Proceedings of the Annual Meeting of the Cognitive Science Society

Memory researchers have studied learning behavior andextracted regularities describing learning and forgetting overtime. Early work revealed forgetting curves and the benefitsof temporal spacing and testing for learning. Computationalmodels formally implemented these regularities to capturerelevant trends over time. As these models improved, theywere applied to adaptive learning contexts, where learningprofiles could be identified from responses to past learningevents to predict and improve future performance. Often times,past performance is expressed as accuracy alone. Here weexplore whether a model’s predictions can be improved ifpast performance is expressed by an integrated measure thatcombines accuracy and response times (RT). We present asimple, data-driven method to combine accuracy and RT on atrial-by-trial basis. This research demonstrates that predictionsmade using the Predictive Performance Equation improvewhen past performance is expressed as an integrated measurerather than accuracy alone.