Variable construction for predictive and causal modeling of online education data

Abstract

We consider the problem of predictive and causal modeling of data collected by courseware in online education settings, focusing on graphical causal models as a formalism for such modeling. We review results from a prior study, present a new pilot study, and suggest that novel methods of constructing variables for analysis may improve our ability to infer predictors and causes of learning outcomes in online education. Finally, several general problems for causal discovery from such data are surveyed along with potential solutions.