This research project will advance statistical modeling for latent variables in big and complex longitudinal data, with a focus on personalized education. Given the increasingly complex structures of computerized tests and online surveys, education is getting ever closer to a time when personalization will become possible. Work in this area is hampered, however, by a lack of state-of-the-art statistical techniques to analyze the data. This project will contribute a set of statistical models and inference methods to stimulate more longitudinal studies and applications in latent trait analysis. The research will be integrated within K-12 education through the application of the developed methods to a personalized learning platform. This platform is widely used by students from Grade 2 to 12 throughout the United States. The investigator also will apply the new methods to data sets from other disciplines, including psychology, ecology, and engineering. Graduate students will be actively engaged in this research project. Results from this research will be incorporated into an undergraduate seminar, an open online course, and professional development and training courses. The results also will be disseminated through conference presentations and journal publications. All computational algorithms with scalability for big data will be distributed as user-friendly and open-source software packages to the public.
This research project will focus on learning the dynamic changes of latent trajectories from big and complex longitudinal testing data. The investigator will develop both parametric and nonparametric statistical models and inference methods, especially for dichotomous and categorical data. First, the project will build up a new class of hierarchical dynamic models. These models will provide more accurate estimates for latent ability (traits), including for real time data. Model criteria will be proposed to assess this improvement. The results will allow educators to design better educational strategies and computerized tests according to students' respective abilities. Second, the project will develop nonparametric modeling to flexibly capture the varying dependence and nonlinear effects of the latent trajectories. The developed methods will be useful for explaining, predicting, and grouping changes in latent ability, which is vital in personalized learning. Scalability strategies will be developed to make computation feasible for the developed statistical models. This last step is critical to ensure wide impact of the project results.
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.
