TL;DR. The claim that you learn better when teaching is matched to your “style” has never passed the experiment that would validate it. That much is settled. What is less widely known is that two of the most popular replacement explanations are also weaker than advertised. A meta-analysis of 8,776 effect sizes found that prior knowledge predicts where you end up but barely predicts how much you gain. And the deliberate practice literature, once its 2018 corrigendum is applied, explains 5 percent of performance differences in education and 1 percent in professions. Most of what people call a fast learner is not a trait at all. It is a gap between what learners believe works and what they actually do, and that gap is measurable.
The experiment learning styles never passed
Start with the part that is not in dispute.
In 2008, Harold Pashler, Mark McDaniel, Doug Rohrer and Robert Bjork were commissioned to review whether learning styles assessments deserved a place in education. Their report, published in Psychological Science in the Public Interest, is the reference point for every serious discussion since.
Their contribution was not an opinion. It was a specification. They set out exactly what evidence would have to look like for the idea to be valid. Students must be sorted into groups by learning style. Students from each group must then be randomly assigned to one of several instructional methods. Everyone must sit the same final test. And then the results must show a particular pattern: the method that works best for one style group must not be the method that works best for another.
That pattern has a name. It is a crossover interaction, and it is the whole ballgame. Without it, the most you have shown is that one teaching method is better for everyone, which says nothing about styles.
The authors reported that they “found virtually no evidence for the interaction pattern mentioned above, which was judged to be a precondition for validating the educational applications of learning styles.” They also noted that “although the literature on learning styles is enormous, very few studies have even used an experimental methodology capable of testing the validity of learning styles applied to education,” and that of those which did, “several found results that flatly contradict the popular meshing hypothesis.”
Their conclusion was unambiguous: “there is no adequate evidence base to justify incorporating learning-styles assessments into general educational practice.”
Two things are worth holding onto, because careless summaries drop both. First, the review confirmed that people do have preferences, and that people do differ in specific aptitudes. Preferences are real. The claim that fails is that matching instruction to those preferences improves learning. Second, the authors explicitly declined to say every possible version has been tested and refuted. Most have simply never been tested.
That is the honest state of the question. A billion-dollar idea rests on an experiment almost nobody ran.
The replacement that also fails
Here is where most articles stop, hand you a list of “real” predictors, and move on. That list usually starts with prior knowledge. The reasoning feels airtight: people who already know a lot about a subject pick up new material in it faster.
It is called the knowledge-is-power hypothesis, and in 2022 it got the same treatment Pashler gave learning styles.
Bianca Simonsmeier, Maja Flaig, Anne Deiglmayr, Lennart Schalk and Michael Schneider examined the relationship between prior knowledge and learning across 8,776 effect sizes, publishing in Educational Psychologist. They separated two things that everyday language runs together.
The first is where you finish. The correlation between what learners knew on the pretest and what they knew on the posttest was high, at 0.534. People who start ahead tend to finish ahead. Nobody disputes this.
The second is how much you gain. The correlation between pretest knowledge and normalised knowledge gains was 0.059 in the negative direction, and the 95 percent prediction interval ran from negative 0.688 to positive 0.621.
Read that interval again. It spans almost the entire possible range in both directions. Prior knowledge sometimes helps a great deal, sometimes hurts a great deal, and on average does essentially nothing to how much you gain.
The authors were careful about what this does and does not license. In their words, “this strong variability falsifies general statements such as ‘knowledge is power’ as well as ‘the effect of prior knowledge is negligible.’” The finding is not that expertise is worthless. It is that the direction of its effect depends on conditions nobody has yet identified.
An experimental follow-up published in the Journal of Experimental Psychology: Applied in 2025 pushed the same question harder. Zachary Buchin and Neil Mulligan randomly assigned participants to receive three days of training in one of two academic domains, then had everyone learn new material in both. Because training covered only three of four topics per domain, the untrained topic gave a clean measure of new learning inside a domain the participant now knew more about. Their result: new learning, measured either as final test performance or as knowledge gains, did not differ between the high and low domain knowledge conditions.
So the second answer is weaker than the first sounded. Prior knowledge is an excellent predictor of your level. It is a poor predictor of your slope.
The replacement after that, corrected
The third answer people reach for is practice volume. This is the popular-science position: nobody is born fast, some people simply put in the hours.
Brooke Macnamara, David Hambrick and Frederick Oswald tested it across every major domain in which deliberate practice had been studied, publishing in Psychological Science in 2014. Their headline numbers circulated widely: 26 percent of performance variance explained in games, 21 percent in music, 18 percent in sports, 4 percent in education and under 1 percent in professions.
Those are the numbers almost every article still quotes. They are also superseded.
In 2018 the authors issued a corrigendum. They had used a sample-size adjustment formula incorrectly, applying it to each individual effect size rather than to the average of dependent effect sizes. They reanalysed and published a full table of corrected values. The substance of their conclusion did not change, but several figures did, and the corrected set is what an accurate 2026 discussion should use.
Deliberate practice and performance: originally reported versus corrected figures
| Domain or measure | As published in 2014 | Corrected in the 2018 corrigendum |
|---|---|---|
| Average correlation, practice and performance | 0.35, 95% CI 0.30 to 0.39 | 0.38, 95% CI 0.33 to 0.42 |
| Overall variance in performance explained | 12%, 95% CI 9% to 15% | 14%, 95% CI 11% to 18% |
| Games | 26% (r = 0.51) | 24% (r = 0.49) |
| Music | 21% (r = 0.46) | 23% (r = 0.48) |
| Sports | 18% (r = 0.42) | 20% (r = 0.45) |
| Education | 4% (r = 0.21) | 5% (r = 0.22) |
| Professions | under 1% (r = 0.05, p = 0.62) | 1% (r = 0.09, p = 0.377) |
Source: Macnamara, Hambrick and Oswald, Psychological Science, 2014, and the 2018 corrigendum, Table 1. Figures are reproduced as printed.
Even after the upward correction, the picture is sobering. Across all domains, deliberate practice leaves 86 percent of performance differences unexplained. In education it explains 5 percent. In professions the corrected correlation of 0.09 is not statistically significant at all, with a p-value of 0.377.
If you work in a profession, the amount of deliberate practice you have logged is, on this evidence, close to useless as a predictor of how good you are relative to your peers.
What the same study found that almost nobody quotes
Buried in the same corrigendum table is a moderator that does more explanatory work than any of the person-level variables above.
The authors split activities by how predictable the task environment is. Corrected, deliberate practice explained 23 percent of performance variance in activities high in predictability, 14 percent in activities moderate in predictability, and 6 percent in activities low in predictability.
That is the finding worth carrying. The strongest single determinant in this literature is not a property of the learner. It is a property of the domain. In stable, rule-bound environments where the same situation recurs and feedback is fast, practice compounds. In unstable environments where the situation never repeats and feedback is slow and noisy, it compounds far less.
This reframes the whole question. Asking “how fast do I learn” is close to meaningless. Asking “how learnable is this particular thing, and what feedback does it give me” is answerable, and it changes what you do next.
It also explains the education and professions figures without any appeal to talent. Professional performance is judged in a low-predictability environment. Hours logged cannot rescue you from that.
The gap that explains the rest
If style does not predict learning, prior knowledge predicts level but not gain, and practice volume explains a minority of the variance, what is left?
Method. And here the evidence is strong, well replicated, and almost universally ignored in practice. That is the interesting part.
Two independent datasets, collected for different purposes, can be joined to show the size of the problem.
The first is belief. Philip Newton and Atharva Salvi reviewed 33 studies containing 37 samples, covering 15,405 educators across 18 countries between 2009 and early 2020, published in Frontiers in Education. A weighted 89.1 percent agreed with the learning styles matching claim, with individual studies ranging from 58 percent to 97.6 percent. Belief was highest among trainee teachers at 95.4 percent, 87.8 percent among qualified teachers, and lowest in higher education at 63.6 percent. Crucially, 79.7 percent said they used, or intended to use, the matching of instruction to learning styles.
The second is behaviour. Jeffrey Karpicke, Andrew Butler and Henry Roediger surveyed 177 undergraduates at Washington University in St. Louis, publishing in Memory. Asked to list and rank the strategies they used, 84 percent reported rereading notes or textbooks, and 55 percent named rereading as their single most-used strategy. Only 11 percent, that is 19 of 177 students, reported practising testing themselves by recalling information. Only 1 percent, 2 students out of 177, named it as their top strategy. Given a forced choice after reading a chapter, 57 percent chose to restudy, 21 percent said they would use some other technique, and 18 percent said they would test themselves. Those figures do not sum to 100 because the authors excluded ambiguous responses they could not score, and the paper reports the headline result as 78 percent indicating they would not test themselves after reading a chapter.
Now set those against the utility ratings from the landmark review by John Dunlosky, Katherine Rawson, Elizabeth Marsh, Mitchell Nathan and Daniel Willingham, also in Psychological Science in the Public Interest. They assessed ten techniques for how well their benefits generalise. Practice testing and distributed practice received high utility assessments. Rereading and highlighting received low utility assessments, and the authors noted that most students report using them anyway.
The belief-behaviour-evidence gap
CEOtudent editorial framework. Belief and behaviour percentages are from Newton and Salvi and from Karpicke, Butler and Roediger respectively; utility ratings are from Dunlosky and colleagues. The ratio column is calculated by CEOtudent from the published percentages.
| Practice | Share who believe in or use it | Evidence rating | Mindshare-to-evidence mismatch |
|---|---|---|---|
| Matching instruction to learning style | 89.1% of educators agree; 79.7% use or intend to | No adequate evidence base (Pashler et al.) | Highest belief, no supporting experiment |
| Rereading notes or textbook | 84% of students report it; 55% rank it first | Low utility (Dunlosky et al.) | Reported 7.6 times more often than self-testing |
| Practice testing by recall | 11% of students report it; 1% rank it first | High utility (Dunlosky et al.) | Ranked first 55 times less often than rereading |
| Distributed practice over time | Not measured in these samples | High utility (Dunlosky et al.) | Evidence without a usage figure in this data |
Two caveats belong with this table, and they are not small. The educator and student samples are different populations, measured a decade apart, answering different questions. This is a mindshare comparison, not a controlled contrast. And Newton and Salvi themselves flagged that most of the studies they reviewed used convenience sampling with small samples, and that only two briefed participants on what matching actually entails before asking about belief. Treat the 89.1 percent as an upper bound on informed belief.
With those caveats applied, the pattern still holds. The technique with the strongest evidence is the one almost nobody uses. The technique with the weakest evidence is the one almost everybody uses. And the idea with no supporting experiment at all commands near-universal agreement among the people who teach.
There is a reason for this, and it is not stupidity. Karpicke and colleagues found that among the minority who did choose self-testing, only 18 percent said they did it because they learn more that way. Sixty-eight percent said they did it to figure out how well they had learned the material. Even the people doing the right thing are mostly doing it for the wrong reason, which is why they abandon it under time pressure. The authors proposed that students experience illusions of competence while studying: rereading feels fluent, and fluency feels like knowing.
The scale of the underlying effect is not in doubt. Nicholas Cepeda, Harold Pashler, Edward Vul, John Wixted and Doug Rohrer synthesised 839 assessments of distributed practice drawn from 317 experiments across 184 articles in Psychological Bulletin, finding that the optimal gap between study sessions grows as the interval before the test grows. Olusola Adesope, Dominic Trevisan and Narayankripa Sundararajan, meta-analysing practice testing in Review of Educational Research, reported that practice tests are more beneficial for learning than restudying and all other comparison conditions.
What to do instead
The CEO half of this is the diagnosis. The student half is the practice.
Stop asking what kind of learner you are. The question has no validated answer and the assessment industry built on it sells a result no experiment supports. Nothing follows from your score.
Ask how predictable the thing you are learning is. This is the variable with the largest measured effect in the deliberate practice data, and unlike your style it is knowable. If the situation recurs and feedback arrives fast, volume of practice will pay. If neither is true, invest in feedback quality before volume, because more repetitions of an unmeasurable activity buy you very little.
Treat prior knowledge as an asset for level, not a guarantee of speed. Your existing expertise reliably predicts where you will finish relative to a beginner. It does not reliably predict that you will gain more from the next hour of study than they will. Do not let experience in a domain excuse you from the method.
Assume your sense of progress is unreliable. The fluency of rereading is the specific illusion the research identifies. If a study session felt smooth, that is evidence about the session, not about your memory. The only way to find out what you know is to try to retrieve it without the page in front of you.
Test yourself to learn, not only to check. This is the single highest-leverage correction available, because the survey data shows most people who self-test are doing it as a diagnostic rather than as the learning event itself. Retrieval is not the measurement. Retrieval is the mechanism.
Space the sessions, and space them wider the further away the test is. That is the specific shape of the Cepeda finding, and it is the part most study schedules get wrong.
None of this is a personality. All of it is a protocol, which means anyone can run it.
For the technique-by-technique ranking that sits underneath these recommendations, see our companion piece on what the evidence says about twelve study techniques. For applying this under time pressure, see our work on fast upskilling protocols and on reskilling after 40.
Frequently asked questions
Is it fair to call learning styles a myth when the authors said not every version was tested?
Yes, with precision about what is being claimed. The specific claim in circulation, that matching instruction to a diagnosed style improves learning, is the meshing hypothesis, and it is the one that failed. Pashler and colleagues found virtually no evidence for the required interaction and several results contradicting it. Their caveat is that untested variants exist, not that the popular version survived.
Do people have learning preferences at all?
They do. The same review found ample evidence that children and adults express preferences about how information is presented to them, and plentiful evidence that people differ in specific aptitudes. The preference is real. The instructional payoff from matching it is what is missing.
Does prior knowledge really not help?
It reliably predicts your final level, with a correlation of 0.534. What it does not reliably predict is your gain, where the correlation was 0.059 in the negative direction with a prediction interval spanning from negative 0.688 to positive 0.621. The authors explicitly rejected both the claim that knowledge is power and the claim that its effect is negligible. The effect exists and its direction depends on conditions not yet identified.
Why do the deliberate practice numbers here differ from the ones I have seen elsewhere?
Because most sources quote the 2014 figures and the authors corrected them in 2018. Games moved from 26 to 24 percent, music from 21 to 23, sports from 18 to 20, education from 4 to 5, and professions from under 1 percent to 1 percent, while the overall variance explained rose from 12 to 14 percent. The corrected figures are the ones used above.
If practice explains so little, is talent the answer?
That is not what the data supports either. The largest moderator in the corrected analysis was the predictability of the task environment, not any property of the person. The unexplained variance includes measurement error, the quality rather than quantity of practice, starting age, coaching, opportunity and much else. Reading 86 percent unexplained as 86 percent talent is an inference the study does not license.
How long before retrieval practice feels natural?
The survey evidence suggests the obstacle is not difficulty but belief. Only 18 percent of the students who chose self-testing did so because they thought they would learn more that way. Until you expect retrieval to feel harder and work better, you will drift back to rereading whenever a deadline arrives.
Sources and further reading
Pashler, H., McDaniel, M., Rohrer, D., and Bjork, R. Learning Styles: Concepts and Evidence. Psychological Science in the Public Interest, volume 9, issue 3, pages 105 to 119, 2008.
Simonsmeier, B. A., Flaig, M., Deiglmayr, A., Schalk, L., and Schneider, M. Domain-Specific Prior Knowledge and Learning: A Meta-Analysis. Educational Psychologist, volume 57, issue 1, pages 31 to 54, 2022.
Buchin, Z. L., and Mulligan, N. W. Prior Knowledge and New Learning: An Experimental Study of Domain-Specific Knowledge. Journal of Experimental Psychology: Applied, volume 31, issue 2, pages 84 to 98, 2025.
Macnamara, B. N., Hambrick, D. Z., and Oswald, F. L. Deliberate Practice and Performance in Music, Games, Sports, Education, and Professions: A Meta-Analysis. Psychological Science, volume 25, issue 8, pages 1608 to 1618, 2014, together with the Corrigendum published in Psychological Science, volume 29, issue 7, pages 1202 to 1204, 2018.
Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., and Willingham, D. T. Improving Students’ Learning With Effective Learning Techniques: Promising Directions From Cognitive and Educational Psychology. Psychological Science in the Public Interest, volume 14, issue 1, pages 4 to 58, 2013.
Karpicke, J. D., Butler, A. C., and Roediger, H. L. Metacognitive Strategies in Student Learning: Do Students Practise Retrieval When They Study on Their Own? Memory, volume 17, issue 4, pages 471 to 479, 2009.
Newton, P. M., and Salvi, A. How Common Is Belief in the Learning Styles Neuromyth, and Does It Matter? A Pragmatic Systematic Review. Frontiers in Education, volume 5, article 602451, 2020.
Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., and Rohrer, D. Distributed Practice in Verbal Recall Tasks: A Review and Quantitative Synthesis. Psychological Bulletin, volume 132, issue 3, pages 354 to 380, 2006.
Adesope, O. O., Trevisan, D. A., and Sundararajan, N. Rethinking the Use of Tests: A Meta-Analysis of Practice Testing. Review of Educational Research, volume 87, issue 3, pages 659 to 701, 2017.
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:
