An Extended Technology Acceptance Model for Investigating Factors Affecting the Adoption of Self-Driving Cars

Self-driving cars are expected to reduce accidents caused by human error and free drivers for other tasks. Fully automated driving has not yet been achieved, so the paper argues it is important to evaluate acceptance early and to identify the factors that can hinder it. Prior studies extended the Technology Acceptance Model (TAM, which explains intention to use through perceived usefulness, perceived ease of use and attitude) with factors such as trust, subjective norms, perceived risks and enjoyment. Others examined technological factors and human factors such as education, income and driving experience. The study aims to bring the frequently used factors together into a new extended TAM for self-driving cars and test their effects. Because behavior-theory factors are not directly observable, it also examines how technological factors affect them.

• Model: hypotheses drawn from prior studies of autonomous vehicle (AV) acceptance, with factors in three groups: (a) behavior-theory factors (latent variables): perceived usefulness (PU), perceived ease of use (PEOU), subjective norms (SNs), trust (TRU), perceived enjoyment (PE), perceived privacy risk (PPR) and perceived safety risk (PSR), alongside attitude (ATT), behavioral intention (BI) and actual use (AU); (b) technological factors (controllable, manifest variables): compatibility (COM), relative advantage (RA), complexity (CPL) and reliability (REL); (c) human factors as moderators: education (EDU), income (INC) and driving experience (DE).
• Data: an online survey of 81 respondents from South Korea with 45 questions, including demographics.
• Analysis: Partial Least Squares Structural Equation Modeling (PLS-SEM, a method for estimating relationships among unobserved constructs measured by survey items) in SmartPLS 3.0. Reliability and validity were checked with Cronbach's alpha (> 0.7), composite reliability (> 0.7), average variance extracted (AVE > 0.5) and the Fornell–Larcker criterion for discriminant validity.

• Empirical (measurement model): all AVEs exceeded 0.5 and all Cronbach's alphas and composite reliabilities exceeded 0.7, indicating good reliability and convergent validity; discriminant validity met the Fornell–Larcker criterion.
• Empirical (hypotheses): 18 of 25 hypotheses were supported; H4, H5, H6, H8, H10, H12 and H13a were rejected at the 0.05 level.
• Empirical (intention and use): BI significantly predicted AU (β = 0.637, p < 0.001; R² = 0.52 for AU). BI was significantly affected by ATT (β = 0.405, p < 0.001), SNs (β = 0.424, p < 0.01) and PE (β = 0.203, p < 0.05), but not by PU (β = 0.015) or TRU (β = 0.016). ATT was affected by PU (β = 0.318), TRU (β = 0.316) and PE (β = 0.512). PEOU affected TRU (β = 0.326) and PU (β = 0.325), and SNs affected PU (β = 0.347).
• Empirical (technological factors): COM affected PEOU (β = 0.508) and PE (β = 0.548) but not PU; CPL negatively affected PEOU (β = −0.311) but not TRU; RA affected PU (β = 0.241), TRU (β = 0.224) and PE (β = 0.210); REL affected TRU (β = 0.282) but not PU.
• Empirical (risk): PSR negatively affected trust (β = −0.152, p < 0.05), while PPR had no significant effect (β = −0.043). The authors suggest this is because participants were mostly young adults who are used to sharing personal data with online services.
• Empirical (human factors): driving experience had a significant moderating effect on BI through SNs (less experienced users depended more on others' opinions); education and income showed no direct or moderating effects, which the authors suggest may reflect the small sample.

The authors conclude that attitude is the main determinant of intention to use AVs, shaped by other behavior-theory factors. They also conclude that users with less driving experience are influenced by people important to them. Users' trust, perceived usefulness, ease of use and enjoyment can be improved by increasing the system's compatibility, relative advantage and reliability. Because young adults mostly are not concerned about sharing data with AVs, they suggest developers focus on safety risks. The authors attribute the absence of education and income effects possibly to the small sample size. Future research will aim to increase the reliability and validity of the measurement model and to propose a newly developed AV acceptance model with more respondents from different user groups.