Data How GenAI Pedagogical Agent Interactions Shape Higher-Order Thinking: Engagement Mechanisms and Configurational Pathways
The participants of this study were university students who had experience using GenAI learning tools, such as Doubao, DeepSeek, Kimi, and ChatGPT. A combination of convenience sampling and snowball sampling was adopted, and the questionnaire was distributed through online channels. Data were collected from May 21 to June 5, 2026. A total of 756 questionnaires were returned. To ensure that the sample matched the research context, a screening question was placed at the beginning of the questionnaire: “Have you used generative AI to support your learning in the past month?” Respondents who answered “No” were directed to the end of the questionnaire and excluded from subsequent analysis. Based on this criterion, 29 responses were removed. During data cleaning, the research team further excluded invalid responses, including questionnaires with completion times shorter than 60 seconds and straight-lining responses, defined as selecting the same option for 10 consecutive items. A total of 159 responses were removed during this process. Finally, 568 valid questionnaires were retained for analysis, yielding an effective response rate of 75.1%.