reported 48%.
Does practice work for office and knowledge-work skills?
It can, but the evidence is thin. In the corrected meta-analysis the professions result (1%) was not statistically significant, and in Ericsson’s stricter reanalysis no professions study met the criteria. Practice works best where tasks are predictable and feedback is fast, so for knowledge work the priority is to build feedback into the practice.
How many hours a day should I practise?
In the 1993 study the top violinists practised alone about 3.5 hours a day, every day, and slept and napped more than the less accomplished group. The training studies reviewed in the same paper found essentially no benefit beyond 4 hours a day. For most working adults, a shorter daily block of high-quality practice is the realistic target.
Sources
- Ericsson, K. A., Krampe, R. T., and Tesch-Romer, C. (1993). The Role of Deliberate Practice in the Acquisition of Expert Performance. Psychological Review, 100(3), 363-406.
- Macnamara, B. N., Hambrick, D. Z., and Oswald, F. L. (2014). Deliberate Practice and Performance in Music, Games, Sports, Education, and Professions: A Meta-Analysis. Psychological Science, 25(8), 1608-1618.
- Corrigendum: Deliberate Practice and Performance in Music, Games, Sports, Education, and Professions: A Meta-Analysis (2018). Psychological Science, 29(7), 1202-1204. doi 10.1177/0956797618769891.
- Macnamara, B. N., Moreau, D., and Hambrick, D. Z. (2016). The Relationship Between Deliberate Practice and Performance in Sports: A Meta-Analysis. Perspectives on Psychological Science, 11(3), 333-350. With Ericsson, K. A. (2016). Summing Up Hours of Any Type of Practice Versus Identifying Optimal Practice Activities: Commentary. Perspectives on Psychological Science, 11(3), 351-354.
- Macnamara, B. N., and Maitra, M. (2019). The role of deliberate practice in expert performance: revisiting Ericsson, Krampe and Tesch-Romer (1993). Royal Society Open Science, 6, 190327.
- Ericsson, K. A., and Harwell, K. W. (2019). 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.
- Hambrick, D. Z., Oswald, F. L., Altmann, E. M., Meinz, E. J., Gobet, F., and Campitelli, G. (2014). Deliberate practice: Is that all it takes to become an expert? Intelligence, 45, 34-45.
- Gobet, F., and Campitelli, G. (2007). The role of domain-specific practice, handedness, and starting age in chess. Developmental Psychology, 43(1), 159-172.
- Kahneman, D., and Klein, G. (2009). Conditions for intuitive expertise: A failure to disagree. American Psychologist, 64(6), 515-526.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:
