Evaluating Vibe Coding in Programming Education: Evidence from a Quasi-Experimental Study in Vocational IT Training
Programming courses at the college level in Vietnam continue to suffer from high failure rates, early drop-out, and incomplete student projects. This paper presents Vibe Coding, a four-stage pedagogical model integrating AI coding assistants into Fundamentals of Programming and Basic Web Design. The model was evaluated using a hybrid quasi-experimental design combining a three-year historical trend (2023–2025, n = 735) with a concurrent control group in the 2025–2026 academic year (232 traditional vs. 150 Vibe Coding students). The Vibe-Coding group significantly outperformed the traditional group on the proportion scoring ≥7.0/10 (53.3% vs. 41.8%; z = 2.21, p = .027) and on course pass rate (83.3% vs. 67.2%; z = 3.48, p < .001), and also exceeded the three-year historical baseline (39.7%; z = 3.08, p = .002), while the traditional group did not differ from that baseline. However, this overall effect was driven almost entirely by Fundamentals of Programming (z = 3.13, p = .002); in Basic Web Design, Vibe Coding did not outperform traditional instruction on the merit measure (z = -0.24, p = .811), with only a non-significant favourable trend on pass rate (z = 1.66, p = .097). Class-level variation was substantial, including one traditional section that improved by 40 percentage points without the intervention. Despite this course-level heterogeneity, over 85% of surveyed students reported greater engagement and confidence in debugging. The findings indicate that Vibe Coding can meaningfully improve programming outcomes, but its benefits are course-dependent and require further validation through randomized, multi-institutional studies.