Psychometric validation of the Chinese generative artificial intelligence acceptance scale among university students

In an era where artificial intelligence has rapidly permeated higher education, researchers and educators face a critical paradox: widespread student use of generative AI does not inherently translate to stable, genuine technological acceptance. Addressing this empirical gap, a recent cross-sectional psychometric validation study published in Frontiers in Psychology has successfully adapted and evaluated the 20-item Generative Artificial Intelligence Acceptance Scale for use within mainland China. Led by researchers XT and QZ, the study evaluates how Chinese university students perceive and embrace rapidly evolving AI tools, providing a vital empirical foundation for future educational technology research.

Background and Context of AI Adoption in Higher Education

The proliferation of generative AI tools such as ChatGPT, DeepSeek, and Doubao has fundamentally altered the academic landscape. However, educational researchers emphasize that raw usage metrics—such as a 2025 UK survey revealing that 92% of full-time undergraduates actively utilize AI tools—fail to capture the psychological nuances of user intent, trust, and perceived utility.

Unlike Western academic environments where global platforms dominate, Chinese higher education operates within a distinct technological ecosystem. While platforms like ChatGPT and Gemini maintain a significant user base (reported at 50.8% and 15.6% among the surveyed cohort, respectively), domestic powerhouses such as Doubao (70.9%), DeepSeek (50.5%), and Tongyi Qianwen (43.2%) are heavily integrated into daily student workflows. These domestic applications often feature varying user interfaces, distinct access protocols, and unique regulatory frameworks. Consequently, researchers needed to determine whether established Western-centric or international technology acceptance frameworks could accurately measure students’ psychological stances within this localized ecosystem.

Methodological Framework and Study Timeline

The research protocol received formal institutional approval from the Ethics Committee of Yunnan Minzu University (Approval No. XYLL2026009). The adaptation process commenced with a meticulous translation and back-translation procedure supervised by bilingual experts in educational technology, followed by academic reviews for cultural and conceptual equivalence. A pilot test involving 20 university students refined minor wording ambiguities without altering core item definitions.

Data collection utilized convenience sampling across six higher education institutions spanning three Chinese provinces: Yunnan Minzu University and Yuxi Normal University in Yunnan Province; Zhejiang Chinese Medical University and Jinhua University of Vocational Technology in Zhejiang Province; and Huainan Normal University and Hefei University in Anhui Province. Distributed via the Wenjuanxing digital survey platform, the initial call yielded 632 submissions. Following rigorous screening procedures that eliminated duplicate IP addresses, straightlining responses, and surveys completed in under two seconds per item, a final sample of 604 valid questionnaires (representing a 95.57% retention rate) was established.

To ensure statistical rigor, the 604 participants were randomly divided into two equal subsamples of 302 respondents each:

• Subsample 1: Utilized for Exploratory Factor Analysis (EFA) to uncover underlying data structures.
• Subsample 2: Reserved strictly for Confirmatory Factor Analysis (CFA) to cross-validate the factor model against independent observations.

Empirical Findings and Psychometric Performance

The empirical results affirmed the psychometric robustness of the adapted 20-item instrument. Corrected item-subscale correlations ranged from 0.555 to 0.774, indicating strong item discrimination across all measures. Furthermore, Welch’s independent-samples $t$-tests demonstrated significant score disparities between upper and lower 27% groups across all items ($p < 0.001$).

EFA using principal axis factoring with Promax rotation extracted a four-dimensional structure explaining 52.22% of total variance. The identified dimensions mapped precisely onto the Unified Theory of Acceptance and Use of Technology (UTAUT) framework:

1. Performance Expectancy (PE): 7 items, measuring perceived usefulness and academic benefits.
2. Effort Expectancy (EE): 5 items, evaluating the ease of learning and operating AI tools.
3. Facilitating Conditions (FC): 3 items, assessing access to compatible technology, resources, and institutional support.
4. Social Influence (SI): 5 items, capturing peer and societal expectations regarding AI usage.

Subsequent CFA conducted on Subsample 2 confirmed that the correlated four-factor model demonstrated exceptional fit indices (scaled $chi^2(164) = 180.41, p = 0.180$, robust $textCFI = 0.993$, robust $textTLI = 0.992$, robust $textRMSEA = 0.020$, and $textSRMR = 0.035$). This four-factor model significantly outperformed an alternative three-factor model that combined performance expectancy and effort expectancy ($Deltachi^2(3) = 102.11, p < 0.001$).

Reliability measures further substantiated the scale’s stability. Cronbach’s alpha coefficients spanned from 0.757 to 0.873, ordinal McDonald’s omega values ranged from 0.810 to 0.905, and composite reliability metrics reached 0.773 to 0.888. Multigroup CFA testing across gender subgroups established strict measurement invariance, confirming that the scale operates equivalently for both male and female student populations.

Implications and Future Directions for Educational Research

While the study provides robust validation for the Chinese adaptation of the scale, the authors noted specific psychometric caveats. The Average Variance Extracted (AVE) for performance expectancy sat marginally below the conventional 0.50 threshold at 0.494, and discriminant validity analyses revealed minor conceptual overlap between performance expectancy and facilitating conditions. Consequently, researchers recommend utilizing the four subscale scores independently for targeted diagnostic assessments rather than relying solely on a consolidated total score, which should be reserved strictly for broad descriptive summaries.

The successful validation of this instrument carries substantial implications for policymakers, academic institutions, and ed-tech developers. By offering a reliable tool to measure general generative AI acceptance, universities can now accurately diagnose student readiness, assess the impact of institutional resource allocation, and design targeted digital literacy programs. As artificial intelligence continues to reshape higher education pedagogy across China, this psychometric instrument provides a reliable baseline for longitudinal tracking and cross-institutional comparative studies.