Statistical Results for the session effect.

Ocular fixations contain microsaccades, drift and tremor. We report an increase in the slope of linear fixation drift as a function of time-on-task (TOT). We employed a large dataset (322 subjects, multiple visits per subject). Subjects performed a random saccade task. The task, in which the target dot jumped randomly every one sec, was 100 sec in duration. For each fixation, we regressed eye position against time across multiple segment lengths (50, 100, 300, and 500 ms). We started with the first sample and continued until no further regressions were possible based on the segment length being evaluated. For each segment length, each fixation was characterized by a single value: the maximum slope over the segment length. The slopes were expressed in deg/sec. We were not interested in the direction of the linear drift, so we took the absolute value of the slope as a measure. For data analysis, each 100 sec task was divided into five 20 sec epochs. In an attempt to partially replicate an earlier study of fixation drift over time [1], we also measured drift as the mean velocity of an eye-position signal that had been low-pass filtered at 30 Hz. We found that both methods detected significant drift over our task. For each visit, subjects were tested in two sessions, approximately 20 min apart. Generally, we found statistically significant session effects indicating that drift started at a higher level and increased at a higher rate over the second session. We report that the mean velocity method detects distinctive types of drifting fixation trajectories including curvilinear drift and irregular oscillation. Our findings extend the observations of increased drift over time by [1] to other measures of drift and to a much shorter time interval (100 sec vs two hrs) and a simpler task.