Perceived educational value, ethical risk, and support for responsible AI governance in higher education: an embedded mixed-methods study

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Introduction Generative Artificial Intelligence (GenAI) is becoming increasingly embedded in higher education assessment, yet the perceptual factors that motivate stakeholder support for its institutional governance remain underexamined. This study investigated the extent to which Perceived Educational Value (PEV) and AI Ethical Risk Perception (AERP) are associated with Responsible AI Governance Support (RAGS) among higher education stakeholders. Methods An embedded mixed-methods survey design was employed. Quantitative Likert-scale data and qualitative open-ended responses were collected concurrently through one structured questionnaire; full-sample quantitative analyses were conducted first, followed by thematic analysis of the open-ended responses to contextualize and deepen interpretation of the quantitative findings. A structured questionnaire was administered to 192 participants (125 students and 67 instructors) recruited from public and private higher education institutions in the United Arab Emirates, with N = 192 yielding statistical power exceeding.99 for the structural model. Qualitative open-ended responses from 179 participants were subsequently analyzed using a codebook-based thematic analysis. Quantitative data were analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM). Results Participants reported above-midpoint levels of PEV, AERP, and RAGS. The structural model explained 59.1% of the variance in RAGS. Both PEV [ β = .255, p < .001, 95% bias-corrected CI (.117,.377)] and AERP [ β = .627, p < .001, 95% bias-corrected CI (.485,.748)] were positively associated with governance support, with AERP showing the larger association. AERP did not significantly moderate the PEV–RAGS relationship [ β = −.030, p = .370, 95% bias-corrected CI (−.098,.037)]. MICOM established full measurement invariance for PEV and AERP but not compositional invariance for RAGS; therefore, structural student–instructor comparisons involving RAGS were not interpreted. Thematic analysis identified responsible conditional integration as the central interpretive frame, with learning enhancement, governance expectations, academic integrity, over-reliance, and critical-thinking concerns among the salient themes. Discussion These findings provide preliminary empirical support for a model in which perceived educational value and ethical risk perception are each associated with governance endorsement. The stronger association between ethical risk perception and governance support suggests that transparent attention to ethical risks may be relevant to institutional governance communication, a possibility that requires intervention-based research. Practical implications include risk-aware AI literacy programs, transparent and contestable governance policies, stakeholder engagement, and human-in-the-loop oversight mechanisms in AI-mediated assessment.