Experimental Measures of Difficulty for Princeton Courses

Abstract

Course evaluations at Princeton lack numerical difficulty measures, but the textual evaluations are rich with student commentary about course difficulty. This suggests that there is an opportunity to generate a numerical difficulty score from the textual evaluations. In this project, we collect a data set via scraping and test several variations of a “bag of words” approach with the goal of determining whether such an approach holds promise as a difficulty measure. These variations are evaluated by testing their correlation with the number of pages of weekly reading assigned in a course. Surprisingly, the correlation is negative, perhaps because STEM courses are perceived as difficult and are prone to lighter reading loads. As a result, a research agenda is suggested for similar tools that can complement human academic advisors in helping students to choose better schedules.