Skip to content

Why Practice Scores Stop Improving, and What to Change

A flat practice score has four common causes, and they need different fixes. Here is how to read your own accuracy and timing record to tell them apart.

By Published Updated 6 min read

Reign Creative LLC makes IQ Test & Brain Training Games, which this article links to, so we have a commercial interest in it. How we write and check articles.

Key takeaways

  • Most of what people call a plateau is the normal shape of a practice curve. Gains on a repeated task are front-loaded and decelerate sharply, so a flat-looking stretch after early rapid improvement is expected rather than a failure.
  • A large survey of practice data found an exponential function fit individual learning series better than the power function in every unaveraged dataset examined, and that averaging across learners biased the result towards a power law.
  • A flat score has at least four distinguishable causes, and they need different fixes: the task has become too easy, you are trading accuracy for speed, you are still improving on a metric you are not looking at, or the practice has genuinely stopped challenging you.
  • Consistency and median answer time often keep moving after a score stops, so a single headline number can hide progress that is still happening.
  • Change one variable at a time. Raising difficulty, switching from single-operation to mixed rounds, and adding a timer are three different interventions, and doing all three at once tells you nothing about which worked.

Most plateaus are not plateaus. They are the ordinary shape of a practice curve, which improves fast at the start and then flattens hard. A score that leapt in week one and has barely moved in week five is behaving exactly as the research on practice curves predicts, and the useful question is not "why have I stopped improving" but "which of four different things is my record actually showing".

This article covers the shape first, because it reframes the problem, then the four readings and the specific change each one calls for.

Practice curves decelerate, and faster than you think

The standard textbook claim is the "power law of practice": performance improves as a power function of the number of trials, with each additional repetition buying less than the last.

That claim has been challenged in an interesting way. Heathcote, Brown and Mewhort's paper The Power Law repealed: The case for an Exponential Law of Practice, published in Psychonomic Bulletin & Review, argues that "the evidence for a Power Law is flawed because it is based on averaged data". The authors fitted power and exponential functions to "40 sets of data representing 7910 learning series from 475 subjects in 24 experiments" and reported that "the exponential function fit better than the power in all the unaveraged data sets", adding that "averaging produced a bias in favour of the power function".

You do not need to care which function wins to take the useful consequence. Both describe a curve that rises steeply and then flattens, and the exponential flattens sooner. More importantly, the paper's central methodological point applies directly to a personal practice record: curves built by averaging many learners look smoother and more gradual than any individual's curve does. Your own series is bumpier and flattens earlier than the tidy diagram in a productivity article, and that is not a defect in your practice.

So the baseline expectation for any repeated task is: large gains quickly, small gains slowly, then a long stretch where nothing headline-level moves. Reaching that stretch is a sign that the early work succeeded.

Four readings of a flat score

A score is a composite. To diagnose a plateau you need the components: accuracy, median answer time, and, if your record has one, a consistency figure. Hold the mode and the difficulty constant while you read them — comparing across modes is the single most common way to invent a plateau that is not there, and how to track practice progress without fooling yourself covers the rest of those traps.

Reading one: accuracy is at the top and answer time is flat. The task is finished. You are getting almost everything right, as fast as the method allows, and there is nothing left in this rung to learn. This is the clean case, and the fix is to change what is being asked.

Reading two: answer time is falling but accuracy is falling with it. You are not improving; you are moving along the speed-accuracy trade-off, buying pace with errors. The score may even rise for a while because speed bonuses reward it. The fix is the opposite of raising difficulty: go back to untimed rounds until accuracy is stable again, as set out in accuracy before speed.

Reading three: the headline number is flat but consistency is rising. You are still learning, and looking at the wrong column. The first thing practice usually removes is the occasional long stall rather than the ordinary answer, which shows up as a shrinking tail rather than a lower typical time. Response time distributions are "invariably positively skewed", as Lo and Andrews put it in To transform or not to transform, so the mean and the median tell different stories — reading your response time numbers separates them.

Reading four: everything is flat and has been for two or three weeks at a difficulty you are comfortable with. This is the genuine article. The session has become maintenance. That is a perfectly reasonable thing for a session to be, but if you want it to be something else you have to change the demand.

Performance is not the same as learning

Before changing anything, one caveat that cuts both ways.

Soderstrom and Bjork's review Learning Versus Performance is blunt about how weak a guide in-session performance is: what we observe during training "is performance, which is often an unreliable index of whether the relatively long-term changes that constitute learning have taken place". The review also notes that "improvements in performance can fail to yield significant learning", and that people "often mistakenly interpret their performance during acquisition as a reliable guide to long-term learning".

Applied here, that means two things. A flat score does not prove you have stopped learning — the stable part might be retention you would have lost without the practice. And a rising score does not prove you have learned anything durable, especially when it rises within a single session on the back of warm-up.

The check worth building in is a cold run: same mode, same difficulty, no warm-up, a couple of days after your last session. What survives a gap is a better estimate of what you have than what you can produce at the end of a hot session.

What to change, and in what order

IQ Test & Brain Training Games is our own app — Reign Creative publishes it — and the ladders below are the shipped ones, which makes them a concrete illustration of the general point: a plateau is fixed by changing a specific variable, not by adding effort.

  1. Step up one difficulty label. The arithmetic modes change what is actually generated as you climb — single digit, then wider ranges, then two-digit by two-digit, then mixed chains — and the memory modes have seven levels that change sequence length, grid size and how long items stay on screen. One rung at a time. If your accuracy at the new rung sits somewhere well short of perfect but not close to guessing, you have found the right one.
  2. Switch from a single-operation sprint to the mixed-operation mode. This adds the step of working out what is being asked, which is a different skill from executing it. Expect the score to fall; that is the task being harder, not you getting worse. Interleaved versus blocked practice explains why the mixed version feels worse while working better.
  3. Change the format rather than the content. Moving from an untimed accuracy round to a timed sprint, or to a survival format that ends on a mistake, changes the pressure without changing the arithmetic. This is the lightest of the three interventions and the easiest to reverse.
  4. Change the content entirely. If arithmetic has genuinely stopped moving, a memory mode asks something different of the same session slot. The IES practice guide on organising instruction and study recommends spacing learning over time and using quizzing to re-expose learners to key content, both of which favour rotation over grinding.

The rule that matters more than the list: change one variable and give it several sessions. If you raise difficulty, switch to mixed and add a timer in the same week, the resulting mess of numbers cannot tell you which change did anything.

What a plateau does not mean

It does not mean the practice was pointless, and it does not mean the next tier of effort will unlock something. Nothing about a flat arithmetic score says anything about general ability, and no amount of pushing through one will — the cornerstone article do brain training apps work sets out why claims of that kind do not hold up, and the whole brain training games hub stays inside the narrower and better-supported claim that you get better at what you practise.

Within that frame, a plateau is a useful signal. It tells you the current demand has been met, and that improvement from here is a matter of asking for something else. The app itself is described on the IQ Test & Brain Training Games app page, and more practice-design writing sits in the education and brain category.

Good to know

Frequently asked questions

Why has my practice score stopped improving?
Usually because the curve is supposed to flatten. Improvement on a repeated task is fastest at the start and decelerates steeply, so after a few weeks the same amount of practice buys a much smaller change than it did in week one. Before treating it as a problem, check whether accuracy, median answer time or consistency are still moving.
Is the famous power law of practice real?
It is contested. Heathcote, Brown and Mewhort's 2000 paper in Psychonomic Bulletin and Review fitted power and exponential functions to 40 datasets covering 7,910 learning series from 475 participants and found that the exponential fit better in all the unaveraged data, concluding that averaging produced a bias in favour of the power function. The practical point survives either way: individual learning curves decelerate, and they decelerate faster than averaged data suggests.
Should I raise the difficulty when I plateau?
Raise it if your accuracy is high and stable and your median answer time has stopped falling, which together mean the task no longer demands anything. Do not raise it if accuracy is already slipping, because that is a sign the current level is still unfinished. Change one thing and give it several sessions before judging.
Does a plateau mean I should stop practising?
No, but it does mean the current session is buying maintenance rather than improvement, and that is a legitimate thing to want. If you would rather keep improving, the fix is to change what the session asks for — a harder difficulty, a mixed problem set, or a different mode — rather than to do more of the same for longer.

Sources

  1. The Power Law repealed: The case for an Exponential Law of Practice (opens in a new tab)Psychonomic Bulletin & Review (author copy hosted by Northwestern University) — accessed
  2. Learning Versus Performance: An Integrative Review (opens in a new tab)Perspectives on Psychological Science (author copy, UCLA Bjork Learning and Forgetting Lab) — accessed
  3. Organizing Instruction and Study to Improve Student Learning (IES Practice Guide) (opens in a new tab)Institute of Education Sciences, What Works Clearinghouse, U.S. Department of Education — accessed
  4. To transform or not to transform: using generalized linear mixed models to analyse reaction time data (opens in a new tab)Frontiers in Psychology — accessed

About the author

— The independent app studio that builds the apps covered on this site and publishes every article here. On Google Play it publishes as Reign Collective Apps.

We use AI tools to help research, draft and edit our articles. Each article lists the sources its factual claims rely on, and details about our apps come from their live Google Play listings. No article on this site has been reviewed by an independent expert unless the reviewer is named on it.

Spotted an error? Tell us — how corrections are handled is set out in our editorial policy.

Keep reading