LAUSR

Bayesian knowledge tracing (BKT) is a knowledge inference model that underlies many modern adaptive learning systems. The primary goal of BKT is to predict the point at which a student has reached mastery of a particular skill. In this paper, we examine the degree to which changes in sample size influence the values of the parameters within BKT models, and the effect that these errors have on predictions of student mastery. We generate simulated data sets of student responses based on underlying BKT parameters and the degree of variance which they involve, and then fit new models to these data sets, and compared the error between the predicted parameters and the seed parameters. We discuss the implications of sample size in considering the trustworthiness of BKT parameters derived in learning settings and make recommendations for the number of data points that should be used in creating BKT models.