Shi and colleagues combined questionnaire survey methods with AI-supported classroom behavior analysis to examine the relationships among teachers’ classroom management self-efficacy, self-reported classroom management strategies, and actual classroom management behaviors observed by AI. The study involved 345 Chinese K-12 in-service teachers, collected questionnaire data on their classroom management self-efficacy and strategies, and analyzed 673 valid classroom video recordings, totaling 461.74 hours. The research team developed an AI-supported multimodal classroom management behavior analysis tool that automatically identified teachers’ praise statements, criticism statements, discipline-related statements, positive tone of voice, proportion of proximity to students, and proportion of visual attention to students through text, audio, and image data. This allowed the researchers to test whether teachers’ beliefs, reported strategies, and actual behaviors were consistent.
The results showed that teachers with higher classroom management self-efficacy reported using all types of classroom management strategies more frequently, and these differences were all statistically significant. For example, in praise strategies, the high self-efficacy group had a mean rank of 214.43, significantly higher than the low self-efficacy group’s 128.84 (Z = –8.74, p < .001). In corrective feedback strategies, the high group had a mean rank of 206.27, compared with 137.54 in the low group (Z = –6.67, p < .001). In preventive management strategies, the high group had a mean rank of 221.84, whereas the low group had 120.95 (Z = –9.81, p < .001). In commands/transition strategies, the high group had a mean rank of 220.69, compared with 122.17 in the low group (Z = –9.68, p < .001). However, when actual classroom videos were analyzed through AI, no significant differences emerged between the high and low self-efficacy groups on most observable classroom management behaviors. The only finding was a marginal tendency for teachers with lower self-efficacy to use more discipline-related statements (p = .07), suggesting that they may rely more on disciplinary language to maintain order.
Regarding the relationship between self-reported strategies and AI-observed behaviors, the results showed that consistency existed only in some domains. Teachers’ self-reported praise strategies were significantly positively related to AI-detected praise statements and positive tone of voice, although the effect sizes were relatively small. In contrast, corrective feedback and preventive management strategies were not significantly associated with their corresponding AI-based behavioral indicators. Notably, self-reported commands/transition strategies were significantly negatively related to AI-observed discipline-related statements, meaning that the more often teachers reported using clear instructions and transition management, the less frequently discipline-related statements appeared in their classrooms. Overall, teachers’ reported use of strategies only partially corresponded to their actual classroom behaviors, and more concrete and observable strategies, such as praise, were more likely to be validated by AI observation.
Overall, this study suggests that teachers’ beliefs, strategies, and behaviors in classroom management are not always highly aligned. Although teachers with higher self-efficacy reported using more effective strategies, they did not necessarily demonstrate clearly different classroom management behaviors in practice. Only specific and easily identifiable strategies, such as praise, were more readily confirmed through AI observation. The study therefore highlights a misalignment between teachers’ beliefs and their actual teaching behaviors, while also demonstrating the considerable potential of AI for large-scale, non-intrusive, and evidence-based research on classroom management. For educational research and teacher development, the study serves as a reminder that relying solely on teachers’ self-report questionnaires may overestimate the consistency between reported strategies and real classroom behavior. Future work should combine self-report data with more objective methods such as AI observation to gain a fuller understanding of actual classroom management practices.
Source (Open Access): Shi, Y., Wang, Z., Chen, Z., Ren, D., Liu, H., & Zhang, J. (2026). Do teachers practice what they believe? Exploring discrepancies between teachers’ classroom management self-efficacy, self-reported strategies, and AI-observed behaviors in K-12 education. Teaching and Teacher Education, 169, 105280.