Student engagement and self-regulated learning across standard and AI-enhanced flipped learning: a within-subjects...

Although AI is increasingly being integrated into flipped classrooms, little is known about how student engagement and self-regulated learning (SRL) develop across extended implementations, particularly in underrepresented higher education contexts. This study examined developmental trajectories of student engagement and SRL across a sequential implementation comprising a standard flipped phase followed by a progressively scaffolded AI-enhanced phase, representing one of the earliest longitudinal investigations of AI-enhanced flipped learning in a Central Asian higher education context.

Using a within-subjects longitudinal design, 211 undergraduate students at a public university in Mongolia were followed across four measurement occasions (Weeks 1, 6, 11, and 15) within a 16-week course, with quantitative data collected across 15 weeks of instruction. Engagement and SRL were assessed across a standard flipped phase (Weeks 1-6) and a progressively scaffolded AI-enhanced phase (Weeks 7-15). Metacognitive self-regulation was examined as a potential mediator of the longitudinal relationship between agentic engagement and academic achievement using a cross-lagged panel model (Mplus 8.3, ML estimator, 1,000 bootstrap samples).

Linear mixed-model analyses revealed significant trajectories for all engagement and SRL subscales (all p ≤ 0.004, η² _p = 0.021-0.111). All six SRL subscales remained stable during the standard flipped phase before increasing significantly during the AI-enhanced phase. Agentic engagement demonstrated the strongest engagement trajectory (η² _p = 0.103), whereas metacognitive self-regulation showed the largest SRL effect (η² _p = 0.111), with the greatest gains observed during the W11-W15 interval. Cross-lagged analyses identified a significant longitudinal association between agentic engagement at Week 1 and metacognitive self-regulation at Week 6 (β = 0.187, p = 0.005), whereas the indirect effect on academic achievement was not significant (β = 0.002, 95% BC CI [-0.028, 0.035]).

The findings describe distinct longitudinal trajectories of engagement and SRL across sequential instructional phases and support a developmental interpretation in which standard flipped instruction and progressively scaffolded AI integration may play distinct roles in supporting student engagement and SRL. This interpretation warrants evaluation in future controlled longitudinal studies.