Processing Speed as a Concept, and How It Gets Measured

Processing speed is a construct, not a stopwatch reading. It refers to how quickly someone carries out simple cognitive operations that are not themselves hard — comparing two symbols, scanning a line for a target, matching a code to a digit. Because each of those takes a fraction of a second, nobody estimates it from one timing. It is inferred from performance across many deliberately easy items under time pressure, and that design choice is the whole reason the number means anything.

How the construct gets operationalised

The trick in measuring speed is to remove everything except speed. If the items are hard, the score reflects whether you knew the answer. If the items are trivial and numerous and the clock is short, the score reflects how fast you can get through them.

So the classic speeded tasks look almost insultingly simple: decide whether a target symbol appears in a row of symbols; write the matching code under each of a long list of digits; cross out every instance of one letter. Almost everyone gets almost every item right. The variance lives in throughput.

Two consequences follow, and both are routinely missed.

First, a speeded task is only a speed measure if accuracy is near ceiling. The moment people start making real errors, the score becomes a blend of speed and something else.

Second, the items have to be uninteresting. Any item that invites a strategy stops being a measure of speed and becomes a measure of whether you found the strategy.

Why a response time in a game is not that

An arithmetic question fails both conditions, and it is worth being specific about why.

When an app records that you answered a multiplication question in 2.4 seconds, that interval contains: reading and parsing the problem; deciding whether to retrieve the fact or compute it; carrying out the computation if you compute; entering digits on a keypad; and, underneath all of it, whatever speed-accuracy setting you happened to adopt on that run. Those are not separable from the outside, and only some of them are cognitive.

The speed-accuracy problem is the deepest of the three. Richard Heitz's review of the speed-accuracy trade-off documents it as a fundamental property of decision-making across species: faster responses involve less accumulated evidence and hence less informed decisions. His methodological warning applies directly to any app statistic — group means obtained at a single criterion, he writes, provide only a snapshot of performance that conflates decision strategy with the nature of the task. Translated: a time only means something when you know where the person was sitting on the speed-accuracy curve, and an app cannot know that.

This is why our position on our own numbers is narrow. IQ Test & Brain Training Games is our own app, built by Reign Creative. It records average response time, median response time, fastest correct answer and a consistency figure for each run, and it ships short-clock formats — a 30-second lightning round, 60-second sprints, and a speed ladder whose clock shrinks as you climb. None of that is a processing-speed measurement. Those numbers describe your play in one mode at one difficulty. They are game statistics, and treating them as anything else would be a claim we cannot support. The app also has a reaction time test, which its listing says measures speed and consistency, and an IQ-style test that includes processing speed; the listing describes those results as estimates based on app activities, not clinical or professional assessments, and the cautions below apply to them as much as to the drill times.

The shape of your own timing data

If you do want to read your own response times usefully, the first thing to understand is that they are not normally distributed.

Lo and Andrews, writing on the analysis of reaction-time data, put it plainly: the distribution of response times obtained in simple decision tasks is invariably positively skewed. Most trials cluster tightly near the fast end; a few run very long. They also note the consequence — that a few extreme outliers might dominate the outcome, partially or even completely obscuring the main trends characterising the majority of data points.

Here is what that does to a summary statistic. Take ten answers timed at 1.2, 1.3, 1.1, 1.4, 1.2, 1.5, 1.3, 1.2, 1.4 and 9.0 seconds — nine ordinary answers and one where you stopped to think.

• The mean is 20.6 ÷ 10 = 2.06 seconds.
• The median is the midpoint of the sorted list, which lands between 1.3 and 1.3: 1.30 seconds.

The mean has been dragged nearly sixty per cent above anything you actually did on nine of the ten trials. The median describes your typical answer. This is why a practice record that reports both is more informative than one reporting only an average, and why a jump in your mean with a stable median usually means one long pause rather than a bad session.

Consistency is the third useful number, and it moves before speed does. Early in learning a mode, the spread between your fastest and slowest correct answers is wide because some items are retrieved and others are computed. As more items become retrieval, the spread narrows. You will usually notice your times getting less variable before you notice them getting shorter.