Digital learning anxiety and UTAUT predictors of university students' intention to use generative artificial intelligence in Ghana

Synthesis: Davor and colleagues extended the UTAUT model with digital learning anxiety and tested it on 350 undergraduates in the College of Humanities at one Ghanaian public university. Performance expectancy was the strongest predictor of intention to use generative AI (β = 0.377), effort expectancy followed (β = 0.281), and social influence was not associated with intention at all (β = −0.038). Anxiety reduced intention directly (β = −0.281) and also sat between effort expectancy and intention, so the students who found the tool harder to use were the more anxious ones and therefore the less likely to adopt it. The practical reading for a low-resource context is that access and ease of use are the adoption levers, and peer and lecturer opinion are not.

Key Findings

1. Performance expectancy drove intention most. Its standardized path to behavioral intention was β = 0.377 (B = 0.343, critical ratio 7.772, p < 0.001), the largest association in the model.
2. Effort expectancy mattered almost as much. The path was β = 0.281 (p < 0.001), which the authors read as evidence that a natural-language interface lowers the barrier that learning-management systems impose.
3. Social influence did not predict intention. β = −0.038, p = 0.413 — lecturers, peers and institutional norms added nothing once usefulness and ease of use were in the model, contrary to the original UTAUT proposition.
4. Anxiety suppressed intention directly. Digital learning anxiety predicted intention at β = −0.281 (p < 0.001), equal in size and opposite in sign to effort expectancy's path, and second only to performance expectancy.
5. Ease of use worked partly through anxiety. Effort expectancy reduced anxiety (β = −0.126, p = 0.017), and the indirect path from effort expectancy to intention through anxiety was β = 0.035 (95% CI [0.030, 0.099]).
6. The sample was one college, not one country's student body. 350 of 3,200 undergraduates in English, linguistics and communication studies, surveyed in paper form during scheduled lectures over three weeks.

What the model adds

UTAUT is a reasonable fit for a tool students can choose to adopt, and this study's contribution is the affective term. Most adoption work treats resistance as a belief problem — students doubt the tool's usefulness — and the coefficient here says the stronger negative is emotional: anxiety about learning with the tool reduces intention over and above how useful or easy students find it.

The pathway is the useful part. Effort expectancy did not only raise intention directly; it lowered anxiety, which in turn raised intention. That makes ease of interaction a mental-health-adjacent design property rather than a convenience, and it suggests that demonstrations, Scaffolding and low-stakes first tasks may do more for adoption than testimonials about performance.

Why social influence failed

The null on social influence is the finding most likely to be context-dependent, and the authors say as much. They did not measure which social influences students actually encountered — lecturer expectations, peer norms, departmental practice — so the construct may have been too diffuse to register in this population. A single university college with a strong shared program culture is also exactly the setting where a social-influence measure might saturate or wash out.

Read alongside the corpus's other adoption work, the result is best treated as a caution against assuming that institutional endorsement moves students. Usefulness and ease remain the levers this model can see.

How far it generalizes

The sample comes from one college in one public university, recruited by availability during lectures, so students absent on the survey days are missing — a bias the authors state. The design is cross-sectional, so the anxiety pathway is a statistical association rather than a demonstrated temporal mediation, and the model omits gender, level, prior AI experience, device access and English proficiency, all of which plausibly affect both anxiety and intention.

The English-medium setting also narrows the claim: these students study in the language most generative AI tools perform best in, which is not true of every multilingual system in the region.

What this means for practice

• Sell usefulness with evidence from the students' own courses. Performance expectancy carried the largest path, so examples drawn from their assessments will move intention more than general claims about AI skills.
• Lower the effort barrier first. Effort expectancy both raised intention and reduced anxiety, which makes a guided first task with a working device more valuable than a lecture on AI's potential.
• Treat anxiety as an adoption variable, not a personality trait. Anxiety reduced intention more than ease of use raised it, so induction and low-stakes practice are adoption work.
• Do not rely on endorsement. Social influence was null here; a policy statement or a lecturer's recommendation is not a substitute for making the tool useful and easy in context.
• Administrators in low-resource settings. Device access and English-medium instruction were not modeled, and both condition every coefficient above; plan infrastructure before campaigns.

Limitations

• Cross-sectional and self-report. All constructs measured once, by questionnaire, so no temporal or causal mediation is established.
• One college at one university. 350 of 3,200 undergraduates in humanities departments, recruited during lectures.
• Unmodeled confounders. Gender, level, prior generative AI experience, device access and English proficiency were not controlled.
• An under-specified social-influence construct. The study did not measure which social pressures students actually experienced, so the null cannot be interpreted as their absence.
• Convenience sampling. Students absent from the surveyed lectures are systematically missing.

Citation

Davor, I., Asare, B., Boateng, F. O., & Boateng, E. A. (2026). Digital learning anxiety and UTAUT predictors of university students' intention to use generative artificial intelligence in Ghana. Frontiers in Education, 11.