Abstract

National and international large-scale assessments (LSA) remain to provide the most comprehensive, representative, and credible data to help educators, legislators, and the public assess what students know and can do in various subject areas. Due to rapid advances in information technology, digital transition, and the need of assessing multifaceted constructs, the increasingly rich LSA data pose great challenges to existing techniques used to handle and analyze heterogeneous assessment data. In this talk, I will start with the typical analytical flowchart of LSA and introduce the novel methods we developed for different segments of the protocol. Specifically, I will present a family of innovative methods for multi-cultural analysis of LSA . I will also introduce novel factor regularization methods to support constructing robust latent regressions and generating plausible values for secondary analyses. Although the methods are motivated by large-scale educational surveys, they can also be used in psychological assessments, such as data harmonizations across multi-site studies.