The Application of Bayesian Networks in Learning Progression–Based Assessment

• In Event: Modern Measurement for Science Learning Progressions (Book by Routledge)

Abstract

The Bayesian paradigm provides a convenient mathematical system for reasoning about evidence. Bayesian networks provide a graphical language for describing complex systems, and reasoning about evidence in complex models. This allows assessment designers to build assessments that have fidelity to cognitive theories and yet are mathematically tractable and can be refined with observational data (Mislevy, Almond, Yan, et al, 2015). This presentation illustrates how Bayesian network modelling is used to develop a learning progression-based assessment tool, including:
• Background information on Bayesian networks, Graphical Models, and related inference and representation methods and provide examples of their use in educational assessment.
• The knowledge engineering processes used to go from a basic conception of a domain to a Bayesian network model which can be used to score a science learning progression assessment.
• Bayesian Network basics, including conditional probability.
• Building and validating networks using data, including both EM and MCMC algorithms.
• Developing a prototype infrastructure for classroom assessment, including automated scoring
• Application examples for teachers using Bayesian networks in classroom assessment, with real-time score reports on students’ proficiency measured by the learning progression and students’ progress over multiple time points.

Author

• Duanli Yan, Educational Testing Service