Data Science to Improve Feedback, Understand Student Behaviour and Address Equity Issues in Computer Programming Education Using Data from Large-Scale Courses

The delivery of programming courses online offers great promise to provide quality programming education in an accessible manner. However, it also introduces new challenges, including how to maintain course quality as the ratio of students to teaching staff increases. In particular, the provision of effective feedback, detailed course evaluations and the promotion of equity can all become more challenging as the size of a course increases. This work takes a data-science approach to addressing the challenges and maximising the potential of Large Scale Programming Courses (LSPCs). Specifically, it explores, integrates and develops potential data science techniques to improve educational tools and materials in LSPCs, including to match the quality of small-scale instruction. In addition, it considers how data from LSPCs, in combination with data science techniques, can be utilised to improve programming education in general, including for both small and large-scale instruction. We begin with an introduction to the challenges and questions that arise when studying LSPCs. In Chapter 2, we then survey a wide range of techniques for generating automated feedback. From this survey, we construct a novel framework for understanding and guiding automated hint production. We then introduce two novel data science techniques for analysing student behaviour in LSPCs. Specifically, in Chapter 3 we introduce Progress Networks for summarising student behaviour on programming exercises. In Chapter 4, we present DETECT, a divisive hierarchical clustering technique for trend detection. In Chapter 5, we then address issues of equity with a particular focus on understanding confounding variables in LSPC data. Finally, we conclude with a general discussion of the work and how it addresses the overarching research questions