UPDATING AND EXTENDING THE KEYSTROKE MODEL WITH A NATURALISTIC STUDY
Open Access
- Author:
- Burns, Marc
- Graduate Program:
- Informatics
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- March 16, 2023
- Committee Members:
- Frank Ritter, Thesis Advisor/Co-Advisor
Xiaolong Zhang, Committee Member
Jeffrey Bardzell, Program Head/Chair
Nick Giacobe, Committee Member - Keywords:
- Human-Computer Interaction
Keystroke-Level Model
HCI
Human Factors
Big Data
Typing Speed
Mouse Moves - Abstract:
- For over forty years, the HCI community has used the Keystroke-Level Model (KLM) first proposed by Card et al. (1980) and revised (1983) to approximate user reaction times. Despite users adopting computers earlier in life, no updates have been proposed for these constants. Current research focuses on the use of the model in new ways, such as mobile devices, collaborative computer play, different input devices, or touchscreen controls. All this research occurs in strictly controlled, prescriptive, observed laboratory experiments. Although still relevant as evidenced by the studies of applying the KLM to mobile devices, few studies can be located permitting participants to select their own tasks. Participants (n=83) recorded their interactions during normal use. Data recorded consists of over two million keystrokes, one million mouse movements, 255,000 mouse draws, and almost 500,000 mouse clicks. From these logs, a distribution of actions and times were computed. Current users in naturalistic settings have distributions that vary by action. They type at a rate consistent with skilled typists of the 1980s (n=75K, 244-247 ms). This study defines continuous typing is defined as two or more keystrokes in sequence without interruption and having a maximum inner-keystroke time of 1,500 ms. Homing to mouse and keyboard (n=59k, 358-362 ms) and mouse pointing (n=256k, 584-587 ms) and drawing (n=27k, 622-634 ms) speeds are reliably faster and left (351k, 107 ms) and right (n=28K, 121 ms) mouse clicks are slower than the KLM constants. Double clicks do not appear to be two clicking events, as these events take three times as long as a single click (n=33K, 318 ms). These new values can foster a discussion on naturalistic computer usage and updates the KLM’s constants to show how users work differently in current naturalistic situations.