Time Course and Hazard Function: A Distributional Analysis of Fixation Duration in Reading

Abstract

Reading processes affect not only the mean of fixation duration but also its distribution function. This paper introduces a set of hypotheses that link the timing and strength of a reading process to the hazard function of a fixation duration distribution. Analyses based on large corpora of reading eye movements show a surprisingly robust hazard function across languages, age, individual differences, and a number of processing variables. The data suggest that eye movements are generated stochastically based on a stereotyped time course that is independent of reading variables. High-level reading processes, however, modulate eye movement programming by increasing or decreasing the momentary saccade rate during a narrow time window. Implications to theories and analyses of reading eye movement are discussed.

Keywords

• reading
• distribution
• fixation duration
• statistical modeling