The Road to Becoming an Expert: Feedback First, Volume Second

Introduction

Reduced to a single paragraph, what the previous four posts set out is this.

The evidence that ability moves across to a distant domain is weak. Expertise is closer to pattern perception than to a quantity of knowledge, and those patterns are attached to their domain. What has been well verified in practice is spacing and retrieval, and the feeling during practice cannot serve as an indicator of learning. And the attempt to summarise a person with the concept of a general ability is not supported by the data.

This post moves that into action. But let me say one thing in advance. None of the five items below guarantees a result. Advice that says it guarantees one does not exist in this field. And the last section sets out the cases where this advice does not hold, which may be more practical than the five items before it.

1. Narrow the domain to where feedback arrives

Of the five criteria Ericsson set out in 2020 when he explained his original framework again, three are related to feedback. A well-defined task with a clear goal, immediate informative feedback that can be turned into action, and the opportunity to repeat the same or a similar task (Ericsson, 2020, read 2026-08-16).

A criterion for choosing a domain comes out of this. Whether a domain is a good one to learn in can be judged not by how fun it is or how promising it looks, but by the speed and the resolution of its feedback.

Why this matters shows up in numbers as well. In the 2014 meta-analysis by Macnamara and colleagues, the variance explained by deliberate practice was 26 percent in games, 21 percent in music, and 18 percent in sports, while it was 4 percent in education and under 1 percent in the professions (Macnamara et al., 2014, read 2026-08-16).

Look at that ordering and one regularity appears. The domains at the top are the ones where the result comes out immediately and clearly. The ones at the bottom are where the result comes out late and mixed in with several factors. The domains where practice has a large effect overlap with the domains where the feedback is good.

Applied to actual work it comes out like this. The goal of wanting to be good at planning gets its feedback months later. Leaving a decision from a meeting written down as a document, by contrast, gets checked at the next meeting. Starting with the latter is better. The goal is not being made smaller; it is being cut into a learnable size.

2. Build the feedback loop before the volume

The most common form of failure is the order being inverted. Do a lot first, and check later.

The replication study Macnamara and Maitra published in Royal Society Open Science in 2019 carried out the original 1993 study again with violin majors. They introduced a double-blind procedure and changed the statistical handling (Macnamara & Maitra, 2019, read 2026-08-16).

The correspondence between cumulative practice volume and skill level was replicated, but the effect size was considerably smaller than in the original. And a further result was reported alongside it: that teacher-designed practice did not explain additional variance beyond practice done alone.

It is important not to exaggerate this result in either direction. It does not mean practice is meaningless. It means that the attempt to explain skill with the single variable of practice volume does not work as well as expected.

So in practical terms it is better to invert it like this. Before deciding how many hours to spend this week, first decide what you will know this week about what went well and what did not. If there is no way to check, do not increase the time. Increasing it will not turn into anything.

3. Practise under the conditions you will perform in

If the practice conditions and the real conditions are different, what you built in practice does not come out in the real thing.

What shows this principle most sharply is the condition-by-condition result of a contextual interference meta-analysis. In the 2024 analysis by Czyż and colleagues, the overall retention effect was a standardised mean difference of 0.63, but under laboratory conditions it was 0.92 and under field conditions it was 0.23, which was not significant (Czyż et al., 2024, read 2026-08-16).

The original meaning of this result is a warning about the generalisability of the contextual interference effect. But there is something that can be read alongside it here. That when the conditions change, the result changes too. If what held in the laboratory did not hold in the field, then in an individual's own practice a difference in conditions is likely to work the same way. This is an inference, not the conclusion of that paper.

Moved into practice it looks like this. If you want to be good at presenting, you have to stand up, put a clock on it and speak, instead of sitting and polishing the script. If you want to handle English in meetings, you have to speak in front of people instead of reading alone. Something you have never once done under pressure does not come out under pressure.

Fourteen Things Table Tennis Taught Me has the same item. That a technique you have not used in a match never ends up being yours. A sentence that came out of experience and a sentence that came out of data point in the same direction.

4. Look for the recurring error, not the enjoyable skill

People lean toward practising what goes well. It goes well so it feels good, and because it feels good they do more of it.

But as we saw in the previous post, learners cannot accurately judge what was effective for them. In the 2023 study by Do and Thomas, participants saw the benefit of interleaved practice and did not recognise that benefit (Do & Thomas, 2023, read 2026-08-16).

If that is so, what to practise must not be decided by feel. It has to be decided by the record.

Let me write down the method as a constructed example. For one week, write down only the moments that failed, briefly. When, what, and how it did not work. Look at that list at the weekend and generally three or four items appear repeatedly. What you practise next week is the repeated item, not the item you want to do. This is not an actual case but an example constructed to explain the procedure.

The core of this method is not the list but the criterion of repetition. Something that failed once may be a situation, but something that has repeated three times is a structure.

5. Expect the plateau

The plateau is closer to a default than to an exception.

The most direct material is on the sports side. In the meta-analysis Macnamara, Moreau, and Hambrick published in Perspectives on Psychological Science in 2016, deliberate practice explained 18 percent of the variance in sports performance. But looking only at athletes at the very highest level, that value dropped to 1 percent (Macnamara et al., 2016, read 2026-08-16).

One more thing came out of the same study. The athletes who reached a high level had not started the sport at a younger age than the athletes at lower levels.

Read practically, these two results come out like this. The higher the skill goes, the smaller the part explained by practice volume becomes. What used to grow when you added time stops doing so from some point onward. This is not a signal of failed self-management but a foreseeable point.

As for what should be done here, I do not have a verified answer. The advice commonly given is to change the type of practice, but this time I could not confirm material that supports that advice. So here I will say only this much, that you should expect it. Expecting and preparing are different things, but an unexpected plateau makes people quit, whereas an expected one does not.

6. On mindset, carefully

Growth mindset is one of the most widely cited concepts from learning psychology in Korea. It is better to write down the current state accurately. The evidence grade of this item is under debate.

The starting point is the two meta-analyses Sisk and colleagues published in Psychological Science in 2018. The first examined the relationship between mindset and academic achievement across 273 samples totalling 365,915 people, and the second examined the effects of interventions across 43 samples totalling 57,155 people. In the authors' own words, the overall effects were weak in both meta-analyses. Alongside that, a result was reported that there may be benefits for students of low socioeconomic status or in academic risk groups (Sisk et al., 2018, read 2026-08-16).

The follow-up meta-analysis Macnamara and Burgoyne published in Psychological Bulletin in 2023 reached a stronger conclusion. In an analysis covering 63 studies and 97,672 people in total, the overall effect was 0.05, and it became non-significant once publication bias was corrected for. Looking only at the 13 studies that actually changed students' mindsets as intended, it was 0.04 and non-significant, and looking only at the 6 studies of the highest quality, it was 0.02 and non-significant. The authors concluded that the apparent effects of growth mindset interventions are likely attributable to inadequate study designs, reporting flaws, and bias (Macnamara & Burgoyne, 2023, read 2026-08-16).

But there is material on the other side too. The nationwide experiment Yeager and colleagues published in Nature in 2019 reported that an online intervention of under one hour, delivered to secondary school students in the United States, improved the grades of lower-achieving students and increased enrolment in advanced mathematics courses. It also reported the condition that the effect held when peer norms matched the message of the intervention (Yeager et al., 2019, read 2026-08-16).

And the commentary Tipton and colleagues wrote in Psychological Bulletin in 2023 pushes back on methodological grounds. When they reanalysed the data of Macnamara and Burgoyne with up-to-date multilevel methods for handling heterogeneity, the result came out at 0.09 overall and 0.15 for the risk group, which was similar to the values Burnette and colleagues had reported (Tipton et al., 2023, read 2026-08-16).

This is the current state. The enthusiasm of the 2010s is not accurate, and neither is the backlash that says it has completely collapsed. The accurate summary is this. The average effect is small, the focus of the dispute is whether there is an effect for particular groups under particular conditions, and the methodological disagreement has not yet been resolved.

Practically, it can be read like this. The expectation that achievement changes just from changing your mindset has weak evidence behind it. On the other hand there is no basis for saying that working on mindset is meaningless. That is about as far as the data allow.

7. Evidence grade by claim

Claim	Evidence grade	Source
The better the feedback in a domain, the more practice explains	Replicated	The domain-by-domain figures in Macnamara et al. 2014
Immediate feedback is a core condition of deliberate practice	Definition within a theoretical framework	Ericsson 2020
The explanatory power of practice volume is smaller than originally claimed	Replicated	Macnamara et al. 2014, Macnamara & Maitra 2019
At the very highest level the explanatory power of practice volume falls sharply	Single meta-analysis	Macnamara et al. 2016
Higher-level athletes did not start earlier	Single meta-analysis	Macnamara et al. 2016
When conditions change, effect sizes change greatly	Replicated	Czyż et al. 2024
Learners cannot judge for themselves which method is effective	Replicated	Do & Thomas 2023
Growth mindset interventions raise achievement	Under debate	Sisk et al. 2018, Macnamara & Burgoyne 2023, Yeager et al. 2019, Tipton et al. 2023
Changing the type of practice breaks through a plateau	Failed to obtain material	No citation

When this advice does not apply

This is the most important section in the post.

First, domains where feedback is slow or absent. All the advice above presupposes feedback. But there are jobs where the result comes out several years later. Changing an organisational culture, long-term research, raising a child. In domains like these the deliberate practice framework does not apply as it stands. The explanatory power of under 1 percent for the professions in the meta-analysis by Macnamara and colleagues is likely to have the same reason behind it. In domains like these, the problem of leaving a record of the grounds for your judgement becomes more important than the problem of raising your skill.

Second, domains where access is more decisive than skill. In some fields, where you are decides the result more than what you can do. Where opportunity is not evenly distributed, advice that talks only about skill is only half honest. None of the material in this series dealt with the question of access, and that is a limitation of this series.

Third, cases where the goal slides toward what is measurable. The advice to prioritise what gives fast feedback has a side effect. You end up doing only what is easy to measure. Many important things are hard to measure, and if you stack up only what is easy to measure, the total grows but the direction disappears.

Fourth, cases where expertise is not the goal. This is the item most often forgotten. The moment you turn a hobby into training, what you were getting from that activity can disappear. There are activities that are fine not improving. The attitude of making every activity an object of improvement is not, in itself, a verified way of living well.

Fifth, cases where your current state needs recovery rather than learning. All the advice above presupposes having something left in reserve. Making a list of your errors while depleted turns easily into self-blame rather than learning.

Closing

Reduced to five sentences it is this.

Cut the goal down to the size that feedback reaches. Build the way of checking before you increase the time. Do it under real conditions. Make the recurring error next week's task. Know in advance that a plateau is coming.

And knowing where these five do not hold matters as much as knowing the five. If your own domain is one where feedback is slow, most of the advice above does not apply as it stands. Not knowing that and blaming yourself is the most common side effect of advice in this field.

One last thing. What this series confirmed over and over comes down to one point. Expertise is narrow. Narrow sounds like bad news, but it is actually good news. What is narrow can be decided on, and what can be decided on can be started.

Previous post: What It Means to Be Smart — what gets measured and what does not.
First post in the series: What Studying Well and Playing Sport Well Share Is Narrower Than It Looks
Fourteen Things Table Tennis Taught Me — a post that reached the same conclusion through experience.
Deliberate Practice Path Tool — a tool for laying out a practice path based on weaknesses.
Shadowing Tool — a practice tool for repeating out loud under real conditions.

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.