Study Clarifies How Digital Literacy and Creative Disposition Shape Student Performance

A quantitative cross-sectional investigation into teacher preparation has revealed that technical proficiency with generative artificial intelligence tools does not directly translate into enhanced academic success. Conducted among 258 pre-service Visual Arts teachers across seventeen teacher-education colleges in Ghana, the study examined the structural interrelationships connecting general digital literacy, generative artificial intelligence literacy, creative thinking disposition, and perceived academic performance. Utilizing partial least squares structural equation modeling alongside importance-performance map analysis and necessary condition analysis, the researchers determined that generative artificial intelligence literacy exhibited no statistically significant direct net association with perceived academic performance. Instead, specialized artificial intelligence literacy operated through an indirect pathway completely mediated by creative thinking disposition. Concurrently, foundational digital literacy demonstrated robust, direct positive associations with both artificial intelligence literacy and perceived academic achievement, establishing that broad digital competence continues to function as an indispensable educational foundation.

The conceptual structure of the investigation was grounded in an integrative synthesis of Connectivism and Self-Determination Theory, deployed by the researchers as complementary interpretive frameworks to articulate a capability-enactment perspective. Connectivism conceptualizes learning within digitally mediated landscapes as an active process of navigating, establishing, and sustaining vital connections across distributed information networks, technological instruments, and collaborative human systems. Within this orientation, general digital competence and specialized generative artificial intelligence literacy function as operational capabilities that facilitate productive participation across distributed networks rather than constituting academic achievements in isolation. Complementing this epistemic view, Self-Determination Theory highlights the fundamental psychological conditions supporting autonomous motivation and intentional learning engagement. The authors noted that possessing technical capabilities does not guarantee that pre-service educators will purposefully deploy them toward rigorous learning goals. Because the study did not empirically measure psychological need satisfaction or network topologies, these theories served as interpretive conceptual mechanisms rather than directly evaluated causal pathways.

The empirical research design targeted pre-service Visual Arts educators enrolled in Ghanaian Colleges of Education, utilizing institutional contacts and convenience sampling across seventeen participating institutions. Prior to data gathering, an a priori statistical power calculation was carried out using G*Power version 3.1.9.7 based on an F-test for linear multiple regression with four focal predictor terms: digital literacy, generative artificial intelligence literacy, creative thinking disposition, and their interaction. Specifying a medium effect size of f-squared equal to 0.15, an alpha level of 0.05, and statistical power of 0.95, the calculation established a minimum necessary threshold of 129 respondents, which was elevated to 142 to accommodate potential administrative attrition. The research team collected data between August 4 and September 5, 2025, using concurrent online administration via Google Forms and physical paper-and-pencil surveys to maximize participation. The final analyzed sample comprised 258 participants, of whom 69.4 percent were male and 64.3 percent resided in urban settings, with a mean sample age of 23.5 years.

Construct operationalization relied on four established psychometric measurement scales adapted to the context of visual arts teacher education, alongside a social desirability response marker scale. Creative thinking disposition was gauged using the Marmara Creative Thinking Disposition Scale, originally comprising twenty-five items across six dimensions, which was refined after measurement evaluation to seven robust indicators yielding a Cronbach alpha of 0.838. Digital literacy was assessed through an adapted ten-item Digital Literacy Scale spanning cognitive, technical, and socioemotional facets, using a seven-point response continuum that was subsequently refined to seven indicators demonstrating a Cronbach alpha of 0.841. Generative artificial intelligence literacy utilized an adapted twelve-item Artificial Intelligence Literacy Scale spanning awareness, usage, evaluation, and ethics, reduced after psychometric screening to six indicators with a Cronbach alpha of 0.817. Perceived academic performance was measured using a four-item instrument with an alpha of 0.814, preferred over institutional cumulative grade point averages due to unverified grading equivalence across institutions.

Rigorous procedural screening and common-method variance diagnostics were instituted prior to structural evaluation. To filter out unengaged or careless submissions across both administration modes, the questionnaire embedded four instructional attention checks requiring specific answer selections across different construct modules. Of the 364 total completed survey packets received by the research team, 106 submissions failed at least three instructional checks, resulting in an exclusion rate of 29.1 percent and yielding the final verified sample of 258 respondents. Potential common-method variance and socially desirable response distortions were subsequently assessed through two complementary statistical procedures. A full-collinearity evaluation was executed by regressing a randomly generated criterion variable on all focal latent constructs, which produced inner variance inflation factor values strictly below the conservative 3.3 benchmark. Furthermore, bivariate correlations between the focal constructs and the seven-item social desirability response marker scale remained minimal, ranging from negative 0.194 to negative 0.062, indicating an absence of severe method contamination.

Evaluation of the first-order reflective measurement model confirmed acceptable indicator reliability, composite internal consistency, convergent validity, and discriminant distinctiveness across all constructs. All latent variables achieved average variance extracted values exceeding the recommended 0.50 threshold, verifying that each construct explained more than half of the variance across its retained manifest indicators. Although several individual indicator loadings fell between 0.40 and 0.70, these items were retained to preserve necessary theoretical domain breadth without compromising overall construct validity. Discriminant validity was examined via the heterotrait-monotrait ratio of correlations, which remained below the conservative 0.90 threshold across construct pairs, with none of the ninety-five percent bootstrap confidence intervals encompassing unity. However, the researchers observed that the heterotrait-monotrait ratio between creative thinking disposition and perceived academic performance reached 0.894, with an upper bias-corrected confidence interval of 0.930. This documented statistical proximity signaled an empirical measurement limitation that warranted measured, non-inflated interpretation of their structural association.

Structural model estimation using SmartPLS 4 and ten thousand bias-corrected bootstrap subsamples explained 66.2 percent of the variance in perceived academic performance, 58.8 percent in creative thinking disposition, and 46.5 percent in generative artificial intelligence literacy, accompanied by a standardized root mean square residual of 0.081. Digital literacy exhibited substantial positive structural associations with generative artificial intelligence literacy with a beta coefficient of 0.682, with creative thinking disposition at a beta of 0.350, and directly with perceived academic performance at a beta of 0.326. Creative thinking disposition demonstrated a substantial positive direct relationship with perceived academic performance, recording a beta of 0.481. In sharp contrast, the hypothesized direct path connecting generative artificial intelligence literacy to perceived academic performance proved completely non-significant, yielding a beta coefficient of 0.001 and a p-value of 0.986. Among demographic control variables included in the structural specification, participant age displayed a modest positive link with creative thinking disposition, whereas gender and academic progression level showed no statistically meaningful associations.

Mediation analyses conducted within the structural model clarified the precise indirect mechanisms linking technological competencies to performance outcomes. Evaluating the pathway connecting digital literacy to creative thinking disposition revealed complementary mediation, as the indirect transmission through generative artificial intelligence literacy was positive and highly significant, yielding a beta of 0.333 and a t-statistic of 6.745, alongside a persistent direct link. More critically, the indirect sequence connecting generative artificial intelligence literacy to perceived academic performance through creative thinking disposition demonstrated an indirect-only mediation pattern, characterized by a statistically significant indirect effect with a beta of 0.235, a t-statistic of 5.083, and a p-value below 0.001. In the complete absence of a direct net relationship, this indirect pathway indicates that technical generative artificial intelligence competence relates to perceived academic success exclusively when accompanied by learner tendencies toward flexible exploration, problem reframing, and reflective cognitive engagement.

The structural analysis also evaluated the moderating role of generative artificial intelligence literacy on the relationship between digital literacy and perceived academic performance. Incorporating the latent product interaction term produced a small but statistically significant negative moderating coefficient of negative 0.076 with a t-statistic of 2.741 and a p-value of 0.006, expanding explained outcome variance by exactly 1.0 percent. Simple slope analysis illustrated that the positive association between general digital literacy and academic performance was marginally stronger among students reporting lower artificial intelligence literacy, while showing slight attenuation among those with higher artificial intelligence capability. Robustness checks evaluating prospective nonlinear relationships confirmed a statistically significant negative quadratic effect for the link connecting digital literacy to generative artificial intelligence literacy, recording a beta of negative 0.067. This curvature demonstrated that while broader digital competence supports artificial intelligence literacy, marginal gains diminish at elevated levels, affirming that generative competencies represent specialized capabilities requiring distinct curricular interventions.

Out-of-sample predictive capacity, managerial decision priorities, and prerequisite operational thresholds were systematically mapped using PLSpredict, importance-performance map analysis, and necessary condition analysis. PLSpredict with ten-fold cross-validation confirmed positive Q-squared values across all perceived performance indicators, yet only one indicator exhibited lower root mean square error relative to the linear model benchmark, demonstrating mixed out-of-sample predictive superiority. Importance-performance mapping established digital literacy as the highest overall priority for instructional intervention, combining the largest total structural importance value of 0.653 with a comparatively modest performance score of 73.029. In complementary fashion, necessary condition analysis utilizing ceiling envelopment-free disposal hull estimation revealed that digital literacy, creative disposition, and generative literacy each exhibited medium necessity effect sizes constraining academic attainment. Specifically, attaining an 85 percent perceived performance level established sample-specific bottleneck thresholds requiring minimum baseline scores of 57.186 percent in creative thinking disposition, 42.924 percent in digital literacy, and 43.855 percent in generative artificial intelligence literacy.

The study authors highlighted several methodological boundaries, including non-probability sampling from Ghanaian colleges, reliance on cross-sectional self-reported survey indicators without longitudinal tracking, substantial psychometric item reductions, and unmodeled institutional clustering. The collective findings caution educational policymakers against viewing generative artificial intelligence tools as self-contained catalysts of learning achievement. Instead, professional teacher preparation in visual arts disciplines should emphasize a balanced curricular sequence that reinforces foundational digital competencies in source verification and information management, deliberately nurtures creative thinking dispositions through studio-based inquiry, and positions generative artificial intelligence instruction within rigorous evaluative frameworks prioritizing ethical reflection, prompt optimization, and independent disciplinary reasoning.

Subject of Research: Structural equation modeling of digital literacy, generative AI literacy, creative disposition, and academic performance

Article Title: Digital literacy, GenAI literacy, and creative thinking disposition in relation to pre-service visual arts teachers’ perceived academic performance

Article References: Setsoafia, P., Essel, H. B., Tachie-Menson, A., Johnson, E. E., Appiah-Kubi, J., Koduah, C. A., Nabang, M., Acheampong, E. B., & Aboagye, G. A. (2026). Digital literacy, GenAI literacy, and creative thinking disposition in relation to pre-service visual arts teachers’ perceived academic performance. Discover Education, 5(1), Article 1302. https://doi.org/10.1007/s44217-026-02396-8

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02396-8

Keywords: digital literacy, generative artificial intelligence literacy, creative thinking disposition, perceived academic performance, teacher education, visual arts education, partial least squares structural equation modeling, necessary condition analysis, Digital, literacy, GenAI, creative