Modeling Cognitive Processes of Learning with SAS® Procedures

Traditionally, the principal goal of educational assessments is to evaluate students’ academic achievement or proficiency in comparison to their peers or against promulgated standards. Both classical test theory (CTT) and item response theory (IRT) modeling frameworks provide measurement in the form of a summative estimate of the outcome variable. In recent years, understanding and exploring the complex cognitive processes that contribute to the learning outcome received growing interest in the field of psychometrics, where various item response modeling approaches have been devised to explain the relationship between the outcome and its componential attributes. Such approaches include the linear logistic test model (LLTM), the crossed random-effects linear logistic test model (CRELLTM), and the multiple regression method (MR). This paper introduces these statistical models and illustrates their implementation with the GLM, NLMIXED, and GLIMMIX procedures over an empirical data set. A simulation study was conducted to examine the performance of these approaches. INTRODUCTION Faced with the prevalence of standardized tests which are either legislature-mandated (such as the No Child Left Behind Act of 2001) or commercially available, educators demand that the assessments their student are required to participate provide instructionally relevant results to have their worth maximized in helping with teaching and learning in return for their expensive costs. Advancement in cognition in educational psychology has increased our understanding of the mechanisms of human learning and problem solving, including cognitive domains, schemas, strategies, sub-operations/tasks, production rules, etc. As the traditional CTT and IRT frameworks are not suitable, the challenge to evaluate these unobservable processes has given rise to a series of componential models that are having a huge impact on test construction and analysis procedures. Before presenting these models, it is necessary to introduce briefly an indispensable modeling element.