Chapter 26 - Eye Fixation Durations in Reading: Models of Frequency Distributions

Abstract

This chapter discusses quantitative versions of three models of fixation-duration frequency distributions. The models are proposed by Suppes, McConkie, Kerr, and Dyre and a modified version of that proposed by Harris, Hainline, Abramov, Lemerise, and Camenzu. These, plus two additional versions of the Harris et al. model, are fit to a large set of eye-movement data from a group of adult readers. The McConkie model fit the data best, though the fit appears to result primarily from the functions used rather than from the architecture proposed. The two-state transition model clearly provides a better fit to the data than either the two-stage gamma or the two-state mixture model. The purpose of including two additional versions of the two-stage model was to determine whether the two-state transition model's ability to describe the fixation-duration frequency distributions was because of its architecture or its use of linear hazard distributions to model the early saccades. Two architectures often used in processing models were examined, a mixture and a race model, to describe the Stage-1 period, prior to the exponentially based Stage 2.