Are online tests accurate?

A test that runs in a browser can only be as exact as the browser. What a good test does is take the time of your press from the moment the browser recorded it and not from the moment the page got round to reading it, and take the start of the signal from the moment the picture was drawn on the screen and not from the moment the page asked for it. The gap between the two stamps is your reaction. Done like this, the clock itself is not the weak point.

What the clock cannot see is the rest of the chain. Your screen takes a moment to light up, your mouse or touch screen takes a moment to report your press, and your computer takes a moment to pass it on. Together these add a delay that is hidden from the page and that differs from device to device.

The study that this site compares reaction times with measured the delay of its own set-up and reports a second, lower figure with that delay taken out^[1]. This site cannot measure your device, so it allows for the delays that robot tests of everyday devices found: 44.1 ms to 115.1 ms more than the study's equipment with a keyboard or mouse, and 39.8 ms to 52 ms with a touch screen^[2][3]. No study pressed a mouse, so a mouse gets the keyboard's figures. That is why a result is shown beside the study as a range and not as one exact rank, and why, on a computer, where the range is wide, it is only a rough guide.

A screen shows a new picture only so many times a second. A 60 Hz screen draws a new picture sixty times a second, and a 144 Hz screen more than twice as often, so on the slower screen a signal can appear a little later than the moment it was planned, and the next picture comes later still. A test can notice this, and the tests here do: if the screen stalls while a signal is showing, that try does not count and you do it again.

The refresh rate test shows how often your browser really gets a new picture, which is the first thing to know about a timing result from your own device.

• A mouse reports its position and its clicks a certain number of times a second. A 1000 Hz mouse reports far more often than a 125 Hz one, and a wireless or Bluetooth mouse may add a delay of its own. The polling rate test shows what yours does.
• A keyboard waits a very short time after a press to be sure it was a real press and not a flicker of the switch. That wait is part of every key press, and it differs between keyboards. The keyboard test shows whether each key registers once.
• A touch screen has to notice your finger, decide that it is a tap and pass it on, which takes a moment. For that reason a result from a phone is a different thing from a result from a mouse, and we say less about a result from a phone than about one from a mouse.

• The first try is for practice and does not count.
• A press that comes before the signal, or so soon after it that it must have been a guess, does not count, and you do that try again.
• If you switch to another tab or app during a try, that try does not count.
• Your result is the middle of several tries, so one lucky or unlucky try does not decide it.
• A run that looks as if a machine made it is kept but not compared with anyone.

The methodology page sets out the same rules and what is still unverified.

A questionnaire is scored by arithmetic, so the sum is exact. What is open to question is what the sum means. The answers are how you see yourself on that day, and they depend on how the statements are worded, whether you understand them as the authors meant, and how you feel. That is why a result moves a little when you take a test again, and why a longer questionnaire is steadier than a short one.

The Big Five test on this site uses the IPIP-NEO-120, a public-domain questionnaire of the psychologist John A. Johnson^[4], with many more statements than the quick tests. A result says where you sit between the two ends of each trait. Where a questionnaire has no published group for the language you use, the result says so and does not compare you with other people.

Tests that sort people into a few types are rougher still, because a person near the middle of a trait can land in either box on another day. A questionnaire is a mirror to think about yourself with. It is not a verdict, and it is no basis for choosing a person for a job.

1. Use the same device, and the same mouse or keyboard, every time you compare results.
2. Close other tabs and programs, so that the page is not interrupted in the middle of a try.
3. Take timing tests when you are rested, and at about the same time of day.
4. Take a timing test several times and look at the middle of your results, not the best one.
5. Read a questionnaire's statements slowly, answer for the person you usually are, and take it again later to see what stays the same.