Bayesian diagnosis tracing: Application of procedural misconceptions in knowledge tracing

Abstract

Bayesian diagnosis tracing model (BDT) replaces the generic “wrong” response in the classical Bayesian knowledge tracing model (BKT) with a vector of procedure misconceptions. Using a novel dataset with actual student responses, this paper shows the BDT model has better interpretability of the latent factor and minor improvement in out-sample predictability in some specification than the BKT model.