The Effect of Different Realism Designs of Virtual Agents on Students’ Grades in Quiz App: The Mediating Role of Arousal and Valence

This study investigates the impact of various realism designs of 2D virtual agents on students' grades in a quiz app, focusing on the mediating roles of arousal and valence. A total of 134 electrical engineering students from four Malaysian polytechnics participated in this experiment. Four different types of virtual agents—cartoon-like, stylized, semi-realistic, and realistic—were embedded into the quiz app, each designed to guide students through answering 20 multiple-choice questions. Valence and arousal levels were measured using the Self-Assessment Manikin (SAM) questionnaire, and the quiz grades were recorded. The findings revealed that virtual agents with higher levels of realism (semi-realistic and realistic) were negatively associated with student grades, while the cartoon-like agent led to better quiz performance. Mediation analysis showed that both arousal and valence significantly influenced the relationship between virtual agent design and grades. The results suggest that overly realistic virtual agents may cause cognitive overload, thus hindering academic performance. In contrast, simpler, cartoon-like agents reduce distractions and improve learning outcomes. These findings have implications for designing virtual agents in educational settings, recommending that simplicity and reduced realism can enhance learning efficiency by minimizing emotional distraction and cognitive load.