Typing Research: What Keystroke Data Reveals

Keystroke research has answered some questions about typing speed well and left others open. A 2018 study of over 168,000 volunteers ¹ finds that fast typists differ from slow ones in how their keys overlap, how often they correct errors and how many fingers they use. Typing rhythm can help confirm who is typing: the three best detectors on one benchmark reached equal error rates of 9.6 to 10.2 percent ². Small studies link changes in typing timing to Parkinson's disease ³. No study we found tests typing speed against IQ, and the claim that pianists or gamers type faster has no controlled test on an ordinary keyboard.

A 2018 online typing test recorded over 136 million keystrokes from 168,000 people

The study is a 2018 analysis by Dhakal and colleagues. It drew on over 168,000 volunteers who took an online typing test ¹, and the dataset holds over 136 million keystrokes ⁴. The average participant typed 51.56 words per minute in that dataset ⁵.

The volunteers were self-selected, and most were younger people from the US who were interested in testing and improving their typing ⁶. Read the averages as describing people who chose to take a typing test, not the whole population.

The study splits typists into groups by speed. Fast typists are those faster than 90 percent of participants, above about 78 words per minute ⁷. Slow typists are the slowest 10 percent, below about 26 words per minute ⁸. The fastest typists reached 120 words per minute or more ⁹. The findings below compare these groups.

Fast typists overlap keystrokes, leave fewer errors uncorrected and use more fingers

Fast typists differ from slow ones in rollover, which means pressing the next key before the previous one is released ( rollover). Slow typists use almost none ¹⁰. Most fast typists use rollover for 40 to 70 percent of their keystrokes ¹⁰. Across the online typists, the share of rollover correlates with speed at r = 0.73 ¹¹.

Timing shows the same split. Fast typists averaged about 120 milliseconds between keystrokes, while slow typists averaged over 480 milliseconds ¹². That makes the slow gaps at least four times longer ¹³.

Slow typists also leave more errors in the final text. The study finds that they leave significantly more errors uncorrected, which the authors suggest could mean they are less able to detect their own mistakes ¹⁴. The largest difference between the groups is in substitution errors, where a wrong character replaces the intended one ¹⁵. The authors hypothesize that substitution errors are common when typists are less consistent and have a poor mental picture of where their fingers are ¹⁶. Across participants, faster typing goes with fewer error corrections, at r = -0.36 ¹⁷.

Fast typists also use more fingers. They used about 8.4 fingers on average, against 5.3 for slow typists ¹⁸. Trained typists show the same pattern, with 8 fingers against 6.5 for untrained typists ¹⁹.

These measures describe the online sample, not a cause. Learning rollover is not shown to make anyone faster. For the everyday version of the speed question, see the guide to how typing speed is measured, which explains words per minute and keystrokes per character.

Skilled typists prepare upcoming keystrokes and keep one finger for each letter

A motion-capture study of typing in a laboratory found three predictors of high performance, including that a letter is always pressed by the same finger and that upcoming keystrokes are actively prepared ²⁰. Preparing the next keystroke early fits with rollover, because a key pressed before the last one is released is one way to get ahead of the text.

Consistent finger use is the habit that touch typing teaches. Our touch typing guide covers how that habit is built, and the guide to how to type faster covers which practices change speed and which do not.

Typing rhythm can verify a typist, with equal error rates near 10 percent on a benchmark

The timing of your keystrokes is called keystroke dynamics. One benchmark for it comes from Carnegie Mellon. Its researchers recorded 51 people each typing one password 400 times ²¹. Of the detectors tested on that data, the three best reached equal error rates of 9.6 to 10.2 percent ². An equal error rate is the point where the rate of wrongly rejected owners equals the rate of wrongly accepted impostors. Lower is better.

The researchers released the data along with the test procedure, so other groups can compare methods on the same set ²².

The benchmark answers a narrower question than a reader might expect. It tests whether a typing sample matches the person claiming to type it, and it used one fixed password typed in a lab. It does not show how well typing identifies someone among many people, or how well it works on free text. The direct answer is on the question about whether typing rhythm can identify you.

Typing timing separates early Parkinson's from controls in small studies

Parkinson's disease affects fine motor control, and that can show up in how a person types. A 2016 study recorded 42 people with early Parkinson's and 43 controls ³ as they typed a folk tale on their own computers, as they normally would at home ²³. Analysis of key hold times separated the groups with an AUC of 0.81 ³. The AUC is a score from 0.5, which is chance, to 1.0, which is perfect separation. The authors reported that this was comparable to or better than two clinical tapping tests, which scored 0.75 and 0.61 ²⁴.

A 2017 study used a computer keyboard with 103 people, 32 of them with mild Parkinson's, typing over an extended period ²⁵. In that study, a machine-learning ensemble separated early-stage Parkinson's from controls with an AUC of 0.98 ²⁶. A 2018 study used touchscreen smartphone typing in matched groups of 18 early patients and 15 controls, and reached an AUC of 0.92 ²⁷.

These numbers are high, but they are not a diagnosis. Each study used its own small group, so the scores cannot be compared as if they came from one test. Parkinson's also affects speech: as many as 89 percent of people with the disease develop speech disorders ²⁸. Our guide to speech recognition for speech impairments covers that side of the condition. The direct answer is on the question about typing patterns and Parkinson's.

No study has tested typing speed against IQ, and trained typists are only slightly faster

We found no study that measures typing speed against IQ. The nearest large study is a 2026 analysis of 10,613 adults ²⁹. Faster typing was generally associated with better cognitive functioning across domains. After adjusting for demographics, those links weakened but generally kept the same direction ³⁰. The study used cognitive tests, not IQ scores ²⁹, and an association across people is not a cause. The question on whether typing speed correlates with IQ goes through this in more detail.

Training shows a small link to speed. Trained typists, meaning those who reported having had typing training, averaged about 5 words per minute faster than untrained typists ³¹. The authors describe that as a relatively small difference ³¹, and they note that untrained typists can be as fast as trained ones ³². Trained typists do leave fewer errors uncorrected, at 1.02 percent against 1.23 percent ³³.

The typing test used in the 2026 study is also stable. A one-sentence test, taken again two years later, gave an intraclass correlation of 0.79 for computer typing speed ³⁴. That means a person's score stayed fairly consistent over the two years.

Pianists and gamers: no controlled test on an ordinary keyboard

We found no controlled study that compares pianists or gamers on QWERTY typing speed. The closest work is a 2014 study of PianoText, a text-entry layout for piano keyboards ³⁵. It reached an achievable rate of over 80 words per minute ³⁵, and the authors say that rate parallels the rates of expert QWERTY typists ³⁶.

That result shows what a piano keyboard can do with a new layout. It does not show that pianists type faster on a standard keyboard, because the layout was new and the study used a small number of trained participants ³⁶. For the direct answer, see the question on whether pianists type faster.

How to read a keystroke study

Three checks help you judge a typing study before you rely on its numbers. First, look at who typed. The large online dataset is self-selected, so its averages describe people who chose a typing test, and a Parkinson's study of 18 patients describes a much smaller group ²⁷. Second, look at what was measured. Speed, error rate, timing and identity are different measures, and a result about one does not transfer to another. The benchmark's equal error rate is about telling a typist apart from an impostor, not about how fast someone types.

Third, look at the kind of number. A correlation such as r = 0.73 ¹¹ or r = -0.36 ¹⁷ says two things move together across people. It does not say that changing one would change the other. The same holds for the cognitive links in the 2026 study, which shrank once demographics were taken into account. An AUC is a separation score, not the chance that a single person has a condition. When a result sits on a few dozen people, treat the exact figure as a rough guide and wait for a larger, independent study before relying on it.

What the research cannot yet show

Four questions remain open, and each needs a different kind of study:

• Typing speed and IQ. No study we found measures them against each other. The 2026 study links typing to cognitive tests, and that is a different measure.
• Pianists and gamers on QWERTY. No controlled test compares them on an ordinary keyboard.
• Parkinson's detection. The reported accuracy comes from small groups, and each study used its own participants. Independent replication in larger groups is the missing step.
• Identity on free text. The benchmark used one fixed password typed in a lab, so free-text accuracy is not established by it.

Open data lets readers recheck one of these benchmarks

Some results here can be checked directly. The Carnegie Mellon benchmark is shared, so its equal error rates can be recomputed from the same data ²². Every study result in this guide is cited to a fact page, and each fact page links to the study it comes from, where you can read the sentence the number depends on.

For the basics behind these numbers, read our guide to average typing speed, the guide to good typing speed and the guide to keyboard layouts. The full set of guides is in typing and writing speed.