Files in this item



application/pdfZHANG-DISSERTATION-2018.pdf (2MB)
(no description provided)PDF


Title:Cognitive diagnosis modeling and applications to assessing learning
Author(s):Zhang, Susu
Director of Research:Chang, Hua-Hua
Doctoral Committee Chair(s):Chang, Hua-Hua
Doctoral Committee Member(s):Culpepper, Steven A; Douglas, Jeffrey A; Anderson, Carolyn J; Zhang, Jinming
Department / Program:Psychology
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):cognitive diagnosis, learning, hidden Markov, Bayesian statistics, Markov chain monte carlo
Abstract:Chapter 1: Cognitive diagnosis models (CDMs) are restricted latent class models designed to assess test takers' mastery on a set of skills or attributes. With a wide range of applications in education and in psychopathology, various CDMs have been proposed and fitted to response data from different scenarios. Recently, Xu (2017) derived sufficient conditions for identifying model parameters of a restricted latent class model, which generalizes many existing CDMs. We propose a Bayesian estimation algorithm for this restricted latent class model. The model is applied to the Examination for the Certificate of Proficiency in English language assessment data (e.g., Henson et al., 2007). Chapter 2: There has been a growing interest in measuring students' growth over time. CDMs were traditionally used to measure students' skill mastery at a static time point, but recently, they have been used in many longitudinal models to track students' changes in skill acquisition over time. In this chapter, we propose a longitudinal learning model, where different kinds of skill hierarchies were considered, and the reduced-reparameterized unified model (r-RUM) or the noisty input, deterministic-``and''-gate (NIDA) model is used to measure students' skill mastery at each time point. This model is fitted to the Spatial Rotation data set (e.g., Wang et al., 2016), and different models were compared using Bayesian model comparison methods. Chapter 3: The increased popularity of computer-based testing has enabled researchers to collect various types of process data, including response times. Extensive research has been conducted on the joint modeling of response accuracy and response times. Recent research on CDMs begins to explore the relationship between speed and accuracy to understand students’ fluency of applying the mastered skills, in addition to mastery information, in a learning environment. In this chapter, we propose a mixture hidden Markov Diagnostic Classification Model framework for learning with response times and response accuracy. Such a model accounts for the heterogeneities in learning styles among students by modeling the different learning and response behaviors among subgroups. The proposed model is evaluated through a simulation study in terms of parameter recovery. Chapter 4: We introduce an R package, hmcdm, that can be used to fit several longitudinal models for learning under the cognitive diagnosis framework. The package allows users to simulate item responses (and response times if applicable) under several learning models, to fit the models using Markov Chain Monte Carlo (MCMC) methods, to compute point estimates of parameters based on the MCMC samples, and to evaluate and compare different models using Deviance Information Criterion and posterior predictive probabilities.
Issue Date:2018-06-05
Rights Information:2018 by Susu Zhang. All rights reserved.
Date Available in IDEALS:2018-09-27
Date Deposited:2018-08

This item appears in the following Collection(s)

Item Statistics