Amir Rafe, Subasish Das
8 min
This paper by Amir Rafe and Subasish Das challenges the dominant view in autonomous vehicle (AV) acceptance research, which focuses almost entirely on 'pull factors' like perceived usefulness and trust from models such as TAM and UTAUT. Instead, it introduces perceived community driving-safety concern (PCSC) as a critical 'push factor'—public frustration with declining human driving safety that propels people toward AI alternatives. Using a nationally representative U.S. survey from Pew Research Center's American Trends Panel (Wave 152), the authors employ weighted structural equation modeling (WLSMV) to test a sophisticated moderated mediation model. Their findings reveal a nuanced 'risk-spillover mechanism': while safety concerns directly boost domain-specific endorsement of AI driving, they simultaneously dampen broader enthusiasm for AI in daily life, resulting in a near-zero net effect on AV acceptance.
The study is timely amid rising U.S. traffic fatalities (over 36,000 annually) largely due to human errors like distracted and aggressive driving. Public polls show 49% perceive worse community driving safety than five years ago, creating ripe conditions for AI-driven solutions—if acceptance barriers are understood.
PCSC captures the public's belief that driving safety in their local area is deteriorating, measured via survey items on local crash trends and risky behaviors. This acts as a push factor in a push-pull framework: dissatisfaction with the status quo (human driving failures) motivates seeking alternatives (AI vehicles). Empirically, PCSC has a small but positive direct effect on evaluations of AI driving capability over humans, aligning with intuitive logic—'human drivers are getting worse, so AI might be better.' This contrasts with prior AV literature's tech-centric focus, highlighting how real-world safety crises can drive adoption.
The core innovation is identifying a dual-pathway process. PCSC positively influences AI driving evaluations directly (domain-specific push). However, it negatively impacts Generalized AI Orientation (GAAIS)—a general excited-vs-concerned stance toward AI in everyday life—which itself strongly predicts positive AI driving views. This creates inconsistent mediation: the indirect path (PCSC → lower GAAIS → lower AI endorsement) offsets the direct positive effect, yielding a total effect near zero.
This 'risk-spillover' explains tension in public opinion: people may endorse AI for cars due to safety woes but grow wary of AI broadly, fearing over-reliance or other risks. Robustness checks (bootstrap inference, Imai sensitivity, E-value, propensity matching) confirm the pathways' stability.
Personal driving frequency moderates the PCSC → GAAIS path. The negative indirect effect holds across low, mean, and high frequencies, meaning safety concerns suppress general AI optimism regardless of how often one drives. This suggests PCSC's spillover is pervasive, not limited to heavy drivers most exposed to risks.
The paper extends TAM/UTAUT with PPM (Push-Pull-Mooring) dynamics, integrating trust, usefulness, and now external push factors. It calls for models capturing domain-general vs. domain-specific attitudes, as spillover risks could hinder AV rollout despite safety imperatives. Policymakers should address public AI concerns alongside AV benefits to unlock push-driven adoption.
Road traffic crashes claim approximately 1.19 million lives annually worldwide, and human error accounts for the vast majority, yet the autonomous vehicle acceptance literature models adoption almost exclusively through technology-centered pull factors such as perceived usefulness and trust. This study examines a moderated mediation model in which perceived community driving-safety concern (PCSC) predicts evaluations of AI versus human driving capability, mediated by Generalized AI Orientation and moderated by personal driving frequency. Weighted structural equation modeling is applied to a nationally representative U.S. probability sample from Pew Research Center's American Trends Panel Wave 152, using Weighted Least Squares Mean and Variance Adjusted (WLSMV)-estimated confirmatory factor analysis on ordinal indicators, bias-corrected bootstrap inference, and seven robustness checks including Imai sensitivity analysis, E-value confounding thresholds, and propensity score matching. Results reveal a dual-pathway mechanism constituting an inconsistent mediation: PCSC exerts a small positive direct effect on AI driving evaluation, consistent with a domain-specific push interpretation, while simultaneously suppressing Generalized AI Orientation, which is itself a strong positive predictor of AI driving evaluation. Conditional indirect effects are negative and statistically significant at low, mean, and high levels of driving frequency. These findings establish a risk-spillover mechanism whereby community driving-safety concern promotes domain-specific AI endorsement yet suppresses domain-general AI enthusiasm, yielding a near-zero net total effect.
Alex: So PCSC pushes directly toward favoring AI drivers, but there's another route through overall AI views we already discussed?
Sam: Precisely—it's an inconsistent mediation, where the paths pull in opposite directions. The direct path is positive: more PCSC leads straight to better ratings of AI driving capability compared to humans. But PCSC also lowers general positivity toward AI overall, and since general AI views strongly boost those driving ratings, that creates a negative indirect path. The result is they mostly cancel, leaving little net change in support for AI drivers.
Alex: Like water pouring into a bucket for AI driving support, but leaking out through the general AI side?
Sam: That's a solid way to picture it—a leaky bucket effect from risk spillover. Local driving fears push water directly into preferring AI for roads, but the worry spills over, draining general enthusiasm for AI, which pulls some back out. This explains why rising road dangers don't clearly lift self-driving car acceptance; the dual paths offset each other.
Alex: And this holds up across different kinds of people, like how often they drive themselves?
Sam: Yes, they tested that too—driving frequency doesn't change the negative indirect path much; it stays consistent whether someone drives rarely, sometimes, or every day. The study used data from over 5,000 Americans in a nationally representative survey, adjusted for things like age and location to avoid bias.
Alex: Huh, so the push from bad human driving is real, but the spillover keeps overall AI car support flat.
Sam: Exactly, and they confirmed it with thousands of resampling checks—bias-corrected bootstraps—to ensure the indirect effect is reliable, not a fluke. This challenges simple ideas that just more worry equals more tech support; the mechanism shows why it's more nuanced.
Alex: So the paths really do cancel each other out... but how exactly did they measure that balance?
Sam: They set up two linked paths to trace the flow. First, they checked how worry about local driving problems affects overall views on AI—finding it lowers those views steadily. Then, they linked that to ratings of AI versus human drivers, where general AI views strongly lift the ratings, but the direct link from driving worry still gives a small positive nudge. Driving frequency was tested as a factor that might alter the worry-to-general-AI path, but it didn't—the negative pull stayed about the same size across groups.
Alex: Wait, so no matter how much you drive, the spillover drag is consistent? That seems key.
Sam: Yes—the indirect effect through general AI views was negative and reliable across the board, shrinking just slightly for heavy drivers but not enough to matter statistically.
Alex: Huh. And they double-checked this wasn't due to some hidden flaw?
Sam: Absolutely—a full robustness suite confirmed stability. For instance, sensitivity tests showed you'd need a huge hidden correlation between errors to wipe out the indirect effect. They also ran reverse mediation to confirm the direction—driving worry affecting AI views, not vice versa—plus checks across urban, suburban, and rural areas, all holding up.
Alex: Okay, so the leaky bucket holds steady regardless of where you live or how much you drive... that's a clear picture of why the push doesn't move the needle much.
Sam: Precisely. The direct gain from seeing bad human driving is real but offset by the spillover caution toward AI broadly, with strong evidence across checks. This nuances why road safety woes alone don't propel self-driving acceptance.
Alex: Yeah... the cancelation feels even clearer now. So beyond the paths balancing out, what does this mean for how we think about models predicting if people will accept self-driving cars?
Sam: It refines common frameworks that explain why people adopt new tech. These models look at factors like whether something seems easy to use or builds trust, helping predict uptake for things like apps or gadgets. The key update here is separating broad feelings about AI from opinions just on driving tasks. The study shows broad AI views act as a bridge, so future predictions should treat them that way.
Alex: Okay, so splitting general AI caution from driving-specific views sharpens those predictions. But practically, if road worries don't boost support much, how should that guide policies or campaigns for self-driving cars?
Sam: The evidence points to focusing on overall AI attitudes as the stronger lever. Efforts to teach people about AI—what it can do, how it makes decisions transparently, and chances to try it positively—could lift driving support more than just fixing local roads. Since the link from general views to driving ratings is much larger than the direct road-worry nudge, these literacy programs might unlock acceptance even if human driving stays risky.
Alex: Huh, that flips the script—build AI comfort broadly instead of just road fixes. Though to be fair, no study's perfect—what limits should we keep in mind here?
Sam: A few notable ones. The data is from one time point, so it shows links but can't prove one thing causes another; you'd need tracking over time or experiments to confirm direction. The indirect path is small and could shift with overlooked factors like general risk-taking habits. Plus, it's U.S.-only, so driving norms or AI familiarity elsewhere could change the pattern.
Alex: Right, so solid evidence for the tug-of-war, but cross-checks needed beyond snapshots or one country. Still, it explains why bad roads don't simply push people to AI cars—the caution spillover keeps things balanced. Thanks for breaking it down, Sam—this has been a thoughtful look at how road worries shape AI car views.