Using naturalistic typing to update architecture typing constants

Despite the near-ubiquity of graphic user interfaces for navigating the digital and virtual space, relatively little is known about their naturalistic usage. We start to address two questions. First, how do people use computers outside of laboratory studies? This includes what we can we determine about user behavior by analyzing detailed user logs. Second, can we update the constants in user and cognitive models for predicting typing time based on naturalistic behavior? We thus recorded naturalistic logs of 45 users over 219 sessions providing 1,865 hours of behavior (average session=8.18 hours). The analyses of keystroke times are sensitive to the definition of typing (e.g., how close keys are to be counted as continuous typing), and comparisons will need to provide clear definitions or tradeoff curves. Using this data, we updated the typing and homing constants of the KeystrokeLevel Model (Card, Moran, & Newell, 1983), a theory of interface behavior used to provide constants for many cognitive architectures. The results suggest that people are typing faster than previously believed based on the 1983 KLM predictions; homing (moving one’s hand from mouse to keyboard and keyboard to mouse) occur frequently, and now appear to be different events and thus require separate constants.