A Bayesian Markov Chain Monte Carlo approach to the generalized graded unfolding model estimation: the future of non-cognitive measurement

by Wei Wang

Institution: University of Illinois – Urbana-Champaign
Year: 2014
Keywords: Item response theory
Record ID: 2039403
Full text PDF: http://hdl.handle.net/2142/46897


Accurately measuring individual differences underpins psychological research, educational and clinical decision-making, personnel selection, and managerial practices. Previous research has concluded that ideal point item response theory (IRT) models are more appropriate than dominance IRT models for measuring non-cognitive variables such as personality, vocational interests, attitudes, person-environment fit (e.g., person-job fit), etc. Although a couple of ideal point IRT models have been proposed in the literature, the only model with estimation software available to the public is the generalized graded unfolding item response model (GGUM) and its corresponding software GGUM2004. However, this software sometimes encounters problems due to the marginal maximum likelihood (MML) estimation method that it utilizes. Therefore, this dissertation research was aimed at developing a new computer program estimating the GGUM model by using a state-of-the-art estimation method??????Bayesian Markov Chain Monte Carlo estimation. A series of studies were conducted to test the estimation accuracy of the new software. The results clearly showed that the Bayesian MCMC estimation method outperformed the traditional MML method, in terms of parameter estimation accuracy, parameter recovery with multidimensional data, and differential item function assessment for an ideal point response process. Implications of these findings and future research directions are discussed.