Abstract
The adoption of autonomous vehicle (AV) policies across various states from 2011 to 2023 has presented a mixture of policy responses, reflecting the diverse approaches to governance in the face of such advanced technological innovation. Certain states demonstrate a rapid adoption of new advanced technologies, while slower policy adoption by others might stem from a lack of expertise, collaboration, and learning methods. Several state governments, recognizing the transformative potential of AVs, have strategically included assistance for private organizations and industry leaders within their policies, and aim to foster innovation and economic growth. This strategic approach stands in stark contrast to states that lag, contributing to a significant gap in the AV policy landscape (McAslan et al. 2021). A possible explanation of this gap is a lack of expertise in advanced technology, concerns raised by labor unions, environmental organizations, potential of job disruption and limits in existing academic research (American Planning Association 2024; Gutierrez Gaviria 2022; Teamsters 2023). The intricate nature of AV technology demands a deeper understanding from policymakers and variation in state policy approaches highlights the complexities and challenges but also accentuates the pivotal role of policy diffusion mechanisms in bridging the gap between leading and lagging states.
Rapid technological advancements, complex technicalities, and multifaceted implications of AVs pose challenges that extend beyond the scope of traditional policy frameworks. State governments, particularly those lagging, may face difficulties in comprehending and addressing the intricacies associated with AV technology. This situation corroborates how insufficient policy learning has also contributed to some state governments having a skeptical attitude toward AV technology (McAslan et al. 2021). This underscores the critical need to understand the mechanisms of policy diffusion, specifically policy learning and competition, and their role in overcoming these challenges and fostering a conducive environment for AV policy adoption across states. While research on AV policy has primarily focused on examining the substantive content and outcomes of these policies (Parinandi et al. 2024), there is a need to explore the underlying dynamics of policy diffusion (e.g., expertise, collaboration, and learning methods) that shape their adoption across state governments (Li and Kelchen 2021). In the context of AV policy, policy learning often involves states observing and adapting successful strategies from pioneers, while diffusion mechanisms may include adopting policies in response to neighboring states’ initiatives or to align with national guidelines. Understanding these processes is key to analyzing the varied landscape of AV policy adoption across states (Moyson, Scholten, and Weible 2017).
This paper underscores the critical role of intergovernmental organizations and private sector partnerships in facilitating AV policy innovation and diffusion (Klijn and Koppenjan 2012; Koliba, Meek, and Zia 2019). By leveraging these collaborations, policymakers can harness a wealth of knowledge and resources, fostering a conducive environment for the effective governance of emerging technologies (Nicholson-Crotty et al. 2014). The findings herein provide a foundation for understanding how such synergies can be optimized to enhance policy learning, underscored by practical examples from states that have pioneered AV legislation.
By focusing on the complex mechanisms of policy learning and competition, this research aims to disentangle the factors driving the diffusion of AV policies, providing insights into the complex interplay between innovation, regulation, and interstate influences. This study is not merely an addition to the diffusion literature. It provides a unique perspective by researching into the distinctive nature of AV policy diffusion. By integrating insights from the policy diffusion literature, our research seeks to investigate the existing AV policy landscape across state governments in the United States. Importantly, it builds on the diffusion literature by untangling the specific mechanisms driving the early spread of AV policy. Diffusion researchers have become increasingly focused on defining and measuring the causal mechanisms that drive policy adoption (Shipan and Volden 2008). But these mechanisms can vary across the diffusion life course (Mallinson 2021b). Our conceptual contribution lies in shedding light on the extent to which early diffusion of state AV policy is characterized by policy learning. This is important given the stakes of getting artificial intelligence (AI) regulations right, as an emergent technology with significant social, economic, and political implications, and the pressure on states to cultivate new industries.
Regardless of AV’s benefits, it is important to underline that the expansion of AVs on the roads without an effective public policy cannot provide the necessary protection to citizens (Nathan and Hyams 2022). There is a growing interest in AV technology opportunities in many states (Chen, Gascó-Hernandez, and Esteve 2023; Mallinson et al. 2024), driven not merely by the potential for economic growth but by a pressing demand for innovative transportation solutions that address the unique needs of each state’s market (Saylor, 2021). For example, there is evidence that AV promises to improve the U.S. transportation industry—local city delivery, trucking, public transportation, and farming—by providing an opportunity to improve current and future labor conditions (Gutierrez Gaviria 2022; Noh 2022). Moreover, some states are exploring the enhancement of their transportation networks by integrating the shared mobility concepts of services like Uber and Lyft (Saylor, 2021). As states pilot and refine policies to both support and regulate AVs, there exists a valuable opportunity for inter-state learning, which can contribute to the betterment of policy development and execution. In addition, policy diffusion and learning mechanisms can occur at multiple levels of government and across different domains (Marsden et al. 2011). Meaning that states will also be learning from local governments, they will be influenced by federal policy, and state policy experiments can influence changes in federal policy (Mallinson et al. 2024).
Drawing not only on policy diffusion theory, but also research on collaborative and network governance, we consider whether and how policy learning and competition are occurring between the states and across different levels of government within the American federal system. Further, we move away from the common centering of state governments in policy innovation and adoption decision-making, to consider how intergovernmental organizations and private sector actors are facilitating policy learning between states. Specifically, we examine the dynamic landscape of AV policy adoption, investigating the underlying mechanisms that drive diffusion across state governments. The study seeks to answer the following questions:
Why have some states adopted AV policies when others have not?
Why are some states leading in this policy while others are lagging?
What diffusion mechanism explains the spread of AV policy in the United States?
The study not only contributes to the understanding of AV policy diffusion but also seeks to disentangle the specific mechanisms—policy learning and competition—shaping the early adoption of AV policies. These insights can enhance public organization effectiveness by encouraging governance methods that accelerate policy adoption, improve AV public services, foster technological innovation, and minimize associated risks (de Almeida, dos Santos, and Farias 2021; Wirtz, Weyerer, and Kehl 2022). By integrating insights from policy diffusion literature and network governance theory, we aim to provide a nuanced perspective on how states navigate the complex landscape of AV regulation. Identifying whether learning is a driving force behind AV policy adoption is of paramount importance in understanding the dynamics of policy diffusion in this emerging field. The nuanced insights into policy learning mechanisms offer policymakers and industry stakeholders a roadmap for navigating the complexities of AV regulation. Exploring these mechanisms enhances our ability to anticipate and shape future policies, ensuring a thoughtful and informed approach to the evolving landscape of autonomous vehicles.
Our study employs a mixed methods design, strategically combining quantitative dyadic event history analysis and qualitative interviews. Such a pairing has proved fruitful for disentangling diffusion mechanisms (Starke 2013). The quantitative component provides a comprehensive view of state AV policy adoptions, while qualitative insights deepen our understanding of the intricate learning processes among state governments. This blend of methods allows us to specifically focus on policy diffusion and learning mechanisms, offering a unique contribution to the scholarly discourse on emerging technology governance. The article unfolds in the following sections: a literature review to contextualize the study, an exploration of the AV industry landscape, an in-depth discussion of policy diffusion mechanisms, a detailed overview of our mixed methods approach, and a conclusion that synthesizes our findings and outlines their implications for future research and policymaking.
The Intersection of Policy Diffusion and Networked Governance
Given the rapid spread of autonomous vehicle policies, understanding the underlying mechanisms is crucial. Diffusion theory provides a foundational lens to explore these dynamics, examining how policies propagate across states through learning, emulation, and response to incentives (Graham, Shipan, and Volden 2013; Maggetti and Gilardi 2016; Marsden et al. 2011;). Diffusion scholars have both explicitly (Boushey 2010; Desmarais, Harden, and Boehmke 2015; Garrett and Jansa 2015) and implicitly (Freeman 1985; Walker 1969; Zimmerman 1998) considered the role of various interstate networks in the spread of policy innovations. The theory speculates that ideological similarities between states can significantly influence their policy decisions, including those related to AV technology.
However, a significant challenge faced by many diffusion studies is the quest for a credible conclusion regarding the specific mechanism driving policy adoption. This challenge becomes particularly pronounced in the nuanced landscape of AV policy adoption. It has been long recognized that the macro-level indicators of diffusion mechanisms relied upon in many studies (e.g., neighbor adoptions) do not adequately disentangle candidate mechanisms (Berry and Baybeck 2005). Additionally, the neighbor adoption measure can falsely signal diffusion when adoptions by neighbors are independent (Volden, Ting, and Carpenter 2008). Compounding this challenge is the presence of multiple mechanisms at play. Some states may be learning from each other, while others might be responding to economic or reputational pressures during the adoption of a given innovation. To address this complexity, we employ a mixed-methods approach to triangulate the specific diffusion mechanism(s) involved in the spread of AV policy (Leiser 2017). We turn now to considering why AV policy is important and why understanding early policy learning is important. We then consider how a mixed-methods diffusion design offers both a macro- and micro-level view of the complex networks driving the policy’s diffusion.
Network Governance
In the context of AV policy, network governance becomes particularly salient. State coalitions, intergovernmental organizations, and public-private partnerships form the backbone of a collaborative effort, steering the direction and pace of policy adoption and implementation. Complementing diffusion theory, network governance theory is a practical approach to governing the collective action of the private and public sectors (Klijn and Koppenjan 2012; Koliba, Meek, and Zia 2019), encompassing both formal and informal institutions that allocate resources and coordinate joint action within a network of organizations. The literature defines governance as “formal and informal institutions to allocate resources and coordinate joint action in a network of organizations” (Kapucu and Hu 2020, 18). This theoretical framework becomes particularly relevant as we consider the nuanced landscape of AV policy adoption.
As state governments navigate the challenges and opportunities presented by AV technology, the involvement of various actors in governance networks becomes instrumental. Network governance complements this by highlighting the role of intergovernmental organizations, state coalitions, and public-private collaborations in creating conduits for information exchange, best practices, and innovation (Klijn and Koppenjan 2012; Koliba, Meek, and Zia 2019). For instance, the National Conference of State Legislatures serves as a pivotal platform where policymakers from different states gather to share insights and strategies on AV regulation, facilitating a networked approach to governance (NCSL 2022).
Moreover, as states develop and implement AV policies, they not only grapple with the technical and regulatory complexities of this emergent technology but also actively signal their commitment to fostering an environment conducive to AV innovation and deployment. This approach acknowledges that clear regulatory frameworks and infrastructure readiness are key to attracting AV manufacturers and investors, thereby aligning state policy initiatives with market demands. In the context of AV policy diffusion, we further explore the potential influences of intergovernmental organizations and private sector collaborations within these networks, aiming to contribute to a more comprehensive understanding of the factors shaping AV policy innovation and diffusion across states.
Integration of Policy Diffusion and Networked Governance
The intricate landscape of AV policy adoption, marked by technological complexity and regulatory challenges, calls for an integrated approach. By weaving together diffusion theory’s insights on policy spread with network governance’s emphasis on collaborative structures, we can gain a more comprehensive understanding of how and why policies gain traction across states. This intersection allows for a deeper understanding of how states, while navigating the challenges and opportunities presented by AV technology, engage in governance networks that involve a variety of actors. These networks are instrumental in fostering an environment conducive to AV innovation and deployment, highlighting that clear regulatory frameworks and infrastructure readiness are key to attracting AV manufacturers and investors and aligning state policy initiatives with market demands.
As defined by Walker (1969), policy innovation occurs when a state government adopts a new policy or program for the first time. For example, public organizations may change traditional bureaucratic approaches by recognizing and collaborating with other policy actors (Kim 2006; Kim et al. 2013). Innovation can result from internal processes within an organization or from external sources such as research or consultation. State governments are increasing in complexity, including in diversity in their socio-political environments and economic factors that affect how they operate and make decisions. As state governments grow in complexity, embracing diverse socio-political environments and economic factors, the involvement of citizens and private actors becomes imperative in the policymaking process, particularly in the realm of emergent technologies like AI. This recognition is vital for understanding the dynamics of AV policy adoption (Perry and Uuk 2019). Thus, it is a complex network of actors that is likely engaged in innovation and diffusion surrounding AI-related policies.
Diffusion scholars have increasingly turned their attention to disentangling underlying causal mechanisms. Policy learning is one mechanism that drives policy innovation and diffusion as public organizations are exposed to new ideas and approaches (Graham, Shipan, and Volden 2013; Maggetti and Gilardi 2016; Marsden et al. 2011). Additionally, governance networks, can promote knowledge exchange and collaboration, enhance the capacity for innovation and policy implementation, and facilitate policy learning and diffusion (Hong et al. 2022). Effective policy learning requires an understanding of the broader policy landscape and the networks and actors involved in shaping it. Public organizations, by leveraging these mechanisms, can craft more effective policies and practices, ultimately leading to improved outcomes for citizens (Wang et al. 2020). Scholars who study policy diffusion and policy learning have highlighted the significance of learning as a means for policymakers to recognize and overcome implementation obstacles, enhance policy effectiveness, and ultimately achieve more favorable results (Wischmeyer and Rademacher 2019). Policymakers can avoid repeating the same mistakes and build on successful approaches by learning from the experiences of other regions and adapting policies to local contexts (Maggetti and Gilardi 2016). This analysis underscores the symbiotic potential for leveraging both policy diffusion and networked governance to understand policy diffusion. We find in the case of AV policy diffusion, that the interplay of these theories not only enriches our understanding of policy dynamics but also offers practical pathways for fostering innovation and cooperation in the face of emerging technological landscapes.
The Rise of Autonomous Vehicles and State Responses
The emergence of autonomous vehicles (AVs) has led public organizations into a new era of transportation by revolutionizing mobility and challenging existing policy frameworks. With each passing year, an increasing number of states in the United States have implemented their own AV policies, yielding a diverse landscape of regulatory approaches. Several state governments are investing resources in AV (e.g., California, Nevada, New Mexico, Pennsylvania, Texas, and others), which has become a focal interest in their policy discussions and innovation agendas (NCSL 2022).
AVs offer significant economic benefits, such as reducing transportation costs, enhancing supply chain efficiency, and decreasing road congestion, which collectively promote regional economic growth (AVIA 2024). By improving road safety, AVs can dramatically lower public health expenditures associated with traffic accidents. This technological shift is likely to boost job creation in the tech and automotive sectors and stimulate investment in related industries like artificial intelligence and sensor technology. Furthermore, efficient goods transportation via AVs can reduce logistics costs, improve delivery speeds, and increase business profitability. States that adopt favorable AV regulations may attract high-tech firms and startups, boosting local economies and enhancing their competitive standing on a national scale (AVIA 2024).
The AV industry is predicted to grow to $300–400 billion globally by 2035, with a projection of 3.5 million self-driving cars on American roads by the same year (Deichman et al. 2023; Insurance Information Institute 2022). Within this transformative potential, the market landscape is intricate, shaped by factors such as individual ownership of AVs, ride-sharing initiatives, mass transit experiments, and advancements in goods delivery through AV technology (Harb et al. 2021). One facet is individuals owning their AVs (Contreras 2020). The second hinges on ride sharing. For example, ZipCar and Uber/Lyft are testing AV services in California, Nevada, and Texas by delivering on-demand mobility services through Shared Autonomous Vehicles (SAVs) (Contreras 2020; Harb et al. 2021). The third involves mass transit, like that being tested in North Carolina (Davidson 2023). Finally, the fourth involves trucking and the delivery of goods (Van Fossen et al. 2023).
As some states actively foster this emergent technology, through steps like incorporating simulators and collaborations with private organizations, a parallel challenge arises in the formulation of effective AV policies. Texas is at the forefront in utilizing simulators to compare several types of AV methods (e.g., ridesharing, trucking, and delivery services) against existing transportation systems (Huang, Kockelman, and Quarles 2020). The test and simulation results unveil significant opportunities to revolutionize roadways and incorporate AV technology. Considering this success, numerous private organizations are collaborating with state governments to invest in AV infrastructure and roadway engineering solutions that can facilitate the safe operation of AVs in multiple U.S. cities and select metropolitan areas. Companies like Waymo One are actively testing AVs in places like San Francisco and the Phoenix, Arizona metropolitan area, with promising outcomes (Acheampong et al. 2021).
The Autonomous Vehicle Industry Association’s (AVIA) 2024 report underscores emerging competitive dynamics within the AV sector. States like Texas, Arizona, and Florida are noted for cultivating favorable regulatory landscapes, instigating a competitive milieu among states eager to attract AV enterprises and secure economic advantages. AVIA (2024) also delineates the substantial economic incentives and job creation potential inherent in AV technologies, positioning them as pivotal factors for states aiming to establish themselves as leaders in this innovative field. Additionally, the report reveals that geographic expansion and market growth forecasts suggest a robust competitive arena where companies and states are vying for dominance and market share. Technological advancements and strategic partnerships, as illustrated by companies like Motional, Zoox, and Nuro, further highlight the intense competition among firms to spearhead developments and secure a commanding position in specific AV market segments (AVIA 2024). This competitive landscape is poised to significantly influence policy decisions, investment strategies, and the trajectory of technological innovation within the AV industry (Mallinson et al. 2024).
Concurrently, state governments face many challenges in considering how to approach AV regulation (de Sousa et al. 2019; Mallinson et al. 2024; Parinandi et al. 2024). Principally, policymakers have limited knowledge of how AVs might behave on the roads. AVs utilize on-board technology that interacts with the surrounding environment “consistent with traditional responses from humans, given the human capacity for contemplation, judgment, and intention” (West and Allen 2020). They operate either completely autonomously (e.g., L4 and L5 classification) or under supervision (e.g., L3 classification) (Crayton and Meier 2017), so governments must decide whether fully autonomous modes are allowed. AV employs a “neural network” to support various sensors (e.g., cameras, radar, laser scanners, GPS) that can gather insight via “deep learning,” detect patterns, apply knowledge, and function independently with no human supervision (Bagloee et al. 2016; Ingrams, Kaufmann, and Jacobs 2021). Significant public investments will be required to improve signage, lanes markings, and more, to maximize the potential of AVs (Wang, McKeever, and Chan 2022).
In summary, AVs require significant policy innovation by national and subnational governments because they can radically transform transportation and mobility. Unlike previous technological innovations, AVs are not simply incremental improvements on existing technology, but rather represent a fundamental disruption in how people will move around (de Sousa et al. 2019). Some of the key public policy issues associated with AVs include safety, liability, privacy, and infrastructure (Guerra 2016). While the federal government provides substantial transportation funding and can help set national standards, the future of AV policy, and AI policy more broadly, will likely be up to the states (Mallinson et al. 2024). States are the ones experimenting with both industry supports and regulations.
It is evident that policymakers do not have answers for each potential AI-related situation and cannot prescribe regulations in all instances (Wischmeyer and Rademacher 2019). This both complicates the job of regulating emergent AI technology and elevates the prospect of state governments serving as laboratories for policy experimentation (Maggetti and Gilardi 2016). There are instances where there is no single answer, mainly when algorithms make wrong decisions, and when AVs cause road accidents and even death to pedestrians (Wirtz, Weyerer, and Sturm 2020). Moreover, AV policy, and and increasinly state technology policy more broadly, moves decisionmaking and policy innovation from being centered on state governments alone to considering a more decentralized and networked decision-making structure that includes intergovernmental oganizations and private firms (Vos and Schroth 2018). The development of a more polycentric governance structure is more likely when national policy is “blocked” (Gillard et al. 2017). While national AI policy is emerging in the United States, polycentric governance is already forming.
Thus, it is important to study the diffusion of AI-related policies among state governments. The case also allows us to consdier the specific mechanisms that are driving early innovation and adoption (Mallinson 2021b). State governments have significant autonomy in policymaking, and they often serve as laboratories for policy experimentation. Studying AV policy helps to understand the policy diffusion mechanisms in several ways. First, it helps to identify the different channels of diffusion, such as direct imitation or learning, social networks, and external pressures from interest groups (Graham, Shipan, and Volden 2013; Shipan and Volden 2008). Notably, influential bodies like International Brotherhood of Teamsters’ stance on AV policy advocating for regulations that balance the advancement of AV technology and the protection of workers and public safety. Their constructive engagement with policymakers underscores how interest groups serve as conduits for policy diffusion, exerting influence and fostering engagement with those crafting policy (Eastman 2023; Freemark et al. 2022; Teamsters 2023).
Second, it allows for the exploration of how policy learning and experimentation can occur in different regions and states, leading to the spread of similar policies (Taylor et al. 2012). Saylor (2021) emphasizes the significant challenges that AV deployment poses to public transportation systems, particularly concerning the rights and mobility of individuals with disabilities as protected under Title VI. The argument presented details how existing legal frameworks, particularly the Americans with Disabilities Act (ADA), may not sufficiently protect these rights in an era of AVs. In response to this concern by researchers and others, the Autonomous Vehicle Industry Association has openly highlighted opportunities for policymakers to collaborate with various interest groups, such as the National Federation of the Blind, aiming to guarantee equal opportunities for individuals with visual impairments (AVIA 2024).
Third, it helps to uncover the role of key non-state actors and institutions in the decision-making process, such as policy entrepreneurs, environmental organizations, intergovernmental organizations, and the private sector (Boushey 2010; Freemark et al. 2022; Karch 2007). For instance, environmental organizations like the Urban and American Planning Association (UAPA), which have a longstanding commitment to the environment sustainability, do not outright oppose AVs. Instead, they champion regulations that promote fairness, safety, and well-being, with a particular focus on supporting underserved communities. The stance of the UAPA highlights the potential of AV technology to benefit the environment through eco-friendly innovations (American Planning Association 2024; Brevini 2021; Freemark et al. 2022).
Finally, it provides insights into the factors that influence the speed and extent of policy diffusion, including political, economic, and social factors (Makse and Volden 2011; Mooney 2021; Nicholson-Crotty 2009). Importantly, there has been growing recognition of the importance of policy chracteristics like complexity and salience on the extent and speed of policy diffusion (Bromley-Trujillo and Karch 2021; Makse and Volden 2011; Mallinson 2016; Menon and Mallinson 2022; Nicholson-Crotty 2009). AV policy is certainly complex, which should slow its diffusion, but AI technology more broadly has increased substantially in salience, thus speeding the spread of related policies. Given its highly technical nature, the increase in public and media salience is helping to overcome technical complexity in the speed at which AV policy is begin pushed onto state agendas (Bromley-Trujillo and Karch 2021).
The selection of AV policy diffusion for this study was deliberate, influenced by the unique characteristics of AV technology and the urgent need for innovative transportation solutions across states. The rapid advancements in AVs present a complex regulatory landscape, requiring states to adapt and innovate policies to meet specific market needs (Taeihagh 2023). Initial AV policies in the U.S., guided by federal regulations, aimed to align with national transportation standards (Hemphill 2020). However, the scalability of these policies across states encounters obstacles due to regulatory fragmentation and administrative complexities (Milakis, Thomopoulos, and Van Wee 2020). For example, differences in hardware standards among AVs pose compliance issues across jurisdictions (Mallinson et al. 2024). The innovative legislation in California and Texas, which established frameworks for AV testing on public roads, exemplifies legislative responses to the growing demand for scalable and efficient transportation policies (Contreras 2020; Harb et al. 2021). This legislation, serving as a model for Nevada and Arizona, illustrates the inter-state learning driven by shared market demands and the transformative impact of AV technology (NCSL 2022). Such policy adoption highlights the dynamic interplay between market demands and the proactive navigation of emerging technologies by states (Hemphill 2020).
Given these challenges, the scalability of AV policy components necessitates a nuanced understanding of legal, bureaucratic, and political capacities at various governmental levels (Mallinson et al. 2024). Our focus on AV policy aims to shed light on the mechanisms of inter-state learning and competition, revealing how states are influenced by the demands of their citizens and market forces for innovative transportation solutions (Moyson, Scholten, and Weible 2017). This demand-driven perspective is essential for understanding AV policy diffusion, suggesting that adoption is significantly influenced by market demands, which vary from state to state. Interstate competition emerges as a potential driver of AV policy adoption, with 18 states already adopting policies facilitating AV testing (Mallinson et al. 2024; Parinandi et al. 2024). These policies aim to create a conducive environment for AV companies, highlighting that policy learning, rather than competitive undercutting, significantly drives the adoption of AV policies in the face of complex technological and policy challenges. We suggest that, given the complex nature of AV technology and the myriad challenges faced by policymakers, the early adoption of AV policy is driven considerably by policy learning rather than a competitive race.
Methods
Policy diffusion studies have traditionally struggled to disentangle the specific mechanisms driving policy adoption (Maggetti and Gilardi 2016). Doing so requires significant consideration of the linkages between macro-level indicators and micro-level concepts (LaCombe and Boehmke 2020). We use a mixed-methods approach to leverage each’s distinct utility in identifying macro- and micro-level phenomena (Creswell and Creswell 2017). The combination of quantitative and qualitative research methods allows them to complement each other’s limitations and thus strengthen the study’s conclusions (Archibald 2016). Meaning, testing our hypothesis relies on the results of both, not just one analysis. The macro-level quantitative model offers initial insight into whether learning and competition are good candidate mechanisms, whereas the qualitative interviews help us to disentangle them. We begin by describing the macro-level quantitative analysis and then continue to the micro-level qualitative approach.
Quantitative Data Collection and Statistical Approach
Using a publicly available dataset from NCSL (Hubbard 2018), we collected state, year, and adoption data on AV policy components adopted by the states between 2010 and 2023. The 10 components include commercial, definitions, licensing and registration, insurance and liability, operation on public roads, operator requirements, other, vehicle testing, infrastructure and connected vehicles, privacy of collected vehicle data, and request for study. To analyze the policy adoption process, we chose to model it using a multiple-components dyadic event history analysis (DEHA) (Boehmke 2009a, 2009b). The DEHA approach is particularly beneficial for our study as it recognizes that innovations in policy components are not monolithic (Boehmke 2009a). Further, it recognizes that innovation on components can occur across multiple years, which cannot be captured by a standard time to first adoption monadic EHA model. The dyadic approach allows us to consider what factors are driving the convergence among states in this policy (Boehmke 2009b; Volden 2006). It also allows us to consider how learning and competition may be occurring across the states (Pollert and Mooney 2022).
This analysis involves a directed dyadic design, where states adopting a given component become signal “senders,” and, in an observed year, states that have not adopted are signal “receivers.” As states adopt a given component, they become signal “senders,” whereas in an observed year any states that have not adopted are signal “receivers.” Each observation is thus a component-state-year dyad. We employed a multilevel logistic regression model, and likelihood ratio testing indicated that the addition of random effects for both state and policy topic improved the model fit (Field, Miles, and Field 2012). The lme4 package in R was used to estimate the multilevel model (Bates et al. 2015). 1
Independent variables were selected from the Correlates of State Policy dataset (Grossmann, Jordan, and McCrain 2021). Because of the dyadic set up, some variables are only measured for the “receiving” states (
Descriptive Statistics for Dependent and Independent Variables.
Note:
Qualitative Interviews and Analysis
While the quantitative analysis can point to learning or competition as candidate mechanisms, the qualitative analysis will help to disentangle the two. The qualitative portion involved interviews conducted with 13 total Chief Information Officers (CIOs), Department of Transportation (DOT) officials, and Chief Security Officers (CSO) from 12 states in spring 2023. 2 The participant selection process employed a stratified method, ensuring representation from each of the four Census regions. Table 2 shows the 12 state offices that participated in interviews. The strategic selection aimed to provide insights from diverse regions and different state perspectives, acknowledging that states actively engaged in AV adoption initiatives were prioritized within the stratified approach. This approach aimed to encompass states at varying stages of AV policy development, recognizing that not all states have advanced initiatives in this domain. Interviews were conducted in a relaxed and non-stressful manner, lasting between 30 and 60 minutes. Zoom was used to capture interview transcripts during the interviews.
Interview Collection by US Census Regions.
We specifically explored how state governments gather insights from the experiences of others, including in the context of public engagement when developing and implementing AI/AV policies. We asked questions such as, “How does your state government engage with the public when developing and implementing AI/AV policy?” and “What is your strategy for learning about best AI policy practices? Do you look to certain groups or other states for lessons? If so, who?” The interviews concluded with additional open-ended questions (Table 5).
Invitations were extended to 100 potential interviewees across the four regions to ensure geographic diversity and representation. Of those invited, 13 officials responded, resulting in a response rate of approximately 13%. Despite this modest response rate, the representation from diverse states ensured a broad spectrum of insights into the regulatory environments and AV policy challenges. To strengthen the validity of our findings and address the limitations of the response rate, our analysis was enriched by incorporating state reports and information from reputable organizations such as NCSL (2022) and AVIA (2024). These sources provided additional context and supported the themes that emerged from the interviews. Future research could enhance engagement strategies or broaden the participant pool to include a wider array of stakeholders, potentially increasing response rates and providing a richer analysis of the diffusion of AV policies.
Thematic analysis was used to develop common themes across many interview topics. NVivo was used to recognize patterns within the subjects and queries discussed with each interviewee. Consequently, seven thematic codes surfaced from the examination, covering topics such as learning and knowledge dissemination, interagency collaboration, engagement with the private sector, evaluation of governance and policies, cooperation between federal and state entities, cooperation across state lines, and involvement of the public and local community. These dialogues unveiled multiple themes, which are elaborated upon in Table 3, offering thematic coding and explanations. Furthermore, the questions used for the interviews are provided in the Appendix under Table 5. We will focus more on the concepts of learning and competition in the Results section below.
Qualitative Codes and Description.
Dyadic Analysis Results
Descriptive Analysis
We first conducted a descriptive assessment of AV policy adoption across the states. Figure 1 displays the total number of components adopted by each state as of 2023. In terms of possible regional spread of policies, the northeastern and western regions are leading the adoption of AV policies, with California in the West and Massachusetts and New York in the Northeast being some of the earliest adopters. Florida and Michigan are also clear leaders regionally. The South and Midwest regions are lagging in terms of AV policy adoption, with Missouri, Wyoming, Montana, and Idaho having not adopted AV policies as of 2023.
Enacted policies on autonomous vehicles across the U.S. until 2023.
The regions have different policy priorities for AVs and are also advancing policy at different paces. Table 4 shows a comparative analysis of AV policies components adopted by states from 2011 until 2023. The observed AV component adoptions by state-year. Those states in the south regions most of them focused on operator requirements, with many states adopting policies in this area. In contrast, the West is most focused on insurance and liability policies. The Midwest is more focused on licensing and registration and vehicle testing policies, while the Northeast has a relatively even spread of policies across the different topics.
Comparative Analysis of Autonomous Vehicle Policies in U.S. States From 2011 to 2023.
Starting in the early 2010s, the West and South displayed interest in enacting policies related to definitions, testing, insurance, and liability, as well as operation on public roads and commercial activities. Historically and due to environmental circumstances, these regions have demonstrated a strong interest in adopting cutting-edge technology to improve environmental conditions, society, and promote economic growth. In contrast, states in the Northeast did not begin adopting AV policy components until 2016. This suggests that these regions were observing the national trend and the development of AV technology, rather than making significant investments in a technology that was still in its early stages of development. The Midwest region has seen a relatively small number of policies adopted in recent years, with many policies adopted in 2018 and 2019.
Explanatory Analysis
Table 5 presents the results of the multi-level DEHA logistic regression model. It includes odds ratios and significance levels for all included independent variables in relation to state policy AV policy component adoption. Odds ratios greater than 1 represent a factor that is increasing the odds of adopting any given AV policy component. Odds ratios less than 1 represent a factor that reduces the odds of adoption. Further, for variables that capture the relationship between the dyad pair (
Results of the Dyadic Event History Analysis of Autonomous Vehicle Policy Component Adoption.
Note: 95% confidence intervals reported in brackets. Variance and (standard deviation) are reported for the random intercepts.
Notably for AV policy, neighbor adoptions are the only statistically significant external correlate. Meaning, when a neighbor adopts a component, it is related to an increased odds of other neighbors adopting that component. Ideological distance has no correlation, suggesting that the diffusion of AV policies might transcend an ideological learning pathway. This finding challenges the assumption that states with similar ideological leanings are more inclined to adopt similar policies (Mallinson 2021c), indicating that factors other than ideological proximity, such as technical expertise or economic incentives, may play a more significant role in the diffusion of AV policies. However, the finding of a regional diffusion effect does not distinguish to what extent this is driven by learning or competition. More complex interstate linkages will be explored in the qualitative interviews.
In terms of internal variables, economic size (i.e., income per capita), state liberalism, and the presence of electronics interest groups are each associated with an increase in the odds of AV component adoption. The latter is important because it reflects non-state actors that are engaging in policy decision-making. Further, greater economic capacity is associated with a higher odds of adoption, potentially out of competitive pressures, but this may also be due to the engagement of private sector collaboration. This is again a subject that will be considered in the qualitative interviews. Finally, economic policy uncertainty is associated with a
The descriptive results show that there are regional leaders and laggards in state AV policy. The explanatory results then find that neighboring states do in fact influence AV policy convergence and raise the possibility of private actor engagement in policy decision-making. The question remains, however, how learning is specifically occurring. We turn now to the interview results for more clarity.
Qualitative Results
Two key themes emerged when discussing interstate interaction regarding AV policy. The first was the role of collaboration within a state’s immediate neighborhood and internally across agencies. The second was the role of intergovernmental organizations and interest groups in linking states. We address each in turn.
Neighborhood and Interagency Collaboration
States learn from and collaborate with each other through organizations like the NCSL. This is especially the case for AV policy. NCSL serves as a hub for states to share information, research, and best practices, particularly in areas where regional or interstate challenges exist. Notably, participants also cited the National Association of State Chief Information Officers (NASCIO) as a key organization for information sharing. By engaging in collaborative efforts, states leverage their collective resources, expertise, and knowledge to tackle complex issues that transcend individual state boundaries (Nicholson-Crotty et al. 2014). The significance of interstate cooperation becomes evident as states pool their political and administrative capacities through organizations like NASCIO. This collaboration approach allows for the exchange of ideas, best practices, and lessons learned, thus fostering an environment of learning and innovation.
Interstate cooperation also enables states to pool their political and administrative capacities, which enhances their ability to develop and implement innovative policies. By working together, states share the costs and risks associated with experimentation and policy implementation, making it more feasible to pursue innovative approaches that might be challenging for individual states to undertake alone. According to an interviewee from Texas:
We have several nonprofit associations that genuinely promote and facilitate collaboration within the public sector. Oftentimes, you hear about conferences and associations, and you may wonder about their real value. During these events, you can learn valuable insights. The mindset we have revolves around more than just attending conferences and listening to presentations. It’s about actively sharing, collaborating as teams, and enhancing resources for our community. This approach significantly impacts our goals.
Building on this perspective, a second interviewee from Texas highlighted the nuanced dynamics surrounding collaboration and competition in AV policy deployment:
So, I’ve seen like there’s this competition between [states]. [. . .] I don’t see it as a competition. [. . .] I know there are states that that are feel like they’ve got to be in competition to deploy autonomy or to deploy connectedness or to deploy drones. [. . .] I know there are states that feel there’s a competition.
These insights from Texas underscore the intricate interplay between collaboration, competition, and the pursuit of innovative approaches in shaping AV policy at the state level.
In another example, the California DOT research team collaborates through the National Cooperative Highway Research Program (NCHRP) where agencies seek information, demonstrations of new technologies, and a test site to observe the opportunities and challenges that may arise from implementing new technology on roadways in other states. Their primary focus is to provide support to state agencies by offering valuable insights and real-world testing environments. Additionally, NCHRP actively engages in collecting feedback and data to refine and improve the effectiveness of new technologies, ensuring safe and efficient transportation systems for all. Further illustrating the role of interstate learning in policymaking an interviewee from California stated:
On the national external side of California, we look at what other states are doing, their best practices, the current state of practices, and the direction in which trends are heading. We gather all this information from outside California through various forums. After that, we shift our focus inwards. On the inward side, we have subject matter experts, such as the 13 program steering committees I mentioned earlier. These experts are well-versed in their respective programs, whether it’s related to design, surveys, traffic, or corporations. So, we have these subject matter experts within our organization. Now, as we continuously engage both internally and externally, our goal is not just to seek solutions but also to identify problems. After all, if we don’t know what problems exist, we won’t know what needs to be solved. This continuous engagement includes focused meetings that help us stay informed and proactive in addressing challenges.
States also frequently rely on collaboration and cooperation among different state agencies to achieve common goals, streamline processes, and deliver effective services to residents (Agrifoglio, Metallo, and di Nauta 2021; Gilardi and Wasserfallen 2019). The interagency process allows for the development of information sharing networks within states to disseminate successful policies or practices (Weber and Khademian 2008). This information exchange enables agencies to learn from each other’s experiences, assess the suitability of adopting similar policies, and learn how to adopt complex technologies in public organizations (i.e., AI and AVs) (Neumann, Guirguis, and Steiner 2024).
In Montana, an interviewee underscored the importance of interagency collaboration, emphasizing the interdependence among agencies for sharing information, resources, and expertise to achieve common goals. Notably, the state’s Chief Information Security Officer (CISO) facilitates collaboration among various agencies, including health and human services, child welfare, and early childhood development, with the goal of streamlining processes and providing holistic services to residents.
Private-Sector Governance, Collaboration, and Learning
Collaboration with and learning from the private sector was another important theme that emerged from the interviews. States establish governance structures, including boards, to evaluate policy effectiveness and monitor technology implementation (Hysing 2021). These structures enable collaboration with private entities and individuals to address social issues and uphold public values, especially in areas like AV and AI policies (Di Giulio and Vecchi 2023; Hysing 2021). Public organizations engage the private sector and individuals to evaluate these policies and address concerns, promoting democratic responsibility (Neumann, Guirguis, and Steiner 2024).
Montana’s commitment to robust governance is exemplified by its plans for a dedicated board overseeing policy effectiveness and alignment with state goals. This governance board, set to evaluate policy outcomes and monitor technological modernization progress, plays a crucial role in ensuring the state’s commitment to effective and aligned policies. In New Mexico, governance methods resemble those of other regions, focusing on policy alignment among state agencies for efficient service delivery, regulatory compliance, and coordination. Collaboration among multiple agencies, including the legislature, executive branch, and judiciary, ensures a comprehensive governance approach.
Ohio stands out for its dynamic approach to policy development, actively involving industry stakeholders and universities. The state’s commitment to a flexible policy framework, driven by industry feedback, ensures swift adaptation to emerging technologies. Industry stakeholders lead policy discussions, fostering collaboration and knowledge exchange, with universities contributing valuable insights to inform policy decisions. As an interviewee from Ohio noted:
The purpose of the test was to extract lessons from the procurement process. Specifically, the aim was to understand and learn from the sustainability perspective and subsequently analyze the findings. Additionally, Ohio, particularly I, led a study on automated vehicle deployment. We collaborated with seven other states, sharing information, and undertaking joint projects. Our initial project involved creating a roadmap for automated vehicle implementation, covering various areas such as policy, digital and physical infrastructure, multimodal transportation, and freight. This roadmap was organized into seven pillars to streamline activities and avoid duplication of efforts.
Delaware’s pragmatic approach to governance emphasizes collaboration, particularly in critical policy matters such as public safety. The close coordination between the Department of Transportation (DOT) and the state legislature exemplifies the state’s commitment to prioritizing collaboration for the well-being of residents. Oklahoma employs specialized committees in governance and policy development, including the oversight of AI and AV technology:
Much of the discussed information pertains to AASHTO, which stands for the American Association of State Highway and Transportation Officials. Apart from AASHTO, numerous other committees focus on various aspects such as traffic, safety, and maintenance, often involving discussions on AI technology. By examining the activities of states like Pennsylvania, Florida, and California, we can gain insights into their autonomous initiatives.
These committees ensure adherence to established guidelines and protocols, while a transportation advisory CAD committee oversees projects, aligning them with state objectives to foster effective governance and project implementation.
Discussion and Conclusion
AV technology, and AI more broadly, is emerging rapidly and forcing governments to respond with innovative policies to regulate the industry without stifling it. Within the United States, state governments are already playing a significant role in AV policy experimentation (Mallinson et al. 2024), as they tend to do in so many other areas. Using a mixed methods design, we find that considerable policy learning is driving early innovation in state AV policy. Moreover, competition among states likely serves as a complementary force, spurring rapid adoption and innovative policy development to attract and retain investments from the AV industry. In fact, as states figure out the contours of this new industry and how to regulate it, competition may take over as a predominate diffusion mechanism, illustrating how diffusion dynamics can change as a policy spreads (Mallinson 2021b).
The mixed methods approach allows us to capture the complex interplay between quantitative trends and the rich, contextual insights provided by state officials, offering a holistic view of the policy landscape. This integration of quantitative and qualitative findings not only validates the significant role of policy learning but also reveals that mechanisms such as interstate collaboration and private sector engagement are consistently emphasized across both methods, pointing to a cohesive understanding of policy diffusion dynamics. These types of micro-level dynamics are difficult to capture in a macro-level EHA model. While the qualitative analysis provides in-depth insights into the motivations, strategies, and contextual nuances of policy decisions, the dyadic results validate trends and offer broader context.
While not a cure all for identifying diffusion mechanisms, pairing macro-level EHA analyses with qualitative interviewing of policy experts offers a means to convincingly link the macro- and micro-levels (Leiser 2017). Importantly, the mixed-methods approach, combined with drawing from both diffusion and complex governance theories, allows us to expand our focus on AV policy decision-making from state-centric to polycentric. This broader perspective, informed by both dyadic analyses and in-depth interviews, underscores the polycentric nature of AV policy decision-making, where multiple actors, including neighboring states and private entities, play pivotal roles. Specifically, our analysis underscores learning, collaboration, and competition as key mechanisms driving AV policy diffusion, consistently highlighted across both our quantitative and qualitative findings. Our findings suggest that both macro-level quantitative analysis and micro-level qualitative insights are crucial for unearthing the significant mechanisms of policy diffusion, with our study highlighting learning, competition, and collaboration as central drivers, validated across both methodological approaches. Similarly, competition emerges as a relevant mechanism, not only stimulating policy innovation but also encouraging a diversity of approaches across states. This diversity can be instrumental in determining the most effective regulatory practices in the rapidly evolving AV industry.
Studying AV policy adoption is critical because it provides insight into how state governments are addressing the challenges and opportunities presented by this emerging technology (Jin 2011). As AVs become more prevalent on our roads, it is essential for state policymakers to have a solid understanding of the legal and regulatory framework that governs their operation (Leiman 2021). Without clear and effective policies in place, there is a risk of confusion, inconsistency, and potential safety hazards. By identifying gaps in existing policies and highlighting areas where new regulations may be required, researchers can help state governments develop a more coherent and effective policy framework for AVs (Leiman 2021).
This is particularly important given the diverse representation of social actors involved in AV policy networks and the potential implications for areas such as transportation, economic development, and public safety (Crayton and Meier 2017). We find that interstate knowledge sharing, and collaboration among public sector, private sector, and non-state actors provide a foundation for informed decision-making and policy implementation in the context of emerging technologies like AVs. This yields important implications for policymakers seeking to catch up with rapidly emerging AI technologies. Competitive economic pressures are certainly rising as governments not only across the states, but across the planet, seek to incentivize and use these technologies. But policy experimentation and cross-border learning remain critical to identifying the best approaches to regulating the potential negative externalities of AI technologies, without stifling their promise.
There are, of course, still limitations to this approach. Just because there was clear evidence of learning among the 12 states that we spoke with, that does not mean that all governments are learning. Our findings suggest that while competition among states can accelerate policy innovation, it may also lead to fragmented policies. Future research should therefore investigate how to effectively balance competition with collaboration to ensure cohesive and effective policy frameworks across states. Importantly, our interviews do not only capture leader states, but also middling and laggard adopters. Even among lagging states learning appeared to be occurring, as opposed to copying (Jansa, Hansen, and Gray 2019). But innovation diffusion is likely driven by more than one mechanism, even if one is dominating. Furthermore, specific innovation components, like enabling vehicle testing, may still be reflective of inter-state competition over fostering emergent technologies. Further, we did not ask interviewees, who are members of executive agencies, to distinguish collaboration, learning, and competition arising in the implementation process from the legislative process, which is the focus of the adoption model. The officials often are involved in both, but the two are not well distinguishable in the qualitative results.
Additionally, this study did not specifically focus on federal coercion, but intergovernmental information sharing between the states and federal agencies was a key theme in the interviews. The federal-state dynamics warrant additional research. Future research should also consider how citizen trust and AI privacy and security concerns are shaping the adoption of these policies (Kreps et al. 2023; Robles and Mallinson 2023). Given the technical complexity and considerable scientific uncertainty of AV policy, a consideration of media influences and misinformation in driving this policy innovation would be valuable (Bromley-Trujillo and Karch 2021). Media narratives around AV safety and ethics likely impact policy development, highlighting the need for policymakers to critically assess such narratives to inform balanced AV regulations. Moreover, this underscores the critical interplay between media portrayal and policy momentum in the context of emerging technologies.
Conclusion
As AI applications continue to emerge at a rapid pace, governments are forced to find ways to balance encouraging innovation while also regulating potential negative externalities. In many ways, the states are at the forefront of this effort within the United States (Mallinson et al. 2024). Using a quantitative and qualitative mixed-methods design, we offer insight into the forces driving the early spread of AV policy in the states. While we demonstrate that learning is a key feature of the early spread of this policy innovation, one cannot ignore the growing competitive economic pressures facing the states. Competitive pressures do not have to result in a race-to-the-bottom, however. Through policy experimentation and learning, a policy consensus could emerge for how to properly regulate AI technologies.
Footnotes
Data Availability Statement
Declaration of Conflicting Interests
Funding
ORCID iDs
