Modeling Factors Influencing Learning Gains in Generative Artificial Intelligence-Supported Undergraduate Programming Courses

The integration of Generative Artificial Intelligence (GenAI) into programming education has introduced new opportunities and challenges for the teaching–learning process in higher education. While GenAI tools can support personalized learning and immediate feedback, their effectiveness depends on how they are pedagogically integrated into instructional environments. This study analyzes the relationship between GenAI-supported instructional strategies, academic context, affective factors, and demographic variables on normalized learning gain in undergraduate introductory programming courses. A multi-group quasi-experimental pre-test–post-test design was conducted involving 648 engineering students distributed across 53 course groups taught by 32 professors from 10 campuses. Four GenAI-supported instructional strategies were implemented and compared with traditional instructional conditions. Learning gains were analyzed through a linear mixed-effects model considering instructional strategy, academic program, student frustration, future intention to use GenAI, student gender, and professor gender as fixed effects, while group/professor was modeled as a random effect. Results showed that instructional strategy was the strongest predictor of learning gain, followed by academic program and student frustration. In contrast, demographic variables and future intention to use GenAI did not show significant associations with learning outcomes. The findings suggest that learning gains in programming education depend primarily on instructional design and pedagogical mediation rather than on students’ demographic characteristics or technological predisposition.