Less Talk, More Code: Practice-Based Instruction Improves Programming Skill Acquisition
Abstract
Introductory computer science education often emphasizes watching experts write and explain code. We investigate whether practice-based instruction improves programming skills relative to traditional lectures and which features of practice make it effective. In this preregistered experiment (N = 250), we compared three approaches for an introductory programming lesson: watching a video, code tracing visualizations, and writing code with immediate, AI-generated feedback. Participants who engaged in practice-based instruction performed significantly better than those who watched the video on a novel code-generation test, with the highest performance among learners who practiced writing code. Practice-based instruction also resulted in reduced distraction and increased interest. Whereas video and code tracing conditions led to overconfidence about programming abilities, code writing led to underconfident self-assessments. These findings suggest that active practice, particularly generative practice with feedback, supports learning with advantages in cognitive processing, metacognition, and interest in programming.