Community of inquiry pathways link project-based learning design to student engagement

A two-wave survey of 490 upper-year animation students in China has found that the three dimensions of the Community of Inquiry framework — cognitive, social, and teaching presence — were significantly associated with learning engagement across all twelve indirect pathways linking project-based learning design to student engagement. The study, conducted in studios that use AI-generated visual tools, also revealed a strikingly conditional pattern for project evaluation: its association with engagement flipped from positive to negative depending on the statistical model, yet evaluation appeared as a core or peripheral condition in five of seven configurations linked to high engagement. The findings were published in Frontiers in Psychology.

The research context is an animation education system at a critical juncture. Generative artificial intelligence tools such as Midjourney and Stable Diffusion can now complete in minutes visual work that traditionally required weeks, rapidly eroding the relative value of the production skills on which animation training has long relied. When students can generate visual solutions almost instantly, they face a tension between efficiency and critical judgment: they must use AI to extend their creative capabilities while independently evaluating its outputs.

Project-based learning, which emphasizes authenticity, autonomy, and collaboration, has long been regarded as an effective strategy for developing complex capabilities, but targeted empirical evidence on which of its individual design dimensions still foster deep engagement in AI-enhanced studios has remained limited. The research team, based at the School of New Media at Beijing Institute of Graphic Communication, surveyed third- and fourth-year animation, digital media art, and game design students at application-oriented undergraduate institutions using stratified convenience sampling across comprehensive universities and art and design institutions. At the first wave, 600 questionnaires were distributed and 550 returned; at the second wave, 490 successfully matched responses were obtained, a matching rate of 89.09 percent.

The final sample included 221 men (45.1 percent) and 269 women (54.9 percent), split almost evenly between third- and fourth-year students. Nearly half of the participants reported using AI tools one to two times per week, and just over half had more than twelve months of experience with them. All participants had completed studio courses using AI-generated visual tools during the 2025–2026 academic year.

Community of Inquiry theory holds that meaningful learning emerges from the interplay of three presences: cognitive presence, the construction and confirmation of meaning through sustained reflection and discourse; social presence, the capacity to present oneself socially and emotionally within a learning community; and teaching presence, the design and facilitation of cognitive and social processes. The authors argue that in AI-assisted studios this framework takes on new weight. When AI performs much of the production labor, teaching presence shifts from knowledge transmission toward designing cognitive scaffolds, social presence expands to include human–AI collaboration around AI-generated proposals, and cognitive presence may require deliberately scheduled reflection because technological acceleration compresses the time available for deep thinking.

Methodologically, the study combined partial least squares structural equation modeling in SmartPLS 4.0 with fuzzy set qualitative comparative analysis in fsQCA 3.0, a pairing that lets researchers estimate average statistical pathways while also identifying distinct combinations of conditions associated with an outcome. The questionnaire used a seven-point Likert scale, with 25 items covering the four project-based learning dimensions — Problem Driven, Project Design, Project Implementation, and Project Evaluation — 15 items for the three presences, and nine items for learning engagement covering behavioral, emotional, and cognitive investment. Perceptions of the learning dimensions and the presences were measured concurrently at the first wave, and learning engagement was measured three weeks later.

The structural model explained 52.4 percent of the variance in learning engagement, with satisfactory reliability, convergent validity, and predictive relevance across the constructs. The direct results were strikingly uneven. Problem Driven, Project Design, and Project Implementation each showed significant positive associations with learning engagement, with Project Implementation carrying the largest path coefficient at 0.305 and the largest effect size.

Project Evaluation, by contrast, showed a significant negative net direct association of −0.255 after controlling for the other learning dimensions and the three presences. Yet this figure demands careful reading. When Project Evaluation was the sole predictor of engagement, its coefficient was strongly positive at 0.483, matching its positive zero-order correlation.

The coefficient turned negative only when the other learning dimensions were added, and it remained negative in the full mediation model, where the total association was −0.113. The authors interpret this sign change as possible statistical suppression rather than evidence that evaluation itself inhibits engagement. All twelve indirect paths from the four learning dimensions through the three presences to engagement were statistically significant, tested with bootstrapping using 5,000 resamples.

Teaching Presence showed the highest point estimates of indirect association, ranging from 0.042 to 0.065, followed by Social Presence at 0.030 to 0.060, with Cognitive Presence comparatively weaker at 0.017 to 0.034. For Project Evaluation, the three indirect paths summed to 0.142, offsetting roughly 55.7 percent of the negative net direct coefficient — though not enough to reverse the direction of the total association after adjustment. The authors stress that because differences between indirect effects were not formally tested, no presence can be claimed as optimal, and because the learning dimensions and presences were measured concurrently at the first wave, these figures represent indirect statistical associations rather than mediation with a fully established temporal sequence.

The configurational analysis told a complementary story. No individual dimension met the consistency threshold of 0.90 for necessity — Social Presence came closest at 0.830 — meaning none was indispensable for high engagement. The intermediate solution identified seven sufficient configurations, with overall consistency of 0.924 and coverage of 0.667, jointly accounting for two-thirds of high-engagement cases.

Every configuration included at least one Community of Inquiry dimension as a core or peripheral condition, and three configurations included all three presences together. One pathway achieved the highest consistency of the seven, at 0.964, despite explicitly requiring the absence of Project Implementation, indicating that strong problem orientation, project design, and teaching presence can compensate when implementation is weak. Robustness checks splitting the sample into training and test halves of 245 cases each produced predictive consistency of 0.897 and coverage of 0.541, and varying the thresholds left the core conditions stable.

The study carries important limitations. The sample was restricted to upper-year students at application-oriented undergraduate institutions in eastern China, limiting generalization to other disciplines and educational levels. Because all learning dimensions and presences were measured at the same wave, reverse associations cannot be excluded — students' community experiences may shape how they retrospectively perceive course design.

Generative AI itself was a contextual feature and eligibility criterion, not a measured variable: the study captured neither the type, frequency, intensity, nor quality of AI use, nor included a comparison group with low AI use, so no conclusions can be attributed to AI itself. Discriminant validity between teaching presence and social presence was only partially supported, as their correlation of 0.751 exceeded the Fornell-Larcker criterion even though all HTMT ratios stayed below the conservative threshold of 0.85, and learning engagement was modeled as a single reflective construct rather than separate behavioral, emotional, and cognitive dimensions. The practical upshot is a call for differentiated design rather than uniform maximization.

Because high engagement was associated with multiple equifinal pathways, resource-constrained programs can strategically prioritize particular presences: courses with limited instructor capacity might invest in structured peer review and team project milestones to strengthen social presence, while independent creative courses with fewer peer interaction opportunities might prioritize instructor-led visual argumentation meetings to build teaching presence. The authors argue that evaluation should be integrated into a broader support system of cognitive, social, and instructional scaffolds rather than treated as an isolated control mechanism — a design principle grounded in the finding that evaluation's association with engagement depends on the conditions in which it is embedded, not on evaluation alone.

Subject of Research: Statistical associations between project-based learning design, community of inquiry dimensions, and learning engagement among animation students in AI-assisted studios

Article Title: How does community of inquiry matter? Mediating effects and configurational pathways linking project-based learning design to learning engagement

Article References: Xu, T., Qin, P., & Ma, S. (2026). How does community of inquiry matter? Mediating effects and configurational pathways linking project-based learning design to learning engagement. Frontiers in Psychology, 17, Article 1976588. https://doi.org/10.3389/fpsyg.2026.1976588

Image Credits: AI Generated

DOI: 10.3389/fpsyg.2026.1976588

Keywords: project-based learning, community of inquiry, learning engagement, animation education, generative artificial intelligence, cognitive presence, social presence, teaching presence, PLS-SEM, fsQCA, instructional design, educational psychology