Unpacking the Heterogeneity of Pre-service Teachers' ChatGPT Acceptance: A Latent Profile Analysis Across STEM and Non-STEM Disciplines

Synthesis: This person-centered latent profile analysis (LPA) of N = 128 Taiwanese pre-service teachers (68 STEM, 60 non-STEM) identifies four distinct ChatGPT-acceptance profiles — Pragmatic Evaluators (47.66%), Technology Pioneers (26.56%), Resistant Skeptics (14.06%), and Environmental Observers (11.72%) — and shows that disciplinary background strongly shapes profile membership (Cramer's V = 0.532). Critically, Resistant Skeptics report high perceived ease of use but very low behavioral intention, proving technical ease does not guarantee adoption. The study proposes a differentiated AI Literacy training framework for teacher education.

Key Findings

1. Four distinguishable acceptance profiles (entropy = 0.985). Person-centered LPA (using TAM and UTAUT2 constructs) revealed Pragmatic Evaluators (47.66%), Technology Pioneers (26.56%), Resistant Skeptics (14.06%), and Environmental Observers (11.72%).
2. The ease-of-use ≠ intention paradox. Resistant Skeptics had relatively high perceived ease of use yet very low behavioral intention — demonstrating that perceived technical ease does not guarantee adoption, and implying resistance is driven by attitudes/values rather than capability.
3. Discipline shapes acceptance. STEM pre-service teachers concentrated in Technology Pioneers (47.1%), while non-STEM teachers were overrepresented in Environmental Observers and Resistant Skeptics (χ² = 36.20, p < .001, Cramer's V = 0.532).
4. Profiles differ strongly in behavioral intention (F = 109.00, p < .001, η² = 0.726), confirming the profiles are meaningfully distinct in downstream intention.
5. A differentiated AI-literacy framework. Profiles such as Resistant Skeptics require targeted interventions that extend beyond operational skills training — addressing attitudes, trust, and contextual factors.

What this means for practice

• Faculty developers. Segment AI training by acceptance profile instead of running one onboarding track: Resistant Skeptics (14.06%) need attitudinal and risk dialogue, Environmental Observers (11.72%) need contextual and institutional support, and Technology Pioneers (26.56%) need advanced practice beyond the basics.
• Faculty developers. Stop treating operational fluency as adoption: the Resistant Skeptics reported relatively high perceived ease of use but very low behavioral intention, so pair every skills module with work on trust, values, and professional stance.
• Faculty developers. Set different expectations by discipline: STEM pre-service teachers concentrated in Technology Pioneers (47.1%), while non-STEM teachers were overrepresented among Environmental Observers and Resistant Skeptics (Cramer's V = 0.532), so a single cross-faculty workshop will underserve both groups.
• Faculty developers. Diagnose both perceived usefulness and ease of use — the profiles rest on five TAM/UTAUT2 constructs (PU, PEOU, AT, SI, FC) — and read a high ease-of-use score as no evidence of readiness to adopt generative AI.
• Faculty developers. Treat the proposed profile-specific training framework as a design hypothesis to test locally, not a validated curriculum: the typology-based interventions it recommends have not yet been implemented or evaluated.

Limitations

• N = 128 pre-service teachers recruited by convenience sampling from a single region of Taiwan; the authors note the size may destabilize the smallest profile (Resistant Skeptics, 14.06%) and limit more complex statistical models.
• The sample is culturally bounded: all participants were Taiwanese pre-service teachers, and the authors state the four-profile structure and the disciplinary differences may not hold in Western individualist or other non-Western educational systems.
• Data are entirely self-report questionnaire responses, so social desirability is uncontrolled and the psychological motivations behind each profile cannot be triangulated with behavioral evidence such as usage logs or practicum observation.
• The cross-sectional design cannot show profile stability or movement, and the study tested none of its own recommendations — the differentiated training strategies are theory-driven and were not empirically evaluated, and only five of the six TAM/UTAUT2 constructs served as LPA indicators.

Citation

Chen, P.-H., Lee, H.-Y., Huang, Y.-M., & Wu, T.-T. (2026). Unpacking the heterogeneity of pre-service teachers' ChatGPT acceptance: a latent profile analysis across STEM and non-STEM disciplines. International Journal of STEM Education, 13, 33.