Treffer: Evaluation of Different Group Formation Methods in the Context of Distributed Pair Programming: Design and Experiment in Higher Education

Group formation is an important issue in collaborative learning, as properly formed groups can lead to better interactions between students and enhance their learning. This issue also applies to pair programming when it is used as a learning activity. When the number of students is important and/or students and teachers do not know each other very well, manual group formation, either by students or teachers, can become difficult. Many of the existing automatic methods used for automatic group formation in pair programming require a certain amount of personal information about the students, which can make the implementation of this kind of system difficult. Hence, in this paper we present an automatic pair formation method integrated to a distributed pair programming application. Our approach is based on personal data from the students collected with questionnaires within our platform, but also on trace data automatically collected from the pair programming activities completed by students. It has been experimented during an introductory programming course in higher education. The objective was to identify, among programming experience, self-efficacy, gender and past engagement in pair programming activities, the formation criteria that led to the best perceived compatibility and work distribution as well as performance. The results show that homogeneous pairs in self-efficacy are significantly better than heterogeneous pairs, both in terms of perceived compatibility and perceived work distribution.