Does an Eye Tracker Tell the Truth about Visualizations?: Findings while Investigating Visualizations for Decision Making

Abstract

For information visualization researchers, eye tracking has been a useful tool to investigate research participants' underlying cognitive processes by tracking their eye movements while they interact with visual techniques. We used an eye tracker to better understand why participants with a variant of a tabular visualization called 'SimulSort' outperformed ones with a conventional table and typical one-column sorting feature (i.e., Typical Sorting). The collected eye-tracking data certainly shed light on the detailed cognitive processes of the participants; SimulSort helped with decision-making tasks by promoting ef cient browsing behavior and compensatory decision-making strategies. However, more interestingly, we also found unexpected eye-tracking patterns with SimulSort. We investigated the cause of the unexpected patterns through a crowdsourcing-based study (i.e., Experiment 2), which elicited an important limitation of the eye tracking method: incapability of capturing peripheral vision. This particular result would be a caveat for other visualization researchers who plan to use an eye tracker in their studies. In addition, the method to use a testing stimulus (i.e., in uential column) in Experiment 2 to verify the existence of such limitations would be useful for researchers who would like to verify their eye tracking results.

An experimental study of the decision process with interactive technology

“…Kim et al (2012) found long periods of eye fixation in top-left areas of the screen in Typical Sort, while eye fixation in SimulSort was of shorter duration. These results suggest that Typical Sort may have caused subjects to overweight the importance of maximally valued objects along only a few attributes on the left of the screen, while SimulSort encouraged patterns of quick browsing.…”

“…Indeed, in a complementary paper to this one, Kim et al (2012) 10 used eye-tracking to investigate another group of subjects' browsing behavior in the same decision task, with Typical Sort and SimulSort (i.e., the Visual Sort treatment). Kim et al (2012) found long periods of eye fixation in top-left areas of the screen in Typical Sort, while eye fixation in SimulSort was of shorter duration.…”

“…In the first, the participant is able to manually sort objects by one attribute at a time, a capability that is similar to that found in spreadsheet software. In the second, we evaluate a new interactive visual tool that (1) automatically sorts all objects by attribute and (2) uses visual cues to assist with comparisons (this tool was first introduced in Hur and Yi, 2009 and was also evaluated by Kim et al, 2012).…”

“…The inter-correlation coefficient across attributes was set to 0-0.1, which was found to be common in datasets such as nursing home choice selection in an earlier paper evaluating SimulSort(Kim et al, 2012).3 Caplin et al (2011) ran two experiments, one in which the final decision was used to calculate earnings, and one in which the decision at a random time was used to calculate earnings. Both yielded a similar result in the choice process data, except that in the former there was a greater incentive to continue search.…”

Fauxvea: Crowdsourcing Gaze Location Estimates for Visualization Analysis Tasks

“…Visualization analysis often requires keeping several pieces of information in mind while solving a task, so it is possible that the hypothesis does not hold for some tasks, as Kim et al found [30]. In this case, it is still valuable to understand which visual information is inspected and needed to answer a task, even if it is difficult to infer deeper cognitive processes of the person.…”