Psychometric properties of the Russian version of the artificial intelligence anxiety scale: dimensionality, measurement invariance, and person-centered profiles

Introduction This study examined the psychometric properties of the Russian version of the 23-item, eight-dimensional Artificial Intelligence Anxiety Scale among undergraduates in Kazakhstan and Russia. Methods Participants were 877 undergraduates from ten universities. The sample was split within country and gender strata into exploratory ( n = 439) and confirmatory ( n = 438) subsamples. Analyses examined factor structure, bifactor reliability, and measurement invariance across gender and country. Sensitivity analyses assessed the stability of country comparisons, while exploratory latent profile analysis examined anxiety profiles and their associations with artificial intelligence familiarity and formal training. Results The original structure was not recovered: four dimensions concerning societal and existential threats collapsed into one. Although the eight-factor model showed better fit, it was rejected because four factor correlations exceeded .85. Removing three items yielded an acceptable five-factor, 20-item structure (CFI = .953, RMSEA = .047, SRMR = .043). Bifactor modelling yielded ωH = .875 and an ωH/ωtotal ratio of .918, indicating that approximately 92% of reliable total-score variance was attributable to the general factor. The three dimensions modelled as specific factors had insufficient residual variance to support subscale scoring. Strict invariance held across gender, and partial scalar invariance held across country. The country comparison for the two-item Algorithmic Unfairness dimension was indeterminate because its latent mean reversed depending on which intercept was released; the four longer dimensions were unaffected. Four exploratory profiles emerged, but the level component accounted for 48.5 times as much between-profile variation as shape. Artificial intelligence familiarity predicted profile membership; formal training did not. Discussion The findings support total-score use in this sample and suggest that artificial intelligence anxiety primarily reflects a continuous intensity dimension. Subscale interpretation requires caution, particularly for country comparisons involving Algorithmic Unfairness. Independent replication is needed to establish the stability of the revised structure and the exploratory profiles.