Accuracy First or Speed First? Getting the Order Right

Train accuracy to a stable level first, then push speed. The order matters because speed and accuracy trade against each other: going faster means deciding on less information, so adding a clock to a method you cannot yet execute reliably simply generates wrong answers at a higher rate. Worse, it rehearses the shortcuts you take under pressure, and those are hard to unlearn.

The reverse order has an obvious appeal — a timer is more exciting than a careful round — and it is the default most people fall into. This article covers why it fails, what the trade-off actually is, and how to run the test on yourself.

The trade-off is a property of the task, not a weakness

The relationship between response speed and error rate is one of the most stable findings in the study of decision making. Richard Heitz's review The speed-accuracy tradeoff: history, physiology, methodology, and behavior, published in Frontiers in Neuroscience in 2014, puts it this way: "the tendency for decision speed to covary with decision accuracy seems an inescapable property of choice behavior." The mechanism it describes is straightforward — "faster responses entail less accumulated evidence, and hence less informed decisions" — and the pattern shows up across species, not just in people trying to answer arithmetic questions on a phone.

Two things follow that are worth stating plainly.

First, a drop in accuracy when you speed up is not evidence that you are careless. It is the expected shape of the curve, and everyone sits on it somewhere.

Second, the curve can move. Practising the method until it costs less to run shifts the whole relationship, so that the same accuracy is available at a shorter time. That shift is what people mean by fluency, and it is the actual goal. Simply choosing a point further along the existing curve — going faster and accepting more errors — is not improvement, even though it produces a better-looking time.

Why accuracy comes first

The reason for the ordering is that the two directions are not symmetrical.

If you build accuracy first, speed arrives partly on its own. A method you can run without hesitation takes less time than one you have to reconstruct, so median answer time falls as a by-product of the method becoming automatic.

If you build speed first, accuracy does not arrive on its own. What arrives instead is a set of coping strategies invented under time pressure: abandoning the borrow and guessing the digit, rounding without correcting, answering from the shape of the numbers rather than the arithmetic. Those strategies work often enough to score points, which is exactly what makes them stick.

The IES practice guide Assisting Students Struggling with Mathematics recommends that interventions "devote about 10 minutes in each session to building fluent retrieval of basic arithmetic facts." The word doing the work there is retrieval. Fluency in that sense means the right answer arriving without computation — which is a statement about correctness first and speed second, and which cannot be trained by hurrying an unreliable calculation.

There is a second-order reason too. National Numeracy defines numeracy as "having the confidence to use basic maths at work and in everyday life", and argues that "it is confidence that unlocks our numeracy skills". A practice diet built entirely on rounds you get wrong at speed is a poor way to build that. The point is not that practice should feel easy; it is that the difficulty should come from the material, not from a clock you were not ready for.

Running the test on yourself

The useful thing about a drill app is that it reports both numbers from the same run, so the trade-off stops being theoretical.

IQ Test & Brain Training Games is our own app — Reign Creative publishes it — and it ships both poles of this deliberately. There is an untimed accuracy round of ten questions with feedback, an explanation and an optional hint on each item; there are 60-second sprints on each operation; and there is a no-mistake format where a single wrong answer ends the run. Every run reports how many you got right, how many wrong, your accuracy, and your answer times.

The experiment takes two sessions.

Session one, untimed. Pick one operation and one difficulty. Run the untimed accuracy round twice. Note your accuracy and roughly how long the questions took. This is your method working without pressure.

Session two, timed. Same operation, same difficulty. Run two 60-second sprints. Note accuracy and median answer time again.

Then read the pair:

• Accuracy held, median time fell. Fluency. This is the result you want, and it means you are ready to push the difficulty rather than the clock.
• Accuracy fell noticeably, median time fell. You moved along the curve rather than shifting it. Go back to the untimed format at this difficulty.
• Accuracy fell and median time barely moved. The clock is costing you attention without buying anything. Almost always a sign the difficulty is too high for a timed format yet.

Use the median rather than the average answer time for this comparison — a single long think distorts an average badly, and reading your response time numbers explains why the two diverge.