Sometimes you come across a study that feels both revolutionary in its research design and instantly recognisable to anyone who has been working on classroom management for years. The study by Shi and colleagues from China fits that description perfectly. The researchers wanted to know how closely teachers’ beliefs align with what they say they do. They also wanted to find out how those beliefs align with what teachers actually do in the classroom.
They combined questionnaires with video recordings of 673 lessons and used an AI system that automatically codes words, tone and movement. The study involved 345 teachers and three types of data: teachers’ sense of classroom management self-efficacy, their self-reported strategies and AI-observed behaviour. That behaviour includes, for example, how often a teacher offers praise, the tone of voice they use and how much time they spend moving among their pupils. The dataset is undeniably thorough.
The first finding is predictable. Teachers with high classroom management self-efficacy report using effective strategies more often. They say they give more positive feedback, work more preventively and provide more precise instructions. That aligns well with earlier research. But once the researchers looked at the behaviour that actually appeared on video, the picture changed.
The AI detected very few differences between teachers with high and low self-confidence. The frequency of praise, the number of corrective comments, the warmth of the teacher’s voice and the time spent among pupils all seemed to differ only minimally. There was a slight indication that teachers with lower confidence used disciplinary language a little more often. However, even that did not reach statistical significance.
This confirms something many teachers intuitively recognise. What you think you do and what you really do in the moment are not always the same. Self-report remains valuable because it provides insight into intentions. However, people often answer such questions too positively. Observation gets closer to reality, yet even observation never captures everything. The study clearly illustrates the tension. Intentions, beliefs and behaviour do not line up neatly. Only in the case of praise was there a clear link between what teachers said and what they did. For other strategies, the connection largely disappeared. That may seem like a small point, but it highlights how complex classroom management is and how little linearity there is between knowing, wanting and doing.
The researchers rightly point out the limitations of their AI system. Small gestures, subtle relational cues and rare but important interventions are easily missed. Even so, the combination of questionnaires and AI-based observation provides a richer picture than we have had so far. It makes the gap between intention and action visible. That is precisely what makes this approach promising for professional development. It encourages conversations not only about strategies and beliefs, but also about what actually happens. Reflection gains strength when it is grounded in concrete behaviour.
Image made by AI.