CMStatistics 2022: Start Registration
View Submission - CMStatistics
Title: The development of ``variationalDCM'' an R package performing variational Bayesian estimation for DCMs Authors:  Keiichiro Hijikata - The University of Tokyo (Japan) [presenting]
Motonori Oka - London School of Economics (United Kingdom)
Kazuhiro Yamaguchi - University of Tsukuba (Japan)
Kensuke Okada - The University of Tokyo (Japan)
Abstract: ``variationalDCM'' is provided, an R package that performs recently developed variational Bayesian (VB) estimation methods for Diagnostic Classification Models (DCMs). DCMs are a class of latent variable models used to reveal students' current knowledge status and applied to various educational tests. Despite increasing attention to DCMs, there are few software programs available on the Internet for DCMs, and, to the best of our knowledge, there do not seem to be any programs that estimate parameters by VB methods. VB methods are techniques for approximating the posterior distribution of parameters in the Bayesian estimation framework. They are characterized by their fast calculation time compared to Markov Chain Monte Carlo methods which are usually used for Bayesian estimation. This package enables fast estimations by VB methods for various DCMs and can be applied to large-scale data. We implement five functions that estimate model parameters for 1) deterministic input noisy AND gate (DINA) model, 2) saturated DCM, 3) multiple-choice DINA model, and 4) hidden Markov type longitudinal general DCM and estimates 5) Q-matrix for DINA model.