Why Test Scores Vary — What Reaction Time Really Measures

1. Where the milliseconds actually come from

When you tap and the page records a time, that number is a stack of delays, and only one layer of it is you:

• Display latency. A 60 Hz screen shows a new frame every 16.7 ms. The stimulus can appear anywhere inside that window, so up to 16.7 ms of your score is just waiting for the panel to refresh. At 144 Hz it is under 7 ms. Check yours with the refresh rate test.
• Input latency. A mouse polling at 125 Hz reports position every 8 ms; at 1000 Hz, every 1 ms. Touchscreens add their own sampling delay, usually larger. The polling rate test shows what yours is doing.
• Browser and system. Background tabs, extensions, a busy CPU, power-saving modes — all of it shifts timing by amounts that dwarf the differences people worry about.
• You. Recognition, decision, and movement. This is the part the test is nominally about, and on a typical setup it is a minority of the total.

That ordering is the whole point. If you change device between runs, you have measured the devices.

2. How to run a test so the number means something

One run is a coin flip. A few small habits turn a noisy number into something you can actually use:

• Run five, take the median. Not the best. The best score is your luckiest single trial, and luck does not repeat. The median is the value you would get on a typical attempt, which is the one that describes you.
• Fix your conditions and write them down. Same device, same browser, same input, roughly the same time of day. A score without its conditions is not comparable to anything, including your own earlier score.
• Warm up, then discard the warm-up. The first one or two trials are almost always slower — you are still learning the layout. Start counting after them.
• Close the other tabs. This sounds like superstition and is not; a video decoding in another tab moves reaction times by tens of milliseconds.
• Do not chase a number. If you retake a test twenty times and keep the best result, you have measured your patience, not your reflexes.

3. Which comparisons are fair, and which are meaningless

Not every comparison survives the noise above. Sorted from most trustworthy to least:

Fair — you versus yourself, same setup. Same device, same browser, medians of five. A change here is probably real. This is the only comparison we recommend anywhere on this site.

Usable with care — you versus yourself, different day. Sleep, caffeine, mood and time of day all move these numbers. A single bad day means nothing; a consistent drift over weeks might mean something.

Weak — you versus a friend on their device. Unless you are both on the same screen with the same input device, you are largely comparing hardware. If you want a real contest, take turns on one machine.

Meaningless — you versus a published average. Lab figures come from calibrated equipment with the display latency measured and subtracted. Your browser number contains all of it. The two are different quantities that happen to share a unit.

Because those measurements are not comparable, our tools show you your own numbers and mostly decline to rank you against a population. A percentile computed from incomparable measurements is a decoration, not information.

4. Difference scores: the trick that cancels the noise

There is one measurement design that sidesteps most of this, and it is worth knowing because it changes which results you should trust.

If a test measures you under two conditions and reports the difference, everything constant between them cancels out. Your screen adds the same 16.7 ms to both. Your mouse adds the same 8 ms to both. Subtract one from the other and those terms disappear, leaving the part that actually differed.

The Stroop test is built this way. It times you naming ink colors when the word matches, and again when the word fights the ink, and leads with the gap between them. That gap is far more stable across devices than either raw number, because the device contributed equally to both.

When you meet a test that reports a single raw speed — reaction time, clicks per second, double-click speed — read the number as "me plus my equipment". When you meet one that reports a difference, you are closer to reading just yourself. Neither is wrong; they answer different questions.

5. When a change is worth paying attention to

Most swings are noise. A few are not. Some rough guidance for telling them apart:

Probably noise: a single run that is worse than usual. A 10–20 % swing on one attempt. A bad result late at night or after a long day. Different result on a different device.

Worth a second look: the median of five is consistently worse across several sessions on the same setup. Your accuracy drops rather than your speed — on tests like aim training or typing, accuracy is usually the more stable signal, so a drop there is more meaningful than a slow day.

Not this page's business: if you are worried about attention, memory or coordination for health reasons, no browser test can answer that. These tools cannot detect or rule out any condition, and they are not built to. That is a conversation for a doctor, who has instruments that are calibrated and a history that a web page does not have.

The honest summary: these tests are good at showing you change in yourself under fixed conditions, and bad at almost everything else. Used that way they are genuinely useful — and used any other way the number is mostly about your hardware.