Exploring user heterogeneity in autonomous vehicle acceptance: A moderated mediation model based on extended TAM

AVs could reduce accidents caused by human error and make travel more efficient, but public acceptance remains relatively low, with many people citing concerns about safety, control and trust. Traditional acceptance models such as the Technology Acceptance Model (TAM) and UTAUT may lack explanatory power unless they are adapted to the psychological and technological context of AVs. Prior work has suggested adding technological factors and individual differences (personality, demographics), and several studies have tested demographic moderators. The paper argues that a significant gap remains in understanding how these human factors interact with behavioral constructs to shape AV adoption intentions.

The authors propose an extended TAM and test it with survey data.

• Model: behavioral factors (attitude, perceived usefulness, perceived ease of use, subjective norms, trust, perceived enjoyment, perceived privacy risk, perceived safety risk), technological factors (compatibility, relative advantage, reliability, complexity) and two Big Five personality traits (openness, neuroticism), leading to intention to use and actual use.
• Measures: 60 items across 20 latent constructs, adapted from validated scales and rated on a 7-point Likert scale. A pilot test with 201 participants led to the removal of several items. The English questionnaire was translated into Korean and back-translated.
• Sample: online survey conducted between August 8 and 16, 2024, using purposive quota sampling; 509 valid responses from participants aged 20 to 70 (51.1% female, 48.9% male).
• Analysis: partial least squares structural equation modeling (PLS-SEM, a variance-based method for testing networks of relationships among latent constructs) in SmartPLS 4.1, with bootstrapping (5000 subsamples). The final measurement model had 20 constructs and 59 items.
• Moderated mediation (testing whether an indirect effect differs between groups): bootstrapped multi-group analysis (PLS-MGA) on five dichotomized variables: age (<40 vs. ≥40), gender, education (below vs. bachelor's degree or higher), monthly income (below vs. above 5 million KRW) and driving experience (less than vs. three years or more).

• Empirical result: Intention to use strongly predicted actual use (β = 0.714, p < .001). Attitude had the strongest positive influence on intention (β = 0.862, p < .001), followed by perceived usefulness (β = 0.186). The model explained substantial variance in key constructs (e.g., R² = 0.702 for intention to use, 0.784 for actual use, 0.847 for attitude).
• Empirical result: Trust positively influenced attitude (β = 0.121) but showed a negative coefficient toward intention (β = −0.184). The authors interpret this as trust working mainly indirectly, through attitudes, rather than independently driving use.
• Empirical result: Technology perceptions shaped beliefs about AVs. Compatibility strongly influenced ease of use (β = 0.647) and enjoyment (β = 0.480) but not usefulness; relative advantage increased usefulness, trust (β = 0.350) and enjoyment; reliability strongly increased trust (β = 0.500) but not usefulness; complexity lowered ease of use (β = −0.203) but did not significantly reduce trust. Perceived safety risk was a significant negative predictor of trust, whereas privacy risk did not significantly affect trust.
• Empirical result: Personality effects were modest. Openness was positively associated with subjective norms (β = 0.231, p < .001), and neuroticism with higher privacy concerns (β = 0.161, p < .001) and higher perceived safety risk; the other hypothesized personality effects were not supported.
• Empirical result (partly exploratory): Several indirect pathways differed by group. For example, social-norm pathways were stronger for women (SNs → PU → ATT → BI: 0.084 vs. 0.018, p = .003), as were safety-risk-to-trust pathways. Age effects were limited and only marginal (p = .064 and .065). The authors report effects at p < .10 as exploratory, because they would not survive a Bonferroni correction.

The authors state three contributions: theoretically, the study extends TAM with technological perceptions, trust and personality traits, highlighting the indirect role of trust and the limited role of personality; methodologically, it applies moderated mediation analysis to show how demographic characteristics reshape adoption pathways; practically, it offers targeted recommendations for AV deployment. These include safety certification and transparent communication for risk-sensitive groups such as women and less-experienced drivers, privacy regulation and data governance for older and highly educated users, and usability standards to reduce complexity barriers. The stated limitations are the cross-sectional (correlational) design, a single-country sample from South Korea, the dichotomizing of continuous moderators, and a perceived safety risk scale with only two items. The authors also note that the extended model was not formally compared against a baseline TAM. They call for longitudinal or experimental designs, cross-cultural replication with measurement invariance testing, latent moderated SEM or alternative cut-offs, and nested model comparisons.