This paper investigates the correspondence between student affect in a web-based tutoring platform throughout the school year and learning outcomes at the end of the year, on a high-stakes mathematics exam. Affect detectors were used to estimate student affective states based on post-hoc analysis of tutor log-data. For every student action in the tutor, the detectors provided an estimated probability that the student was in a state of boredom, engaged concentration, confusion, and frustration, and estimated the probability that they were exhibiting off-task or gaming behaviours. Boredom during problem solving was found to be negatively correlated with performance, however, boredom was positively correlated with performance when exhibited during scaffolded tutoring. A similar pattern was seen for confusion. Engaged concentration and, surprisingly, frustration were both associated with positive learning outcomes.
Comment from Martyn Cooper: This paper reports an investigation of the relationship between affective states of students as noted in online tutor logs and the performance in summative high stakes exams. An interesting result was that students who were bored or confused while answering the main problems, tended to do poorly on the test; however, boredom and confusion on scaffolding were associated with positive performance on the test. It is a good example of using analytics to assess students’ attitudes while learning. However, it is just a study with a single cohort over one year in a single limited learning environment used in maths education. The intention is to work towards better support for tutors to intervene appropriately given the students’ affective states and their contexts.