Cognitive diagnosis models (CDMs) are psychometric models that are developed for assessing examinees’ mastery and nonmastery of attributes. The generalized Deterministic Input, Noisy and Gate (G-DINA; de la Torre, 2011) model has the most general formulation and can be applied for different cognitive diagnosis purposes. The G-DINA model is a mathematical probabilistic function that can produce a variety of CDMs in the literature. When the attributes are assumed to be a function of one general ability, a higher-order G-DINA model can be formulated. This study aims to develop a multilevel higher-order G-DINA model for longitudinal data and conduct a series of simulations to assess the parameter recovery using Markov chain Monte Carlo (MCMC) methods. Fifteen items (simulated dichotomously scores items) with five attributes (simulated cognitive skills) were used to generate item responses at four time points. The Q-matrix and the generated values came from an empirical analysis of the fraction subtraction test (Tatsuoka, 1990). Three levels were manipulated, where a within-occasion model at Level 1, a within-person model at Level 2, and a between-person model at Level 3. Two groups of persons were included at Level 3 with each having 1,000 persons. Parameter recovery of the multilevel higher-order G-DINA model was assessed by computing the bias and root mean square error (RMSE). The results showed that the multilevel higher-order G-DINA model had fairly good item parameter recovery (lower RMSE values) and effectively captured the variations in the individual changes over time. The authors provide several suggestions for future study.

Authors:
Hung-Yu Huang, University of Taipei, Taiwan
Su-Pin Hung, National Cheng Kung University, Taiwan

About the Presenter(s)
Professor Hung-Yu Huang is a University Professor/Principal Lecturer at University of Taipei in Taiwan

See this presentation on the full schedule – Friday Schedule

Bookmark