Date of Degree
Access restricted until 07/03/2019
PhD (Doctor of Philosophy)
Jacob J. Oleson
First Committee Member
Brian J. Smith
Second Committee Member
Third Committee Member
Data resulting from eye-tracking experiments allows researchers to analyze the decision making process as study participants consider alternative items prior to the ultimate end point selection. The aim of such an analysis is to extract the underlying cognitive decision making process that develops throughout the experiment. Resulting data can be difficult to analyze, however, as eye-tracking curves have very high autocorrelation values which consists of measurements that are milliseconds apart, as mandated by the nature of eye movements. We propose an analytic approach to eye-tracking data that tests for statistically significant differences at every time point along the curve while calculating an appropriate familywise error rate correction which is based upon an autoregressive correlation assumption of the test statistics. Our technique has been implemented in the R package BDOTS with various extensions relevant to the real-world analysis of highly correlated nonlinear data. A popular alternative approach to analyzing eye-tracking data is to fit mixed models to the area under the curve. Through simulation studies we provide evidence for the benefit of using information criterion measures in selection of the random effects structure and make an argument against current industry-standard approaches such as sequential likelihood ratio tests or always using a maximal random effects structure.
Eye-tracking, Familywise error rate, Nonlinear, Psycholinguistics, Time series
viii, 135 pages
Includes bibliographical references (pages 130-135).
Copyright © 2018 Michael Thomas Seedorff
Seedorff, Michael Thomas. "Methods for testing for group differences in highly correlated, nonlinear eye-tracking data." PhD (Doctor of Philosophy) thesis, University of Iowa, 2018.
Available for download on Wednesday, July 03, 2019