Modeling Students' Learning and Variability of Performance in Problem Solving

Given data about problem solving times, how much can we automatically learn about students’ and problems’ characteristics? To address this question we extend a previously proposed model of problem solving times to include variability of students’ performance and students’ learning during sequence of problem solving tasks. We evaluate proposed models over simulated data and data from a “Problem Solving Tutor”. The results show that although the models do not lead to substantially improved predictions, the learnt parameter values are meaningful and capture useful information about students and problems.