MEASUREMENT INVARIANCE OF AN AI-GAMIFIED ADAPTIVE LEARNING ACCEPTANCE SCALE ACROSS ENGLISH PROFICIENCY AND YEAR OF STUDY

Ainur Zhorabekova, Marzhan Moldakassymovna Duisenova, Moldir Kurmasheva, A. Yesnazar

2026

Closed access Metadata (CSL-JSON)
DOI 10.63597/uto3105-4161.2026.7.3.014
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Students' acceptance of adaptive learning tools that combine artificial intelligence with gamification is usually measured with multidimensional questionnaires, but whether such questionnaires show the same measurement properties across student subgroups is rarely examined. This study tested the measurement invariance of a 13-item scale of acceptance of AI-gamified adaptive learning across English proficiency and year of study and estimated latent mean differences where invariance held. Previously collected data from 401 EFL students at a Kazakhstani pedagogical university were analysed with multi-group confirmatory factor analysis. A second-order factor model served as the primary measurement model, and a four-correlated-factor model estimated with MLR and WLSMV was used as a sensitivity analysis. Across proficiency levels A2, B1 and B2+, the scale showed full scalar invariance, with CFI decreasing by no more than .003 and no significant chi-square difference at any step, and the latent mean of general acceptance did not differ between levels. Across year of study, comparing first-year students with those in years two to four, metric invariance was supported and scalar invariance was partial. Once the intercept of one item concerning AI feedback was freed, constrained and unconstrained models no longer differed significantly. Within the reported partial invariance model, the latent mean of general acceptance was higher for upper-year students than for first-year students by about0.3 standard deviations. The findings provide a full psychometric basis for comparisons across proficiency and a partial basis for latent mean comparisons across year of study. The main limitations are a single-institution convenience sample, self-reported proficiency, a ceiling effect and the possibility of common-method variance.