Processing Speed and IQ: What Reaction Time Actually Reveals About Intelligence

Here is a task with no vocabulary, no arithmetic, no reasoning, and no knowledge of any kind: a light comes on, you press a button. A five-year-old can do it. It takes a fraction of a second.

It should tell you nothing whatsoever about intelligence. It tells you something.

Not much — the correlation is moderate, and this article is blunt about how moderate. But it's real, it has replicated across six decades and many countries, and it survives in a task deliberately stripped of everything IQ tests normally measure. That's a genuine puzzle, and the answer turns out not to be "smart people have fast brains." It's more interesting than that, and it connects to the g factor, to cognitive ageing, and — improbably — to how long you live.

Chronometric tasks are lab instruments, not IQ tests — nobody estimates intelligence from button-presses alone, because the correlations aren't nearly strong enough. For an actual estimate, our free IQ test uses a mixed verbal and non-verbal battery: ~20 minutes, instant, no email.

Galton's failure, Jensen's revival

Francis Galton had the idea first, in the 1880s, and got nowhere with it. He reasoned that if the mind is fed by the senses, then sharper senses and quicker responses should mean a sharper mind. He set up an anthropometric laboratory and measured reaction times on thousands of visitors. The relationships he found were weak enough to be discouraging, and the approach was abandoned. Intelligence testing went the other way — toward Binet's complex, school-like tasks, which worked.

The idea sat dormant for most of a century. Then Arthur Jensen revived it in the 1970s with better equipment and a specific theoretical claim: that individual differences in the speed of elementary information processing underlie individual differences in g. His apparatus — the "Jensen box" — had a home button and a semicircle of lights with buttons beside them. Its clever feature was separating decision time (releasing the home button after a light appears) from movement time (travelling to the target button). Only decision time was supposed to be cognitive.

The work was contentious immediately. Longstreth (1984) published a detailed methodological critique arguing that order effects, visual attention artefacts, and response-bias problems contaminated the results; Jensen replied that the critique contained errors of fact and that meta-analysis refuted its main verdicts. Some of these disputes were real and led to better paradigms — later work controlling those artefacts tended to produce higher RT–IQ correlations, not lower. But the episode is a reminder that this literature was built under adversarial conditions, which on balance strengthened it.

The correlations — and what they're worth

Here is the honest picture, with the variance each figure actually accounts for. That last column is the one most articles omit, and it's the one that matters:

Inspection time deserves a note since it's the strongest of these. It isn't a speed-of-response measure at all — you're shown a stimulus very briefly, then it's masked, and the question is how much exposure you need to judge it correctly. Motor speed is irrelevant. Grudnik and Kranzler's meta-analysis, pooling over 90 studies and more than 4,100 participants, found −0.30 raw and −0.51 after correcting for artefacts, with adults (−0.51) and children (−0.44) not significantly different, and visual and auditory versions comparable (−0.49 and −0.58).

The complexity gradient: more choices, stronger link

Notice the pattern in that table. Simple reaction time — one light, one button — correlates weakly, around −0.22. Add choices, and the correlation climbs: eight-choice RT reaches about −0.40; Deary's four-choice task hit −0.49 in a population sample. Sheppard and Vernon's review of fifty years of research reported a consistent trend for higher correlations in tasks with more choices.

The Hick paradigm formalises this: reaction time increases roughly linearly with the logarithm of the number of alternatives (the information load in bits). Jensen's theoretical bet was that the slope — how steeply your RT rises as information load increases — would be the purest index of processing capacity.

It mostly wasn't. Later work found that plain mean choice RT and simple RT relate more strongly to ability than the Hick slope does (Deary, 2000). The elegant theory-driven parameter underperformed the crude one — which is a nice illustration of how this field actually progressed, by having its best ideas fail informatively.

But the gradient itself is the finding that matters. The correlation tracks cognitive demand, not speed. The more a task requires discriminating, selecting, and deciding, the more it looks like intelligence. Which already suggests that "how fast is your brain" is not what's being measured.

The worst performance rule

Now the finding that should genuinely change how you think about this.

Give someone a few hundred reaction-time trials and you get a distribution — some fast, some slow. Sort them. Now ask: which end predicts IQ better?

Intuition says the fast end. If reaction time indexes maximum processing rate, a person's best trials show what they're capable of when everything lines up, and the slow ones are just noise — sneezes, distractions, blinks.

The data say the opposite. A person's slowest trials predict intelligence better than their fastest ones. This is the worst performance rule, and it's been replicated repeatedly. The relationship between RT and IQ grows stronger as you move up through the slower quantiles of someone's own distribution.

Relatedly, intraindividual variability — the standard deviation of a person's reaction times across trials — often predicts g about as well as, or better than, mean speed. Jensen concluded from more than 30 studies that RT variability generally has a larger negative correlation with g than mean RT does. (One caveat worth keeping: variability is a less reliable measure than a mean, so uncorrected comparisons understate it — a direct application of the reliability arithmetic in our piece on the standard error of measurement.)

Put those together and the interpretation shifts hard. If what mattered were peak speed, best trials would win. Instead, what predicts intelligence is the absence of lapses — how rarely the system drops out. Higher-ability people aren't reliably faster at their best. They're less erratic. Consistency, not velocity. That's a claim about attentional control far more than about nerve conduction.

It isn't speed — it's drift rate

The strongest modern reinterpretation comes from formal modelling, and it dissolves the phrase "mental speed" entirely.

A reaction time isn't one thing. Ratcliff's drift-diffusion model, fitted to a person's full RT distribution and error rates simultaneously, decomposes every response into three independent parameters:

Raw reaction time smears all three together. And when you pull them apart, intelligence tracks drift rate — the efficiency of evidence accumulation — not the other two (Schmiedek et al., 2007). Someone can be slow because they're cautious, which is a decision about how to take the test, not a fact about their cognition.

This is why the ageing literature is instructive: diffusion analyses have repeatedly found that older adults respond more conservatively — wider boundaries — and are slower in non-decision components, while drift rates sometimes show no age difference at all. Much of what looks like "cognitive slowing" is strategy and motor speed, not degraded processing.

So the honest formulation isn't "smart people are fast." It's: smart people extract usable signal from noisy input more efficiently. Speed is a downstream symptom of that, mixed in with caution and motor factors that have nothing to do with ability.

Processing speed on actual IQ tests

None of this is confined to the lab. Processing speed is a broad ability in its own right — Gs at Stratum II of the Cattell–Horn–Carroll hierarchy, sitting alongside fluid reasoning, crystallized knowledge, and working memory. It's one of the roughly eight to ten broad factors below g in the structure covered in our g factor explainer.

On the Wechsler scales this appears as the Processing Speed Index — simple, fast, high-volume clerical tasks like symbol coding and symbol search. Two things about it are worth knowing:

• It's the most trainable index. Studies re-administering the WAIS-IV at 3- and 6-month intervals found Processing Speed rose by roughly 9 points, the largest gain of any index and well above the ~7-point Full-Scale gain — because speeded tasks reward familiarity most. A rising PSI on retest is usually practice, not improvement.
• It's the most divergent index. A low PSI alongside a strong Verbal Comprehension is a common and clinically meaningful profile. It's precisely the pattern a single composite number erases — one reason a domain breakdown carries information the composite doesn't.

Speed and cognitive ageing

The most ambitious claim ever made for processing speed is Salthouse's (1996) processing-speed theory of adult age differences in cognition, one of the most-cited papers in the field. Its proposal: most age-related decline in fluid cognition is caused by slowing, through two mechanisms — the limited time mechanism (later operations don't get executed because earlier ones ate the clock) and the simultaneity mechanism (early products have decayed by the time later processing finishes).

The supporting evidence is substantial: speed measures share a great deal of age-related variance, and statistically controlling for speed strongly attenuates age effects on other cognitive measures.

But statistical mediation isn't causation, and the diffusion work above complicates the story considerably — if a chunk of measured "slowing" is caution and motor time rather than degraded processing, then speed is partly a proxy rather than a cause. Processing speed does follow a clear lifespan arc, rising through childhood and declining from young adulthood onward, which is part of the picture in IQ and age. Whether it's the engine of decline or one of its more visible symptoms remains open.

The result nobody expects

In 1988, Deary and Der measured IQ and reaction time in 898 Scots aged around 56. They followed them for 14 years. By 2002, 185 had died.

Higher IQ predicted living longer — already known, and it survived adjustment for education, occupational social class, and smoking. But reaction time was the stronger predictor. And the finding that makes the paper famous: once reaction time was controlled for, IQ's association with mortality was no longer significant.

Sit with that. Whatever it is about lower IQ that goes with dying earlier, a button-press task captured it more directly than the IQ test did.

The interpretation Deary favours is the system integrity hypothesis: reaction time isn't causing anything. It's a sensitive readout of how well the whole organism is functioning — a bodily integrity indicator that happens to show up in both cognitive performance and survival. On this view, low IQ doesn't shorten life; both are downstream of the same underlying system quality. Later work found reaction time measures matched IQ in predicting cardiovascular, cancer, and respiratory mortality specifically, with effect sizes more similar than their mutual correlation would predict — and RT variability predicts dementia and death independently of mean speed.

This founded the field of cognitive epidemiology, and it's the strongest evidence going that chronometric tasks tap something biologically real rather than a testing artefact. It also connects directly to the material in IQ and longevity.

The honest verdict

• The correlation is real and replicated — six decades, many countries, many paradigms, and it survived a hostile methodological audit that improved it.
• The correlation is moderate. Around −0.3 typically, −0.5 at the very best. Five to twenty-six percent of variance. It says nothing about any individual, and a fast reflex is not a credential.
• It isn't really speed. Worst trials beat best trials; variability rivals mean RT; and formal modelling says intelligence tracks drift rate, not velocity. Consistency and signal quality, not a fast clock.
• It might be about the body. The mortality findings suggest chronometric performance indexes general system integrity — which would explain why such a trivial task predicts so much.
• Causation is unresolved. Whether efficient processing produces intelligence, intelligence produces efficient performance, or a third factor produces both is genuinely open.

What makes this literature worth your attention isn't that it lets anyone measure your IQ from a button-press — it emphatically doesn't. It's that a task with no content at all keeps carrying signal, which tells you something real about what IQ is measuring underneath the vocabulary and the matrices. And it's a standing caution against the brain-training pitch: making yourself faster at a speeded task is easy and mostly meaningless, which is exactly the trap covered in brain training and IQ.

Frequently asked questions

Related reading

← Browse all articles on the blog

References

1. Galton, F. (1883). Inquiries into Human Faculty and Its Development. Macmillan.
2. Hick, W. E. (1952). On the rate of gain of information. Quarterly Journal of Experimental Psychology, 4(1), 11–26.
3. Ratcliff, R. (1978). A theory of memory retrieval. Psychological Review, 85(2), 59–108.
4. Jensen, A. R., & Munro, E. (1979). Reaction time, movement time, and intelligence. Intelligence, 3(2), 121–126.
5. Longstreth, L. E. (1984). Jensen's reaction-time investigations of intelligence: A critique. Intelligence, 8(2), 139–160.
6. Jensen, A. R. (1987). Individual differences in the Hick paradigm. In P. A. Vernon (Ed.), Speed of Information-Processing and Intelligence (pp. 101–175). Ablex.
7. Kranzler, J. H., & Jensen, A. R. (1989). Inspection time and intelligence: A meta-analysis. Intelligence, 13(4), 329–347.
8. Kail, R., & Salthouse, T. A. (1994). Processing speed as a mental capacity. Acta Psychologica, 86(2–3), 199–225.
9. Salthouse, T. A. (1996). The processing-speed theory of adult age differences in cognition. Psychological Review, 103(3), 403–428.
10. Deary, I. J., & Stough, C. (1996). Intelligence and inspection time: Achievements, prospects, and problems. American Psychologist, 51(6), 599–608.
11. Deary, I. J. (2000). Looking Down on Human Intelligence: From Psychometrics to the Brain. Oxford University Press.
12. Deary, I. J., Der, G., & Ford, G. (2001). Reaction times and intelligence differences: A population-based cohort study. Intelligence, 29(5), 389–399.
13. Grudnik, J. L., & Kranzler, J. H. (2001). Meta-analysis of the relationship between intelligence and inspection time. Intelligence, 29(6), 523–535.
14. Deary, I. J., & Der, G. (2005). Reaction time explains IQ's association with death. Psychological Science, 16(1), 64–69.
15. Shipley, B. A., Der, G., Taylor, M. D., & Deary, I. J. (2006). Cognition and all-cause mortality across the entire adult age range: Health and lifestyle survey. Psychosomatic Medicine, 68(1), 17–24.
16. Schmiedek, F., Oberauer, K., Wilhelm, O., Süß, H.-M., & Wittmann, W. W. (2007). Individual differences in components of reaction time distributions and their relations to working memory and intelligence. Journal of Experimental Psychology: General, 136(3), 414–429.
17. Sheppard, L. D., & Vernon, P. A. (2008). Intelligence and speed of information-processing: A review of 50 years of research. Personality and Individual Differences, 44(3), 535–551.
18. Ratcliff, R., Schmiedek, F., & McKoon, G. (2008). A diffusion model explanation of the worst performance rule for reaction time and IQ. Intelligence, 36(1), 10–17.
19. Estevis, E., Basso, M. R., & Combs, D. (2012). Effects of practice on the Wechsler Adult Intelligence Scale-IV across 3- and 6-month intervals. The Clinical Neuropsychologist, 26(2), 239–254.
20. Doebler, P., & Scheffler, B. (2016). The relationship of choice reaction time variability and intelligence: A meta-analysis. Learning and Individual Differences, 52, 157–166.
21. Der, G., & Deary, I. J. (2018). Reaction times match IQ for major causes of mortality: Evidence from a population based prospective cohort study. Intelligence, 69, 134–145.