How much does reaction time really slow with age?
In a laboratory, very little before 60. In a browser, a lot more, and the gap widens after 50. The table lists the large studies that report a slope, with the conditions that shape it.
| Study | People | Ages | Setting | Slowing per year |
|---|---|---|---|---|
| Woods et al. 2015 | 1,469 | 18 to 65 | Lab, calibrated hardware | 0.55 ms |
| Krieg et al. 2001 | 4,896 | 20 to 59 | Lab | 0.34 ms |
| Dykiert et al. 2012 | 312 | 18 to 59 | Lab | 0.50 ms |
| Deary and Der 2005 | 1,930 | 16 to 63 | Lab, population sample | 1.70 ms |
| MindCrowd, Talboom et al. 2021 | 75,666 | 18 to 85 | Web browser, own device | 7 ms linear term, curve bends with age |
| UK Biobank, same paper | 158,249 | 40 to 70 | Clinic touchscreen, card-matching task | 3.40 ms |
The first four rows are the source of the "reaction time barely changes" claim. Woods and colleagues note that their 0.55 ms figure sits inside the 0.34 to 1.7 ms range of earlier laboratory work. At 0.55 ms per year, the average 60-year-old is about 22 ms behind the average 20-year-old. In the same study, one person's own trials swing about 40 ms around their own average. The age gap is smaller than that. On any given day, the day matters more than the birthday.
The MindCrowd row is the source of the "you lose 7 ms a year" claim. It is the biggest reaction-time dataset ever published, gathered through a web page where people pressed a key or tapped the screen when a pink sphere appeared. Its age effect is not a straight line. The paper fits a curve with a 7 ms per year linear term and a negative bend. In plain words, the slowing is gentle in the thirties and forties and steep later. Reading the 7 as a flat rate across all ages overstates it for younger adults and understates it for older ones.
The UK Biobank row is a third kind of number. More than 150,000 people did a card-matching task on a clinic touchscreen. That is a recognition task, not a simple reaction, and the slope came out at 3.4 ms per year between 40 and 70. More decision per trial means more slowing per year. The kind of task changes the answer as much as the age range does.
Why do browser datasets find ten times more slowing?
Four reasons, and none of them is that one side measured badly.
The lab studies stop before the steep part. Every laboratory row in the table ends at 59 to 65. Der and Deary (2006), using 7,130 adults from a national UK survey, found that simple reaction time shows little slowing until around 50 and then picks up. Choice reaction time, by contrast, slows across the whole adult range. A sample that runs to 85 includes thirty years of the curve that a sample ending at 65 leaves out. Fit a straight line through both and the longer sample will always look steeper.
Older participants were on slower devices. In MindCrowd, the people who took the test on a phone or tablet averaged 54 years old, well above the cohort, and their reaction times were longer. A browser test cannot tell an older nervous system from an older touchscreen, so some of the age slope in web data is a device slope wearing an age costume. Laboratory studies use one calibrated machine for everyone, which is why their slopes are smaller and also why their absolute numbers are lower. The accuracy guide covers how much a device adds.
Familiarity with the interface is not evenly spread. A five-trial web task rewards people who already know where the key is and how the page behaves. Younger adults on average have more of that practice, and five trials leave no room to catch up. Woods and colleagues ran forty trials per condition on a fixed setup, which washes the interface out of the number.
Who shows up is different. A 75-year-old who takes an online brain test is not a random 75-year-old, and neither is a 22-year-old laboratory volunteer. Both samples select, in different directions, and neither paper claims otherwise. This is a reason to read every age chart as "people like the ones who took part", not as a rule about you.
Put those together and the ten-fold gap stops being a contradiction. The lab number is the slope of the nervous system, measured on one machine, mostly before the slowing accelerates. The browser number is the slope of nervous system plus device plus practice plus who turned up, measured across the whole adult lifespan. A page that shows an age table without saying which of these it is has left out the part you need.
What should you expect at your age on this test?
Under 50, almost no age discount at all. On the laboratory scale the difference between 25 and 45 is about 11 ms, which is invisible inside the noise of five rounds. If your median is slower than a friend's, look at your screen, your mouse and your sleep before you look at your birth year. The scores guide lays out where each number falls.
From 50 on, expect a real but modest shift. The laboratory data say the slope increases after this point, and the browser data agree that the curve steepens. A healthy 65-year-old on a good setup can still post a median in the standard band on this test. What changes more than the median is the spread: older adults show more trial-to-trial variability, so the median of five rounds becomes more important, not less.
Past 70, the numbers in the table mostly stop applying, because the laboratory samples end there and the browser samples thin out. The honest statement is that reaction time keeps slowing and the rate varies a lot between people, with health, medication and sleep mattering more than the calendar. MindCrowd found that education, smoking, dizziness and the number of daily medications all moved the number alongside age.
This site does not ask your age, so it cannot show you an age-adjusted percentile, and it will not guess one from your score. The comparison that works at any age is the one with yourself: same device, same time of day, tab in front, median of five. If that median drifts over months while the conditions stay the same, you have measured something. If it jumps around between days, you have measured the days.
Two practical notes. First, the 0.55 ms figure quoted on the reaction time test page is the laboratory slope. It is the right one for the on-screen scale here, because the device delay grows with your hardware, not with your age. Second, if you want to know whether you have slowed, the only useful chart is your own history, and the test keeps it on your device for exactly that purpose.