AI Insight
This study reanalyzed survey data from 301 Chinese university students regarding AI acceptance and found that five theoretically separate dimensions (such as perceived usefulness and readiness) could not be empirically distinguished from each other. The questionnaire items measuring different aspects of AI acceptance were so highly correlated (0.98-1.00) that they essentially measured one underlying factor rather than five distinct constructs. While the combined measure strongly predicted students' intention to continue using AI, the results indicate the original measurement instrument failed to capture independent psychological mechanisms of AI acceptance.
Why it matters
This methodological critique highlights measurement problems in AI acceptance research, suggesting that many studies may be claiming to measure distinct factors when their surveys actually capture a single general attitude. Researchers developing AI adoption instruments need stronger discriminant validity to identify which specific factors drive technology acceptance and continuation.
Understand the Science
PurposeThis measurement-aware secondary analysis examines whether five theoretically distinct AI-acceptance and continuance domains are empirically separable in 301 records from a public dataset described by its repository as university-student responses from mainland China.MethodThe analysis used 301 records in a public workbook described by its repository as responses from university students in mainland China. Five 4-item questionnaire blocks were reconstructed from the public workbook. The measurement audit combined internal-consistency assessment, HTMT, principal-component analysis, one- versus five-factor CFA benchmarks, and a collapsed descriptive 16-item summary score. HC3-robust regressions were retained as descriptive composite benchmarks, while bootstrap and fsQCA analyses were treated as supplementary diagnostics. Perceived practical usefulness refers to respondents’ perceived career, efficiency, competitiveness, and problem-solving value rather than verified learning, employment, or institutional outcomes.ResultsThe five domains were internally reliable but not empirically distinct. HTMT estimates ranged from 0.983 to 1.007, the first unrotated component explained 58.75% of item variance, and the five-factor CFA provided negligible improvement over the one-factor model, with latent correlations ranging from 0.980 to 0.997. A collapsed 16-item descriptive composite was strongly associated with continuance intention (β = 0.896, R2 = 0.805). The regression coefficients therefore describe how shared variance is distributed across the questionnaire composites rather than independent acceptance mechanisms. fsQCA produced one high-readiness configuration and its symmetric low-readiness counterpart, with no evidence of equifinality or configurational asymmetry.ContributionThe study provides a reproducible measurement critique of a public AI-acceptance dataset. Its central finding is inadequate discriminant separation among the five affirmative questionnaire blocks. Their shared covariance can be summarized descriptively, but the present data cannot determine whether that common dimension is substantive, method-related, or both, and it should not be treated as a newly validated AI-readiness construct.