Articles · 4 September 2026
Sleep, caffeine and reaction time
Two nights of six hours in bed is not "a bit tired". Fourteen of them leaves you performing like someone who has not slept for two days — and rating yourself as fine.
Sleep restriction accumulates
The landmark study put 48 adults on 4, 6 or 8 hours in bed for fourteen consecutive nights, with a fourth group kept awake for three nights straight. Attention lapses in the 4-hour and 6-hour groups increased nearly linearly across the two weeks. The 8-hour group showed no cumulative deficit at all.
By the end, chronic restriction to six hours or fewer produced cognitive deficits equivalent to up to two nights of total sleep deprivation. The whole effect was explained by a single variable: cumulative wakefulness beyond about 15.8 hours per day.
You cannot feel it
The finding worth carrying around is that subjective sleepiness tracked the actual impairment poorly. Participants in the restricted groups rated themselves as only mildly sleepy while their performance kept degrading day after day.
That is why a reaction test is more useful than an opinion. If you want to know whether last week has caught up with you, measure it — your sense of how alert you are is the least reliable instrument in the room.
What sleep loss hits hardest
A meta-analysis of 70 studies and 147 cognitive tests found the effect of short-term sleep deprivation varied enormously by task. Reasoning accuracy barely moved. Lapses in simple attention showed by far the largest effect.
Simple attention and vigilance is precisely what a reaction-time test measures. Of all the cognitive abilities you could pick, this is the one most sensitive to how you slept — which makes it a good daily instrument and a poor one for comparing yourself against strangers.
Caffeine, measured
The most directly relevant study here tested 24 first-person-shooter esports players in a randomised, counterbalanced crossover against a water control, at two doses.
Reaction time improved at both. The nuance is the interesting part: caffeine improved reaction time and reactive tracking, but had no significant effect on static clicking accuracy. It makes you faster. It does not make you more precise at a task with no time pressure.
A meta-analysis of 31 trials across roughly 1,450 participants found significant improvements in both accuracy and reaction time, with a positive dose-response relationship for reaction time and larger effects above 200 milligrams.
| Measure | Water | 1 mg/kg (≈83 mg) | 3 mg/kg (≈250 mg) |
|---|---|---|---|
| Reaction time | 290.8 ms | 279.8 ms | 273.0 ms |
| Reactive tracking accuracy | 35.7% | 36.9% | 37.8% |
| Static clicking accuracy | 88.3% | 88.9% | 88.8% |
Time of day is genuinely unsettled
You will find confident advice about when to test. The literature does not support it. A systematic review of diurnal variation found studies placing peak simple reaction time in the evening, in the morning, and at mid-morning, with the review concluding that the answer depends on which cognitive function you are measuring.
What is reasonably consistent is a mid-afternoon dip in alertness. Beyond that, if you want to compare your scores over weeks, the useful discipline is testing at the same time of day rather than finding the optimal one.
Sources
- Van Dongen, Maislin, Mullington & Dinges (2003), Sleep — The cumulative cost of additional wakefulness. 48 adults, 14 nights at 4, 6 or 8 hours in bed; six hours or fewer produced deficits equivalent to up to two nights of total sleep deprivation; subjective sleepiness tracked impairment poorly.
- Lim & Dinges (2010), Psychological Bulletin — Meta-analysis of 70 articles and 147 cognitive tests. Largest effect was on lapses in simple attention; smallest on reasoning accuracy.
- Rogers, Trotter, Johnson, Desbrow & King (2024), Frontiers in Sports and Active Living — Caffeine improves the shooting performance and reaction time of first-person shooter esports players. 24 participants, randomised crossover against water.
- Systematic review of diurnal variation in cognitive performance (2024) — Finds conflicting peaks for simple reaction time across studies and concludes the peak depends on the function measured.