A0499
Title: An adaptive regrouping subsampling method for linear mixed models
Authors: Rundong Zhou - King\'s College London (United Kingdom) [presenting]
Abstract: In latent group modeling scenarios, datasets are often assumed to follow a linear mixed model, although group memberships are unknown. In addition, group information may be missing or unreliable, which has been largely overlooked in previous studies. A key assumption is introduced: Grouping information provided in a dataset, particularly in linearly structured data, should not be fully trusted. Instead, mathematical procedures are applied to determine whether the data conform to a single linear model or a linear mixed model, and to identify appropriate groupings within a linear mixture framework. To support this, an unsupervised subsampling algorithm is developed that incorporates clustering to automatically assign group labels without requiring prior grouping information. This method ensures that the model fitting process retains the most informative results under potential group uncertainty.
