Automated intelligent feedback in university statistics education

Abstract

Introductory statistics courses are both essential and challenging for many university students. Students struggle to understand the abstract concepts involved, such as significance level and p-value, and the role of uncertainty in statistical procedures. Appropriate feedback could support students in gaining understanding, but is difficult to provide for teachers, since the number of students enrolled in such courses is often large. In this thesis, a solution is sought in automated feedback in an Intelligent Tutoring System, guided by the question: How can automated feedback support students in higher education in gaining understanding of statistics? In two first-year introductory statistics courses for social-sciences students, two feedback types were implemented: inner loop feedback on steps in hypothesis-testing tasks by a domain reasoner and outer loop feedback over series of tasks in the form of inspectable student models.

Separate studies focused on the design, implementation, and student’s use of the two feedback types. Design was based on promising paradigms, such as model-tracing and constraint-based modeling for the domain reasoner. Students’ use of the feedback was evaluated by investigating their feedback-seeking and decision-making behavior. Finally, the influence of both feedback types on student’s course performance was assessed. Lower-achieving students were found to benefit from student models, and students who had had enough time to familiarize themselves with the feedback were found to benefit from the domain reasoner. Hence, the combination of feedback types has the potential to provide many students with useful guidance in the process of learning statistics.