Reaction Time Statistics

The distribution is not a bell curve, and it never will be

Reaction times are bounded below and unbounded above. There is a floor set by how fast a signal can physically get from your eye to your hand, nobody beats it, and yet any round can be arbitrarily slow if attention wanders for a moment. That asymmetry produces a curve with a steep left edge and a long right tail.

It means the mean sits above the peak, and the more careless attempts a sample contains the further it drifts. Quoting an average reaction time without saying so overstates what a typical attempt looks like, which is why the median is given first here and the mean second.

It also means the interesting comparisons live at the left edge, where the distribution is steep. A twenty millisecond improvement near the peak moves you several percentiles; the same twenty milliseconds out in the tail barely moves you at all.

What the percentile beside your score is computed against

Attempts in the same input mode, and nothing else. A mouse click is compared to other mouse clicks, a tap to other taps. Pooling them would let the slower hardware in the sample flatter every desktop visitor and penalise every phone visitor, silently.

Within a mode, the position is interpolated inside its bucket rather than rounded to it. Without that, two visitors nine milliseconds apart would be told they were in the same percentile and the first attempt in a bucket would jump several points for no reason.

Below two hundred attempts in a mode, no curve is drawn and no percentile is offered. A distribution built from a few dozen results is noise wearing the costume of a statistic, and the honest thing to display is the attempt count and a note that it is not enough yet.

Why these numbers should not be compared to a laboratory study

Published reaction time research typically uses a dedicated response box on known hardware, with the display latency characterised in advance. A browser on unknown equipment cannot match that, and every figure on this site includes the visitor's display and input device as well as the visitor.

The offset is not random noise that averages out. It is a systematic addition of roughly ten to twenty milliseconds depending on the equipment, always in the same direction, so a browser sample sits shifted relative to a laboratory one no matter how many attempts it contains.

This site reports the offset it can estimate rather than correcting for it. Subtracting an estimate would produce a number that looks more authoritative and is less checkable, and the whole point of publishing the method is that a reader can decide what the figure is worth.

This is a measurement toy, not a medical or psychological test. No score here says anything about your health, attention or cognition, and it is not a screening tool of any kind.

Questions

What is the average human reaction time?

Browser click tests generally centre somewhere in the 200s of milliseconds, but the average is a poor summary of a skewed distribution — the median and the percentile breakpoints on this page describe it far better.

Why is the median lower than the mean here?

Because the distribution has a long slow tail and a hard fast floor. A few very slow attempts drag the mean upward while the median stays near the peak.

Does this data say anything about my health?

No. It is a general-interest measurement of a browser test on unknown hardware, not a screening tool, and nothing on this site should be read as an assessment of any medical condition.