How much performance can deliberate practice explain?

Deliberate practice matters. It just does not explain performance with one number.

Across 88 studies, the corrected average was 14% of the differences in performance between people. The estimate ranged from 1% in professions to 24% in games. Studies that measured practice with contemporaneous logs found only 4%. A stricter reanalysis found 29% before measurement correction.

Then a study published online in June 2026 for the journal’s July 2026 issue changed the quality of the measurement. It followed the time-stamped activity and ratings of 44,213 Chess.com players. Practice-aligned activity was associated with 3.61 times as much rating improvement per hour as playing games.

Those results are not contradictions. They answer different questions.

The 10,000-hour rule was never a rule

The number came from a 1993 paper by K. Anders Ericsson, Ralf Krampe and Clemens Tesch-Römer.

In its best-known study, violinists reconstructed how many hours they had practiced in each year of their musical development. At age 20, the best group reported a cumulative average above 10,000 hours. The good group reported about 7,800; the least accomplished group about 4,600.

This was an important pattern. It was not a threshold experiment.

The researchers did not randomly assign novices to 10,000 hours of training. They selected existing groups, asked them to remember earlier practice and compared their averages. Some individuals would have been above or below the group mean. The study did not show that any person becomes an expert when a clock reaches five digits.

It also did not define deliberate practice as simply doing more.

The original framework separated practice from work and play. Deliberate practice targeted a current weakness, demanded attention, repeated a suitable task and delivered informative feedback. Performing, competing or repeating what already felt comfortable could consume hours without qualifying.

The corrected average is 14%, not 12%

In 2014, Brooke Macnamara, David Hambrick and Frederick Oswald combined 88 studies and 157 effect sizes from games, music, sports, education and professions.

The published paper initially reported that deliberate practice explained 12% of performance variance overall. A 2018 corrigendum fixed the adjustment for dependent effect sizes. The corrected estimate is 14%, with a 95% confidence interval from 11% to 18%.

That correction matters because the superseded 12% figure still appears in ranking explainers. It did not reverse the conclusion: practice was associated with performance, while most of the variation remained unexplained.

Explained variance needs careful translation. Fourteen percent does not mean practice caused 14% of one person’s ability. It means that, across the studies and measures analyzed, differences in reported deliberate practice accounted statistically for 14% of differences in measured performance between people. Correlation, measurement error, sample selection and shared causes all sit inside that estimate.

Domain changed the result by a factor of 24

The average hid a wide corrected range.

The professional estimate was not statistically significant. That does not show that practice cannot improve professional skill. It shows that accumulated practice, as the included studies measured it, did not explain much of the between-person performance variation in those settings.

Predictability produced another split. The estimate was 23% in highly predictable activities, 14% in moderately predictable activities and 6% in low-predictability activities.

A runner can repeat a stable movement against a clock. A professional handling changing people, incomplete information and competing goals receives a noisier signal. The same practice hour is less directly connected to one reproducible outcome.

Better measurement made the relationship smaller

How researchers counted practice mattered almost as much as what people practiced.

Retrospective questionnaires produced a corrected estimate of 15%. Practice logs produced 4%.

A log asks what happened near the time it happened. A retrospective estimate may ask someone to reconstruct years of weekly practice after they already know how successful they became. Memory is imperfect, and current identity can color the reconstruction.

Performance measurement moved the estimate too. Studies using standardized objective scores produced 10%, compared with 12% for laboratory tasks and 11% for expert ratings.

The meta-analysis was highly heterogeneous. Its 14% is a map of varied evidence, not a dose label.

Elite performance creates a different comparison

A 2016 sports meta-analysis found that deliberate practice explained 18% of performance variance overall. The estimate was 19% in subelite samples, 29% in mixed samples and just 1% among elite athletes.

That 1% has two meanings—and one common misuse.

Within the elite samples analyzed, accumulated practice hours did little to distinguish who performed better. But selecting only national-level and higher athletes compresses the range. Everyone has already passed demanding filters, and their practice histories may be more similar. A restricted range tends to reduce correlations.

The result therefore does not prove that practice was irrelevant to becoming elite. It does challenge the claim that more accumulated deliberate practice continues to explain most differences once everyone in the comparison is already elite.

Why another review found 29%—then 61%

The deepest disagreement is not arithmetic. It is definition.

Macnamara and colleagues included structured activities created specifically to improve performance. Ericsson and Kyle Harwell argued that many of those activities lacked the individualized design, focused solitary work or reproducible performance required by the original concept.

In 2019, they reapplied stricter criteria and retained 14 effects classified as purposeful or deliberate practice. The pooled association was 29% before correcting for measurement error. After an attenuation correction, they reported 61%.

That does not make 61% the corrected version of 14%.

The reanalysis selected a much smaller subset, changed the exposure definition and adjusted for assumed measurement unreliability. A 2020 counter-review showed how much the corrected value depends on those assumptions. For the eight effects labeled specifically as deliberate practice, assuming .80 reliability for both practice and performance produced 49%; assuming higher reliability required less correction and produced a smaller estimate.

There is a useful principle inside the dispute: measuring hours without measuring what happened during them can flatten meaningful differences in practice quality. There is also a methodological warning: a narrower definition can make the measured relationship larger while answering a narrower question.

The 2026 chess study replaces memory with timestamps

The newest evidence takes a different route.

Daniel Southwick and colleagues analyzed 44,213 Chess.com players with objective, time-stamped records of activity and rating performance. More than 90% of player time went into playing games. Yet the category of deliberate-practice-aligned activities was associated with 3.61 times the rating improvement per hour of gameplay. Not every practice-aligned activity was equally effective.

This is rare real-world evidence that how people train can matter more than the raw time they spend performing the activity.

It does not settle the old variance debate. The study measured marginal learning efficiency inside online chess, not the share of all expert performance caused by practice across domains. Players chose what to do, so motivation, prior skill and goals could affect both activity and improvement. It was longitudinal, not randomized.

The result is still important because it removes one major weakness of earlier work: no one had to remember how they practiced years ago. Activity and outcome were recorded as they happened.

What the evidence supports

Original sources

• Ericsson, Krampe and Tesch-Römer, 1993
• Macnamara, Hambrick and Oswald, 2014
• 2018 corrigendum
• Macnamara, Moreau and Hambrick, 2016
• Ericsson and Harwell, 2019
• Hambrick, Macnamara and Oswald, 2020
• Southwick et al., 2026

Independent editorial summary. The authors, journals and Chess.com are not affiliated with LifeScore.