Observations on Typing from 136 Million Keystrokes

Abstract

We report on typing behaviour and performance of 168,000 volunteers in an online study. The large dataset allows detailed statistical analyses of keystroking patterns, linking them to typing performance. Besides reporting distributions and confirming some earlier findings, we report two new findings. First, letter pairs typed by different hands or fingers are more predictive of typing speed than, for example, letter repetitions. Second, rollover-typing, wherein the next key is pressed before the previous one is released, is surprisingly prevalent. Notwithstanding considerable variation in typing patterns, unsupervised clustering using normalised inter-key intervals reveals that most users can be divided into eight groups of typists that differ in performance, accuracy, hand and finger usage, and rollover. The code and dataset are released for scientific use.

How do People Type on Mobile Devices?

“…Our methodology closely follows prior work on large-scale online studies of desktop typing [11]. To assess a user's typing speed, we use transcription typing, a common task to study motor performance that excludes cognitive aspects related to the process of text generation.…”

“…Our test supports the main mobile operating systems and browsers and was available globally on the Internet in a collaboration with a Web company offering typing testing and training. In the design of the software, we directly built on work by Dhakal et al [11] who studied transcription typing on physical keyboards: (1) we used the same phrase set representative of the English language; (2) we updated performance feedback only after users committed a phrase; and 3we chose well-understood performance metrics covering speed and errors. However, we needed to adapt the software to support mobile devices, making it responsive to different screen sizes, changing the logging and updating the database structure.…”

“…Table 1 summarizes the demographic background of the 37,370 voluntary participants left in the database after preprocessing (see below). Similar to the general user-base of the company hosting the webpage (see [11]), the test was completed by more females than males, the majority of which came from the U.S. and were mostly experienced in typing in English (native -56%, always -21%, usually -12%). The majority reported entering text using two thumbs (74%) and using the QWERTY layout (87%).…”

“…We followed the same procedure as Dhakal et al [11]. The task was to transcribe 15 English sentences, shown one after another.…”

Generation of Synthetic Query Auto Completion Logs

“…Users delete characters either due to typing errors [6,20] made within that word, or to switch to an entirely different word. Among various typing errors that can occur in any character level entry systems [22], the following error categories that are predominantly discussed in past research [6,20,21,46] are included in the generation process.…”