This paper presents a Bayesian framework for evaluative classification. Current education policy debates center on arguments about whether and how to use student test score data in school and personnel evaluation. Proponents of such use argue that refusing to use data violates both the public’s need to hold schools accountable when they use taxpayer dollars and students’ right to educational opportunities. Opponents of formulaic use of test-score data argue that most standardized test data is susceptible to fatal technical flaws, is a partial picture of student achievement, and leads to behavior that corrupts the measures.
A Bayesian perspective on summative ordinal classification is a possible framework for combining quantitative outcome data for students with the qualitative types of evaluation that critics of high-stakes testing advocate. This paper describes the key characteristics of a Bayesian perspective on classification, describes a method to translate a naïve Bayesian classifier into a point-based system for evaluation, and draws conclusions from the comparison on the construction of algorithmic (including point-based) systems that could capture the political and practical benefits of a Bayesian approach. The most important practical conclusion is that point-based systems with fixed components and weights cannot capture the dynamic and political benefits of a reciprocal relationship between professional judgment and quantitative student outcome data.