Title: Joint latent variable modeling of binary, ordinal, count, and continuous data for social science and psychological research
Abstract: Research in the social sciences increasingly utilizes mixed data types, such as a combination of ordinal item scores, continuous response times, and discrete variables based on the response process. Generalized linear latent variable models (GLLVMs) fitted to such data can be used to infer relationships between multiple latent constructs and their development over time. These models are often high-dimensional and require efficient estimation methods. We propose an estimator for GLLVMs based on maximizing an approximation to the marginal likelihood and discuss the finite sample and large sample properties of the estimator. The modeling approach is illustrated through joint modeling of response times, action counts, and item scores, where we examine the impact on measurement precision of the proficiency estimates when including multiple types of variables. Extensions of the modeling approach to longitudinal data analysis and joint modeling with non-ignorable missing data are also discussed.
Biography
Björn Andersson is a professor at the Centre for Educational Measurement, University of Oslo (CEMO), Norway. He obtained his Ph.D. in statistics from Uppsala University in 2014 and has worked as a post-doctoral researcher (2015-2017) at the Collaborative Innovation Center of Assessment towards Basic Education Quality, Beijing Normal University in Beijing, China. His research interests include estimation methods for latent variable models, methods to ensure comparability of test scores in applied measurement and applications of item response theory in education, mental health, and psychology.
