Chaos and Order

💡 왼쪽 원문을 읽으면서 오른쪽에 따라 써보세요. Tab 키로 힌트를 받을 수 있습니다.

Introduction

Korean self-improvement talk contains one old sentence. People who are good at studying are also good at sport, and people who are good at sport are also good at their work. Someone who has done one thing properly will do the next thing properly too.

The sentence feels good. It tells you that the one thing you are doing right now will someday help with everything else. The problem is that this sentence is one of the longest-running, most repeatedly tested, and most frequently failed hypotheses in psychology.

This post explains that failure. And in place of the broad claim that failed, it puts forward the narrow claim that survived. I want to stress the narrowness in advance. By the time you finish reading, the number of things you believe will have gone down. What remains will be a little firmer.

1. Transfer has to be split in two first

In learning research, transfer means the phenomenon of something learned in one place being used in another. But this word says nothing at all unless it is split in two.

Near transfer is the case of moving to a task that shares many elements with the trained task. Doing working-memory training and getting better at other memory tasks; practising a forehand and improving on forehands at a similar angle.

Far transfer is the case of moving to a situation entirely unlike the training situation. Playing chess and getting better at mathematics, playing brain games and making better everyday judgements, learning an instrument and gaining in logical ability.

What popular discourse always means is far transfer. And what research has repeatedly confirmed is near transfer. That mismatch is the backbone of this whole post.

2. Why the evidence for far transfer keeps disappearing

The most direct material is a second-order meta-analysis in the field of cognitive training. That means a study that meta-analyses meta-analyses. The study Sala and colleagues published in Collabra: Psychology in 2019 integrated several meta-analyses covering working-memory training, video games, music, chess, and exergames (Sala et al., 2019, read 2026-08-16).

The results split as follows. In the model dealing with near transfer (k = 99), an effect of working-memory training carrying over to memory tasks was confirmed, and its size varied by target population. By contrast, in the models dealing with far transfer (k = 119, k = 233) the effects were small or absent. To carry the authors' own wording over directly: once placebo effects and publication bias were controlled for, both the overall effect size and the true variance became zero.

This conclusion is not the failure of one particular programme. The point of the study is that changing the type of programme, and whether the participants were children, adults, or older adults, left the result the same.

The second source is a comprehensive review of the brain-training industry. The review Simons and colleagues published in Psychological Science in the Public Interest in 2016 examined only the papers that the brain-training companies themselves put forward as evidence. In effect, it looked at the most favourable evidence available (Simons et al., 2016, read 2026-08-16).

The conclusion comes in three steps. Evidence that performance on the trained task itself improves is abundant. Evidence that it carries over to closely related tasks is thinner. Evidence that distant tasks or everyday cognitive performance improve is almost nonexistent. And among the studies cited, not one satisfied all of the best-practice procedures the authors set out.

The third source comes from the sports side. Fransen's 2024 piece in Sports Medicine reviews general perceptual-cognitive training tools such as stroboscopic visual training, neurofeedback, and executive-function training, and concludes that the evidence that commercial cognitive-training equipment carries over into actual competitive performance is poor (Fransen, 2024, read 2026-08-16).

Three sources from three different fields arrive at a conclusion of the same shape. What you trained gets better. What sits next to it gets a little better. What sits far away does not get better.

3. The sentences this conclusion knocks down

The single fact that far transfer is weak invalidates a fairly large number of popular claims.

Playing chess or Go improves your thinking in general.
Building concentration with a brain-game app raises your performance at work.
Learning an instrument makes you better at mathematics.
Studying a difficult subject trains the brain in itself.

These sentences share the same structure. Do a particular activity X and a domain-independent general ability goes up. And the failure of that structure is one of the most robust findings of the last twenty years of cognitive-training research.

Let me make one thing clear here. This is not to say that playing chess is bad or that learning an instrument is wasted. You become good at chess and you become good at playing your instrument. That is value enough on its own. What collapses is not the value of the activity but the rhetorical move of selling that activity as a detour toward some other ability.

4. So why does the intuition keep surviving?

Even with the evidence for far transfer this weak, the sense that studying and sport connect does not go away. Mixed in here is not a data problem but a problem in the structure of observation. The three items below are not research findings but interpretations of why things look that way to us.

First, the selection effect. The reason we so often see people who are good at both domains is that the people who were able to attempt both domains are already a filtered group. Only those with access to time, physical capacity, and coaching try both. Look at the correlation inside that group and it looks as though ability has carried over.

Second, base-rate neglect. Someone who is good at studying and good at sport stands out and stays in memory. The person who studies well but is average at sport, and the person who plays well but has average grades, make no story and so go uncounted.

Third, the retrospective account of successful people. When someone successful in a field explains their own success, they generally end up emphasising what carried over. It makes for the better story. But the people who trained the same way and did not succeed leave no retrospective account.

Let me sum it up with a constructed example. Suppose that at some company, five developers who run amateur marathons all get good performance reviews. Before concluding that running built their work ability, there are things you need to count. How many developers who do not run get good reviews, and how many people started running and quit, and what their reviews looked like. Without counting those two cells, no correlation has been established at all. This is not a real case but an example constructed to display the logic.

5. So what actually does carry over?

From here the grade of the evidence drops by one notch. Let me say that up front.

That far transfer barely exists is a replicated conclusion. What I am about to say, by contrast — the claim that learning behaviours carry over — does not have direct evidence of anything like that strength. It is closer to a synthesis reasoned backwards from a number of research findings. So the list below should be read not as established fact but as a grounded working hypothesis.

5.1 How you handle feedback

The condition that keeps recurring in expertise research is immediate, informative feedback. It is exactly one of the five criteria Ericsson set out in 2020 when he re-explained his own 1993 framework: the condition that you must receive immediate, informative, actionable feedback on your performance (Ericsson, 2020, read 2026-08-16).

Feedback takes a different form in each domain. In table tennis, where the ball went; in code, whether the test passes; in language, whether the other person understood. The forms differ, but there are individual differences in how a person handles them. Some people take negative feedback as a judgement on themselves; others take it as the input to the next attempt. Someone who has already built the latter habit in one domain finds that position quickly in another domain too.

5.2 How you structure practice

Deciding what to repeat and how much, in what order to mix things, and when to rest is closer to procedure than to domain knowledge. I will cover this in detail later, but spacing out repetition and practising retrieval are learning techniques whose effects have been confirmed across several domains (Dunlosky et al., 2013, read 2026-08-16).

Someone who has taken this procedure into their body once does not have to reinvent it from scratch in a new domain. But we should distinguish whether this is an ability carrying over or simply knowing a good method and applying it. It is closer to the latter. And the fact that it is the latter is, practically speaking, the better news. Methods can be learned.

5.3 How you diagnose your own errors

This is the one that looks most likely to carry over: the ability to say specifically what you cannot do.

One difference between a beginner and a skilled performer is the resolution at which they describe failure. The beginner says it did not go well today; the skilled performer says that when the ball came to the backhand the body opened first and the timing was late. The second sentence already contains what to practise next.

This habit requires domain knowledge. Without the word backhand you cannot build that sentence. So this too is not full far transfer. What carries over is not the content of the diagnosis but the habit of pushing the diagnosis to that resolution.

6. Why conscientiousness or general ability is not enough of an explanation

The explanation that "in the end, conscientious people do well at everything" is widely used but has two problems.

First, it easily becomes circular. If you look at someone doing well and say they were conscientious, then say they do well because they were conscientious, you have predicted nothing.

Second, the explanatory power of practice volume itself is smaller than popular expectation. The meta-analysis Macnamara and colleagues published in Psychological Science in 2014 calculated, domain by domain, what percentage of the variance in performance deliberate practice explains. Games 26 percent, music 21 percent, sports 18 percent, education 4 percent, and under 1 percent in the professions (Macnamara et al., 2014, read 2026-08-16).

The authors' conclusion was not that practice does not matter but that it does not matter as much as has been claimed. And these figures differ greatly by domain. The clearer the rules and the faster the feedback, the larger; the less so, the smaller. In other words, the very attempt to explain every domain with the single variable of conscientiousness does not fit the data.

The attempt to explain things by general ability runs into a similar problem. I will handle that part separately in the fourth post of this series. Here I will note only one thing. The statistical observation that a general ability exists and the claim that this ability rises with training are completely different matters. The former is solid and the latter is not.

7. Evidence grade by claim

Claim	Evidence grade	Source
Near transfer is observed	Replicated	Sala et al. 2019, model 1
Far transfer effects converge on zero once controlled	Replicated	Sala et al. 2019, models 2 and 3
Brain training improves everyday cognition	Almost no evidence	Simons et al. 2016
General cognitive training carries over to competitive performance	Almost no evidence	Fransen 2024
The explanatory power of deliberate practice differs by domain	Replicated	Macnamara et al. 2014
Feedback-handling habits are reused across domains	Working hypothesis	No direct evidence available
Error-diagnosis resolution is reused across domains	Working hypothesis	No direct evidence available
Conscientiousness explains achievement in every domain	Insufficient evidence	Conflicts with Macnamara et al. 2014

What has not been verified

The weakest part of this post is section 5. I could not find a meta-analysis that directly tests the claim that learning behaviours carry across domains. So the three items in section 5 are grounded inference, not established results.

There are also open points within far-transfer research itself. What Sala et al. 2019 dealt with is cognitive-training programmes. Far longer and more complex interventions, such as years of formal education or occupational experience, were not the object of that analysis. That short training programmes produce no far transfer, and that long experience does not change a person, are not the same statement.

One more thing. This field is especially sensitive to placebo effects and publication bias. That the effect became zero once controlled means, read the other way, that effects did appear in the studies that did not control. Individual studies with positive results will keep appearing in the future too. A single study will not overturn this conclusion, but neither is this conclusion fixed forever.

Finally, what this post does not cover. The influence of factors such as motivation, identity, environment, and access on achievement is not treated here at all. Transfer research does not answer that question.

Closing

To sum up. The claim that there is a shared general ability between being good at studying and being good at sport has weak evidence. If something is shared, it is more likely to be a manner than an ability. How you handle feedback, how you lay out practice, the resolution at which you see your own errors.

This is a far narrower conclusion than you might have hoped for. But it is actionable precisely because it is narrow. General ability cannot be trained; manner can be changed.

There is one post that reaches the same conclusion by a different route. Fourteen Things Table Tennis Taught Me starts from experience rather than research and arrives at conclusions like: playing only for fun will not make you better, and you have to practise what you cannot do. Where a conclusion drawn from experience and a conclusion drawn from data overlap is worth trusting; where they do not overlap, both are worth doubting.

Next post: Experts Do Not Know More, They See Differently — a story about perception that starts with the chessboard experiment.
Deliberate Practice Path Tool — a tool for fixing on a weakness and laying out a repetition cycle.

References

Below are the sources checked directly on 2026-08-16 while writing this post. The links are the addresses actually read, and some are public API responses from Europe PMC.