Modeling Teachers’ Diagnostic Judgments by Bayesian Reasoning and Approximative Heuristics

The diagnostic judgments teachers make can be regarded as inferences from manifest observable evidence on students’ behavior (e.g., responses to a task) to their latent traits (e.g., misconceptions). The judgment process can be modeled by Bayesian reasoning. We use this framework to analyze the situation of teachers’ diagnostic judgments on students’ potential misconceptions based on students’ responses. Humans typically deviate from normative Bayesian reasoning and apply heuristic strategies, often by only partially processing the available information (e.g., neglecting base rates). From the perspective of ecological rationality, such heuristics possibly constitute viable cognitive strategies for assessing student errors. We investigate the adequacy of a cognitively plausible heuristic strategy, which amounts to approximating the average probability information on prior hypotheses (base rates of student misconceptions) and evidence (student errors). With a computational simulation, we compare this strategy to optimal Bayesian reasoning and to information-neglecting strategies. We interpret the resulting accuracy within the ecology of informal student assessment.