A History of Bayesian Inference in Educational Measurement
Roy Levy, Robert J. Mislevy · 2021
This chapter surveys the history of Bayesian inference in educational measurement and testing. The mid-century revival of Bayesian inference led the measurement community to look backward and forward: backward to recast earlier developments in Bayesian terms, and forward to advance the field of measurement by way of using Bayesian approaches. In the 1970s, researchers began investigating Bayesian approaches to tackle problems in criterion-referenced testing. This work was closely related to the research on prediction reviewed. Bayesian efforts addressed this principal problem of classifying students with respect to the criterion, as well as problems of determining test length, setting cut scores, evaluating reliability, and marrying probabilistic classifications with utilities in a decision-theoretic frame. Concurrently, researchers such as Embretson were drawing on developments in cognitive psychology to define constructs, construct items, and build measurement models to analyze resulting data. In the 1980s, work in artificial intelligence began to develop in ways that would soon influence educational measurement in several ways.