Title: Estimation of generalized linear latent variable models for performance and process data with a mixture of ordinal, continuous, and count variables
Abstract: Computer-based assessments often record not only task responses (ordinal data) but also other information on human-computer interactions such as the time on task (continuous data) and the number of operations (count data). Such mixed-type data often require multidimensional models that can account for residual dependencies between the observed variables. In this circumstance, conventional methods such as CFA and IRT are not suitable for simultaneous analyses of the mixed-type data. In this study, we employ generalized linear latent variable models (GLLVMs) for a combination of ordinal, count and continuous variables. We implement an efficient estimation method that uses a higher-order Laplace approximation and examine the performance of the proposed method in terms of estimation efficiency, convergence, and the recovery of model parameters.