Peer-reviewed research

Observations on Typing from 136 Million Keystrokes

Vivek Dhakal et al., CHI 2018, ACM Conference on Human Factors in Computing Systems, 2018 · DOI 10.1145/3173574.3174220

Read the original

Read on October 16, 2026

What it is: Analyses 136 million keystrokes from an online typing test taken by about 168,000 volunteers. It reports distributions of words per minute, error rate and keystroke timing for keyboard typing.
How much weight it carries: Peer-reviewed CHI paper with a very large sample. The participants were self-selected volunteers who took a test online, so the averages are higher than population averages. The paper says so in its own sampling discussion.

Facts taken from this source

The Dhakal and colleagues study analysed typing data from over 168,000 volunteer participants.
“We have reported extensive statistical analysis for over 168,000 volunteer participants” (Section 'Summary')
The online typing study analyzed 136,857,600 keystrokes from 168,960 participants, most of whom finished all 15 sentences.
“The final dataset includes 136,857,600 keystrokes from 168,960 participants” (Results)
The volunteers in the online typing study had a mean age of 24.5 years, with 75 percent aged 11 to 30.
“Demographics Result Remark Females 52.7% Rest preferred not to specify Males 41.5% Age: mean 24.5” (Table 1: Demographics)
In a study of 136 million keystrokes from an online typing test, the average typing speed of participants was 51.56 words per minute.
“The average WPM value for participants is 51.56 (SD = 20.2).” (Results: Words per minute)
The keyboard study compares its measures by fast versus slow typists and by trained versus untrained typists.
“It shows mean and SD for each measure and compares them for fast versus slow and trained versus untrained typists.” (Results, Table 3 text)
The distribution of typing speeds in the online dataset is not bell-shaped; it is slightly positively skewed.
“The distribution is not a normal-form one; it shows slight positive skewness” (Section 'Words per minute')
Clustering the keystroke data found that most users can be divided into eight groups of typists that differ in performance and accuracy.
“most users can be divided into eight groups of typists that differ in performance, accuracy” (Abstract)
In a study of 136 million keystrokes from an online typing test, typists faster than 90 percent of participants were those above about 78 words per minute.
“Fast Typists are the typists with speeds higher than that of 90% of participants (above approx. 78 WPM in our data)” (Section 'Typist Groups')
Fast typists used about 8.4 fingers on average, while slow typists used about 5.3.
“we observe fast typists to use more ﬁngers than slow typists (averaging 8.4 versus 5.3)” (Results, Hand and finger use)
Fast typists had an average gap of about 120 milliseconds between keystrokes, while slow typists had gaps of over 480 milliseconds.
“For fast typists, the average IKI is ∼120 ms, with a standard deviation of only 11 ms, while slow typists have an IKI of over 480 ms, sometimes as high as 900 ms” (Results, Words per minute and inter-key intervals)
Across participants, faster typing was negatively correlated with uncorrected errors (r = -0.21) and with error corrections (r = -0.36).
“We ﬁnd a negative correlation between WPM and uncorrected errors ( r = −0.21), error corrections ( r = −0.36), and KSPC (r = −0.4)” (Results, Errors and error-correction behaviour)
The fastest typists in the online study reached 120 words per minute or more.
“The fastest typists in the dataset reach speeds of even 120 WPM or more.” (Performance Measures: Words per minute)
The fastest typists in the online typing dataset of about 168,000 volunteers reached 120 words per minute or more.
“The fastest typists in the dataset reach speeds of even 120 WPM or more.” (Section 'Words per minute')
In the online keyboard study, 52.7 percent of volunteers were female and 41.5 percent were male.
“Females 52.7% Rest preferred not to specify Males 41.5%” (Demographics table, first column)
The study found that letter pairs typed by different hands or fingers predict typing speed better than letter repetitions do.
“more predictive of typing speed than, for example, letter repetitions” (Abstract)
In the online study, typing speed varied widely, with a standard deviation of 20.2 words per minute around the 51.56 mean.
“The average WPM value for participants is 51.56 (SD = 20.2).” (Performance Measures: Words per minute)
Most typists in the study performed at least a small share of their keystrokes with rollover.
“most typists performing at least a small percentage of keystrokes by means of rollover” (Results)
Across the study's typists, the share of keystrokes typed with rollover correlated highly with typing performance, r = 0.73.
“Rollover ratio is found to have a high correlation with performance (r = 0.73, p < 0.001).” (Results)
Across the online typists, the share of keystrokes typed with rollover correlated with typing speed at r = 0.73, a high correlation.
“Rollover ratio is found to have a high correlation with performance (r = 0.73, p < 0.001).” (Results, Rollover Behaviour)
In the 136 million keystroke study, slow typists used almost no rollover, while most fast typists used rollover on 40 to 70 percent of their keystrokes.
“Slow typists use nearly no rollover, whereas the majority of fast typists use rollover for 40–70% of keystrokes.” (Abstract)
In the online dataset of 168,000 typists, slow typists used almost no rollover, while most fast typists used rollover for 40 to 70 percent of their keystrokes.
“While slow typists use almost no rollover, the majority of fast typists use rollover for 40–70% of keypresses.” (Results, Rollover Behaviour)
The online typing test's users were mainly younger people from the US who wanted to test and improve their typing, so the sample is not representative of all typists.
“The user base of the company is composed mainly of younger people from the US who are interested in testing and improving their typing skills.” (Method, participants)
The online typing study reports a skewness of 0.513 for its words per minute distribution, a positive skew with a longer tail of faster typists.
“The distribution has a skewness of 0.513” (Words per minute results)
In the same online study, the slowest 10 percent of typists were those below about 26 words per minute.
“Slow Typists are those whose performance is among the slowest 10% (less than approx. 26 WPM in our data)” (Section 'Typist Groups')
Slow typists left significantly more errors uncorrected than fast typists, which the authors suggest may mean they detect their own mistakes less well.
“Slow typists leave signiﬁcantly more errors uncorrected, which could indicate that they are less able to detect their mistakes.” (Results, Errors and error-correction behaviour)
The largest difference between fast and slow typists was in substitution error rate, where a wrong character replaces the intended one.
“Comparing fast with slow typists, we can see that there is a particularly large difference in substitution error rate” (Results, Errors and error-correction behaviour)
The authors hypothesize that substitution errors are common when typists are less consistent and have a poor mental picture of where their fingers are.
“We hypothesise that substitution errors are common when participants are less consistent in typing and have a poor mental representation of the ﬁngers’ position.” (Results, Errors and error-correction behaviour)
Each participant in the 2018 online study transcribed 15 English sentences.
“The task was to transcribe 15 English sentences.” (Task and Procedure)
The sample was people interested in testing and developing their typing skills, so it is not a random sample of the population.
“This sample represents people interested in testing and developing their typing skills.” (Participants)
Trained typists in the online sample averaged about 5 words per minute faster than untrained typists, a small difference next to the spread of about 20 WPM within each group.
“Trained typists in our sample type, on average, 5 WPM faster than untrained typists, with a similar SD, of about 20 WPM, for a relatively small difference (d = 0.27).” (Results, Words per minute)
Trained typists left fewer errors uncorrected than untrained typists: 1.02 percent against 1.23 percent.
“Also, trained typists leave fewer errors uncorrected (x = 1.02%) than do untrained typists (x = 1.23%).” (Results, Errors and error-correction behaviour)
Trained typists used about 8 fingers on average, against about 6.5 for untrained typists, a pattern similar to the fast and slow comparison.
“similarly to trained relative to untrained typists (on average, 8 versus 6.5)” (Results, Hand and finger use)
72 percent of the volunteers in the online typing study had taken a typing course.
“85% native language English Took a typing course 72% Hours typing/day” (Table 1: Demographics)
In the 2018 study, the uncorrected error rate was the Levenshtein edit distance between the presented and transcribed strings, divided by the length of the longer string.
“UNCORRECTED ERROR RATE is calculated as the Levenshtein edit distance [23] between the string presented and the ... divided by the larger number of characters between the two strings.” (Performance Measures)
The authors found that untrained typists can be as fast as trained typists, although trained typists are faster on average.
“We found that untrained typists can be as fast as trained typists, but trained typists in general are faster than untrained ones.” (Results, trained versus untrained typists)
The 2018 online typing study defines one word as any five characters when it computes words per minute.
“where one word consists of any ﬁve characters” (Performance Measures)
In the 2018 online study, WPM was calculated for each typed sentence as the transcribed length in words, divided by the time from the first to the last keypress.
“WORDS PER MINUTE (WPM) is calculated for each typed sentence as the length of the string transcribed (in words, where one word consists of any ﬁve characters) divided by the time from the ﬁrst to the last keypress (in minutes).” (Performance Measures)
In a study of 136 million keystrokes from an online typing test, the average participant typed 51.56 words per minute, with a standard deviation of 20.2.
“The average WPM value for participants is 51.56 (SD = 20.2).” (Performance Measures)

Pages that rest on this source

Peer-reviewed research

Observations on Typing from 136 Million Keystrokes

Vivek Dhakal et al., CHI 2018, ACM Conference on Human Factors in Computing Systems, 2018 · DOI 10.1145/3173574.3174220

Read the original

Read on October 16, 2026

What it is: Analyses 136 million keystrokes from an online typing test taken by about 168,000 volunteers. It reports distributions of words per minute, error rate and keystroke timing for keyboard typing.
How much weight it carries: Peer-reviewed CHI paper with a very large sample. The participants were self-selected volunteers who took a test online, so the averages are higher than population averages. The paper says so in its own sampling discussion.

Facts taken from this source

The Dhakal and colleagues study analysed typing data from over 168,000 volunteer participants.
“We have reported extensive statistical analysis for over 168,000 volunteer participants” (Section 'Summary')
The online typing study analyzed 136,857,600 keystrokes from 168,960 participants, most of whom finished all 15 sentences.
“The final dataset includes 136,857,600 keystrokes from 168,960 participants” (Results)
The volunteers in the online typing study had a mean age of 24.5 years, with 75 percent aged 11 to 30.
“Demographics Result Remark Females 52.7% Rest preferred not to specify Males 41.5% Age: mean 24.5” (Table 1: Demographics)
In a study of 136 million keystrokes from an online typing test, the average typing speed of participants was 51.56 words per minute.
“The average WPM value for participants is 51.56 (SD = 20.2).” (Results: Words per minute)
The keyboard study compares its measures by fast versus slow typists and by trained versus untrained typists.
“It shows mean and SD for each measure and compares them for fast versus slow and trained versus untrained typists.” (Results, Table 3 text)
The distribution of typing speeds in the online dataset is not bell-shaped; it is slightly positively skewed.
“The distribution is not a normal-form one; it shows slight positive skewness” (Section 'Words per minute')
Clustering the keystroke data found that most users can be divided into eight groups of typists that differ in performance and accuracy.
“most users can be divided into eight groups of typists that differ in performance, accuracy” (Abstract)
In a study of 136 million keystrokes from an online typing test, typists faster than 90 percent of participants were those above about 78 words per minute.
“Fast Typists are the typists with speeds higher than that of 90% of participants (above approx. 78 WPM in our data)” (Section 'Typist Groups')
Fast typists used about 8.4 fingers on average, while slow typists used about 5.3.
“we observe fast typists to use more ﬁngers than slow typists (averaging 8.4 versus 5.3)” (Results, Hand and finger use)
Fast typists had an average gap of about 120 milliseconds between keystrokes, while slow typists had gaps of over 480 milliseconds.
“For fast typists, the average IKI is ∼120 ms, with a standard deviation of only 11 ms, while slow typists have an IKI of over 480 ms, sometimes as high as 900 ms” (Results, Words per minute and inter-key intervals)
Across participants, faster typing was negatively correlated with uncorrected errors (r = -0.21) and with error corrections (r = -0.36).
“We ﬁnd a negative correlation between WPM and uncorrected errors ( r = −0.21), error corrections ( r = −0.36), and KSPC (r = −0.4)” (Results, Errors and error-correction behaviour)
The fastest typists in the online study reached 120 words per minute or more.
“The fastest typists in the dataset reach speeds of even 120 WPM or more.” (Performance Measures: Words per minute)
The fastest typists in the online typing dataset of about 168,000 volunteers reached 120 words per minute or more.
“The fastest typists in the dataset reach speeds of even 120 WPM or more.” (Section 'Words per minute')
In the online keyboard study, 52.7 percent of volunteers were female and 41.5 percent were male.
“Females 52.7% Rest preferred not to specify Males 41.5%” (Demographics table, first column)
The study found that letter pairs typed by different hands or fingers predict typing speed better than letter repetitions do.
“more predictive of typing speed than, for example, letter repetitions” (Abstract)
In the online study, typing speed varied widely, with a standard deviation of 20.2 words per minute around the 51.56 mean.
“The average WPM value for participants is 51.56 (SD = 20.2).” (Performance Measures: Words per minute)
Most typists in the study performed at least a small share of their keystrokes with rollover.
“most typists performing at least a small percentage of keystrokes by means of rollover” (Results)
Across the study's typists, the share of keystrokes typed with rollover correlated highly with typing performance, r = 0.73.
“Rollover ratio is found to have a high correlation with performance (r = 0.73, p < 0.001).” (Results)
Across the online typists, the share of keystrokes typed with rollover correlated with typing speed at r = 0.73, a high correlation.
“Rollover ratio is found to have a high correlation with performance (r = 0.73, p < 0.001).” (Results, Rollover Behaviour)
In the 136 million keystroke study, slow typists used almost no rollover, while most fast typists used rollover on 40 to 70 percent of their keystrokes.
“Slow typists use nearly no rollover, whereas the majority of fast typists use rollover for 40–70% of keystrokes.” (Abstract)
In the online dataset of 168,000 typists, slow typists used almost no rollover, while most fast typists used rollover for 40 to 70 percent of their keystrokes.
“While slow typists use almost no rollover, the majority of fast typists use rollover for 40–70% of keypresses.” (Results, Rollover Behaviour)
The online typing test's users were mainly younger people from the US who wanted to test and improve their typing, so the sample is not representative of all typists.
“The user base of the company is composed mainly of younger people from the US who are interested in testing and improving their typing skills.” (Method, participants)
The online typing study reports a skewness of 0.513 for its words per minute distribution, a positive skew with a longer tail of faster typists.
“The distribution has a skewness of 0.513” (Words per minute results)
In the same online study, the slowest 10 percent of typists were those below about 26 words per minute.
“Slow Typists are those whose performance is among the slowest 10% (less than approx. 26 WPM in our data)” (Section 'Typist Groups')
Slow typists left significantly more errors uncorrected than fast typists, which the authors suggest may mean they detect their own mistakes less well.
“Slow typists leave signiﬁcantly more errors uncorrected, which could indicate that they are less able to detect their mistakes.” (Results, Errors and error-correction behaviour)
The largest difference between fast and slow typists was in substitution error rate, where a wrong character replaces the intended one.
“Comparing fast with slow typists, we can see that there is a particularly large difference in substitution error rate” (Results, Errors and error-correction behaviour)
The authors hypothesize that substitution errors are common when typists are less consistent and have a poor mental picture of where their fingers are.
“We hypothesise that substitution errors are common when participants are less consistent in typing and have a poor mental representation of the ﬁngers’ position.” (Results, Errors and error-correction behaviour)
Each participant in the 2018 online study transcribed 15 English sentences.
“The task was to transcribe 15 English sentences.” (Task and Procedure)
The sample was people interested in testing and developing their typing skills, so it is not a random sample of the population.
“This sample represents people interested in testing and developing their typing skills.” (Participants)
Trained typists in the online sample averaged about 5 words per minute faster than untrained typists, a small difference next to the spread of about 20 WPM within each group.
“Trained typists in our sample type, on average, 5 WPM faster than untrained typists, with a similar SD, of about 20 WPM, for a relatively small difference (d = 0.27).” (Results, Words per minute)
Trained typists left fewer errors uncorrected than untrained typists: 1.02 percent against 1.23 percent.
“Also, trained typists leave fewer errors uncorrected (x = 1.02%) than do untrained typists (x = 1.23%).” (Results, Errors and error-correction behaviour)
Trained typists used about 8 fingers on average, against about 6.5 for untrained typists, a pattern similar to the fast and slow comparison.
“similarly to trained relative to untrained typists (on average, 8 versus 6.5)” (Results, Hand and finger use)
72 percent of the volunteers in the online typing study had taken a typing course.
“85% native language English Took a typing course 72% Hours typing/day” (Table 1: Demographics)
In the 2018 study, the uncorrected error rate was the Levenshtein edit distance between the presented and transcribed strings, divided by the length of the longer string.
“UNCORRECTED ERROR RATE is calculated as the Levenshtein edit distance [23] between the string presented and the ... divided by the larger number of characters between the two strings.” (Performance Measures)
The authors found that untrained typists can be as fast as trained typists, although trained typists are faster on average.
“We found that untrained typists can be as fast as trained typists, but trained typists in general are faster than untrained ones.” (Results, trained versus untrained typists)
The 2018 online typing study defines one word as any five characters when it computes words per minute.
“where one word consists of any ﬁve characters” (Performance Measures)
In the 2018 online study, WPM was calculated for each typed sentence as the transcribed length in words, divided by the time from the first to the last keypress.
“WORDS PER MINUTE (WPM) is calculated for each typed sentence as the length of the string transcribed (in words, where one word consists of any ﬁve characters) divided by the time from the ﬁrst to the last keypress (in minutes).” (Performance Measures)
In a study of 136 million keystrokes from an online typing test, the average participant typed 51.56 words per minute, with a standard deviation of 20.2.
“The average WPM value for participants is 51.56 (SD = 20.2).” (Performance Measures)