Statistical Advances for Intensive Categorical Spatiotemporal Data from an Eye-Tracking Technique

该研究项目将开发分析眼球追踪数据的方法，为人类行为的潜在认知过程提供更准确的视角。眼球运动倾向于跟随注意力，而眼球追踪是一种灵活的追踪眼球运动的工具。尽管眼动追踪方法无处不在，但目前的数据分析实践是有限的。待开发的方法是一般性的，可用于在收集可比较数据的其他领域推进知识。研究人员将运用这些新方法与心理语言学、教育学和计算机科学的研究人员进行跨学科的合作。研究生和本科生将接受这些分析技术的培训，该项目的结果将通过期刊文章、会议、研讨会和网络研讨会传播给专业和多学科社区。该项目的所有发现、数据、代码和软件都将公开提供。本研究项目将开发具有复杂时间、空间和依赖结构的密集分类时间序列数据的统计模型。研究人员将开发一种通用的模型，称为动态广义线性混合效应模型或动态GLMM。这些模型将捕捉时间效应（序列依赖性和趋势效应）和空间效应（空间依赖性和距离效应），同时考虑多种形式的数据模式变化，包括试验、项目、人员、项目组（或集群）和人员组。研究人员将为二进制数据提供动态GLMM规范，这将表明参与者是否在每个时刻都关注关键兴趣区域。研究人员还将使用IRTree方法提出有序类别或部分有序类别数据的动态GLMM规范。这种方法使研究人员能够检查驱动眼睛在多种刺激中做出选择的认知过程。为了改进数据分析实践，使用动态GLMM的方法将按照数据分析过程的顺序发展：数据描述、模型规范、模型估计、模型选择和模型评估。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This research project will develop methods for analyzing eye-tracking data that will provide a more accurate view of the underlying cognitive processes of human behavior. Eye movements tend to track with attention, and eye-tracking is a flexible tool for tracking eye movements. Despite the ubiquity and utility of eye-tracking methodology, current data analytic practices are limited. The methods to be developed are general and may be used to advance knowledge in other fields where comparable data are collected. The investigators will apply the new methods in interdisciplinary collaborative efforts with researchers from psycholinguistics, education, and computer science. Graduate and undergraduate students will be trained in these analytic techniques, and the results of this project will be disseminated to both specialized and multidisciplinary communities through journal articles, conferences, workshops, and a webinar. All findings, data, code, and software from this project will be made publicly available. This research project will develop statistical models for intensive categorical time series data with the complex temporal, spatial, and dependence structures exhibited by eye-tracking data. The investigators will develop a general class of models called dynamic generalized linear mixed effect models or dynamic GLMM. These models will capture both temporal effects (serial dependence and trend effects) and spatial effects (spatial dependence and distance effects) while accounting for multiple forms of variation across data modes including trials, items, persons, item groups (or clusters), and person groups. The investigators will present a dynamic GLMM specification for binary data, which will indicate whether the participant is or is not fixating on a critical interest area at each moment in time. The investigators also will present a dynamic GLMM specification for ordered-category or partially ordered category data using an IRTree approach. This approach allows the researcher to examine cognitive processes which drive the choice among multiple stimuli for the eye to fixate on. To improve data analytic practices, methods using the dynamic GLMM will be developed following the sequence of the data analytic procedures: data description, model specification, model estimation, model selection, and model evaluation.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.