CMStatistics 2017: Start Registration
View Submission - CMStatistics
Title: Fitting mixtures of flexible circular distributions with an application to eye tracking data Authors:  Kees Mulder - Utrecht University (Netherlands) [presenting]
Irene Klugkist - Utrecht University (Netherlands)
Ingmar Visser - University of Amsterdam (Netherlands)
Daan van Renswoude - University of Amsterdam (Netherlands)
Abstract: In eye tracking research, a common goal is to characterize how groups differ in their visual search strategies. One derived measure of interest is the direction in which a saccade (ie. an eye movement) is made. Usually, humans are more likely to make saccades in the left-right directions, as well as up-down directions. Previously, this type of data was modeled with mixtures of von Mises distributions using the `movMF' package. We improve upon the usual strategy by fitting mixtures of peaked inverse Batschelet distributions. Frequentist inference can proceed through the EM algorithm. Bayesian inference can proceed through MCMC sampling. In this new approach, fewer parameters are needed, while the data is fit more accurately. Interpretation for this new model is also improved. The R package `flexcircmix` is available to perform the analysis. In an application it is shown that infants have less precision in saccade directions compared to adults.