Does Deliberately Failing Improve Learning in Introductory Computer Science?

Abstract

We report our experience with technology-enhanced Productive Failure (PF) in an introductory computer science course. First, we sought to assess whether the use of algorithm visualization tools during the PF problem-solving phase enhanced learning. Second, we used an experimental study to measure learning effects of administering failure-driven scaffolding (FDS) during the PF sessions, that is, explicitly nudging generation with suboptimal representations deliberately designed to lead to failures. Results from surveys and log data indicated that our visualization tools helped students explore the problem space and performance data signaled that FDS improved students’ constructive reasoning (Cohen’s d 0.194, $BF_{01}$ 2.55) and did not harm posttest scores ($BF_{01}$ 3.17) relative to no explicit scaffolding during problem-solving prior to instruction. Further, similar levels of induced frustration ($BF_{01}$ 3.29) and curiosity ($BF_{01}$ 3.27) were observed across the conditions.