Investigating Student Usage of Metacognitive Problem-Solving Strategies in Algorithm Design Problems

Algorithms is a core course in the undergraduate computer science curriculum and is known both for its relative difficulty and its career implications. A key learning goal in the course is algorithm design, but the problem-solving skills related to the task of algorithm design are nebulous and typically left implicit in algorithms courses and materials. Prior results hint at the possibility that students are learning the content material itself but not the metacognitive problem-solving skills necessary to generalize what they've learned to new problems, but much is left unstudied.

To address this, we present a series of work aimed at answering foundational questions about metacognitive problem-solving skills in algorithms courses. To begin, we conduct a systematic literature review mapping the landscape of algorithms education research, finding broadly that the literature is sparse. In our second work, we aim to create a baseline by characterizing students' natural strategy development in algorithms courses, finding that students are learning important metacognitive strategies on their own but the adoption is inconsistent and improper use of the strategies often hinders student progress. With this in mind, we aim next to demonstrate the potential for metacognitive strategies to benefit students. We conduct a study investigating analogical reasoning, a core metacognitive problem-solving strategy, and present evidence that analogical reasoning is very closely tied to success in algorithm design. These findings justify the importance of explicit modeling and instruction about metacognitive strategies in algorithms classrooms. Finally, we to begin developing pedagogical approaches for explicitly teaching metacognitive strategies, we conduct a randomized control trial investigating how different levels of scaffolded instruction affect adoption of a well-established strategy, finding both clear evidence the scaffolding can be helpful and some weaker evidence that over-scaffolding is also possible.