Convergent Validity of a Student Model: Recent-Performance Factors Analysis

Models of student performance can incorporate a skill decomposition that lists the skills that each activity requires. A good model must be sensitive to improvements in skill decomposition. We validate the Recent-Performance Factors Analysis model of student performance by checking its sensitivity to the skill decomposition. We use a dataset from a tutoring system where the skill model has been improved by the Learning Factors Analysis algorithm for skill model refinement and by expert validation. We find that R-PFA reflects improvements in the skill model, providing evidence of convergent validity of R-PFA. We argue that R-PFA may be sensible as a predictive model in Learning Factors Analysis because of its convergent validity and because the R predictor of R-PFA represents mastery-aligned learning curves.