Analysis of Grading Times of Short Answer Questions

Abstract

We present an analysis of factors correlated to grading speed in short answer questions from college level STEM courses using a novel dataset collected by an online education company. By analyzing timestamp data, we were able to estimate how long instructors grade individual student responses, which we typically found to be less than 10 seconds. This dataset provides us with a unique opportunity to determine which steps in the grading workflow could benefit from intervention. We found that sorting responses by rubric similarity has the potential to drastically reduce grading time by up to 50% per response. We plan to follow this work by implementing an intelligent agent to present responses in a sorted order to minimize grading time.