Flexible Approaches to Cognitive Diagnosis: Nonparametric Methods and Small Sample Techniques

• Activity Number: 399
• Type: Topic Contributed
• Date/Time: Tuesday, August 2, 2011 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Chia-Yi Chiu*+
• Companies: Rutgers University
• Keywords: cognitive diagnosis ; Q-matrix

Abstract

In educational testing research, specialized latent class models for cognitive diagnosis have been developed to classify mastery or non-mastery of each attribute in a set of attributes the exam is designed to assess. The ultimate goal of applying diagnostic models is to classify subjects into one of several different categories describing their attribute profiles. However, in many cases where such models might be applicable, sample sizes are too small to adequately fit models, and even when adequate sample sizes are available, the models often display substantial misfit. An alternative is to use nonparametric methods that only assume a Q-matrix, a matrix that associates items with the specific attributes they measure. Nonparametric methods include cluster analysis techniques and algorithms for classifying examinees based on proximity to ideal response patterns. These robust methods do not rely on fitting models, and result in classification rates that are nearly independent of sample size, and compare favorably to parametric models. These methods and their theoretical properties will be discussed, including rapid algorithms for Q-matrix modification and efficient model-free methods