A0729
Title: Optimal next stage designs for sparse longitudinal data
Authors: MingHung Kao - Arizona State University (United States) [presenting]
Abstract: The focus is on optimal designs for collecting informative sparse functional/longitudinal data to precisely predict the trajectories of individual random curves. Constructing such designs typically requires knowledge of unknown quantities of the model. For this issue, previous studies considered locally optimal designs by replacing these quantities with their estimates from the pilot study. However, the information on the design used for collecting the pilot data is completely discarded when developing the design for a future study. To address this drawback, a multi-stage design approach is proposed to adapt the optimal design for the next study based on the design for the pilot study. A new optimality criterion is developed to select a design that gives a precise curve prediction for the next study and improves the curve prediction in the pilot study. Useful theories and computational approaches to facilitate the identification of optimal designs are also provided.
