Human-centered electric vehicle adoption framework for smart mobility: Modeling perceived range and charging anxiety as a psychological barrier

Abstract

Electric vehicles (EVs) offer a transformative pathway toward reducing the environmental, economic, and health-related externalities of internal combustion engine vehicles in urban settings. Despite substantial advances in battery technology, charging infrastructure expansion, and supportive policy incentives, EV penetration remains limited which poses challenges for smart and sustainable mobility planning. A critical yet insufficiently modeled barrier to adoption lies in the psychological perceptions surrounding electric driving range and charging reliability, which is commonly framed as “range anxiety,” but more broadly reflecting perceived range and charging anxiety. To address this gap, this study introduces a latent psychological construct capturing individuals’ perceived range and charging anxiety and integrates it into a two-level nested logit (NL) model of vehicle transaction and fuel type choice. The model jointly represents (a) vehicle transaction decisions (keep, sell, trade, and add) and (b) fuel technology choices (conventional vehicle, hybrid EV, plug-in hybrid EV (PHEV), and battery EV (BEV)), thereby providing a behaviorally realistic representation of household vehicle dynamics. Bootstrap-based inference is employed to ensure robust estimation of both measurement and structural parameters. Using the sample dataset from the State of California, the results show that perceived range and charging anxiety significantly reduces the likelihood of adopting BEVs, particularly in fleet-expansion (add) decisions, while exerting no significant effect on PHEV adoption. This distinction suggests that PHEVs may function as transitional technologies for consumers concerned about electric mobility reliability. The findings highlight the importance of smart mobility strategies that address not only technological performance and infrastructure deployment, but also psychological perceptions shaping consumer behavior. By embedding latent behavioral factors into the vehicle choice process, the proposed framework advances citizen-centered mobility modeling and provides a robust decision-support tool for urban planners and transportation policymakers seeking to accelerate sustainable vehicle transitions.

Keywords

sustainable urban mobility electric vehicle adoption range and charging anxiety psychological barriers human-centered mobility policies

1. Introduction

Urban mobility systems across the globe face persistent and growing externalities, such as traffic congestion, air pollution, fossil fuel dependency, noise, and safety concerns, many of which are exacerbated by the widespread use of conventional gasoline and diesel vehicles (CVs)—also referred to as internal combustion engine vehicles—in urban areas (EEA, 2025; IEA, 2023b; Santos et al., 2010). In response, vehicle electrification has become a central pillar of smart and sustainable mobility strategies. Electric vehicles (EVs), which include hybrid electric vehicles (HEVs), plug-in hybrid electric vehicles (PHEVs), and battery electric vehicles (BEVs), offer a promising pathway to decarbonize transportation, improve air quality, and enhance energy resilience.

Despite recent growth in market share, the large-scale adoption of EVs, particularly plug-in EVs (PEVs) encompassing PHEVs and BEVs, remains limited in the United States and globally (Alliance for Automotive Innovation, 2023; IEA, 2023a). Forecasts suggest that penetration will remain modest in the short to medium term, hindered by several barriers such as high purchase costs, limited charging infrastructure, long recharge times, and concerns about electric driving sufficiency and charging reliability (Boudet, 2019; Costa et al., 2021; Hamed et al., 2023). Of these, the interplay between technical performance and consumer perceptions of electric driving range and charging accessibility—here conceptualized as perceived range and charging anxiety—stands out as a critical inhibitor of widespread PEV adoption (Egbue and Long, 2012; Hackbarth and Madlener, 2016; Higueras-Castillo et al., 2021; Kester et al., 2019; Singh et al., 2020). This issue is especially pronounced in urban and peri-urban contexts where infrastructure limitations, behavioral uncertainties, and user perceptions intersect in complex and dynamic ways. Consistent with this socio-technical framing, recent smart-city syntheses of human mobility behavior map a broad taxonomy of mobility scenarios, which includes driving behavior and EV charging-infrastructure planning, highlighting the need for behavioral models of EV uptake within smart urban mobility systems (Borges et al., 2024).

While substantial technical progress has extended electric driving range and expanded access to public charging infrastructure, these advancements alone have not fully alleviated consumer concerns. On the vehicle side, rapid progress in battery technology (Sanguesa et al., 2021) has enabled the introduction of relatively affordable BEV models exceeding 200 miles of range in the 2020 U.S. market (US EPA, Office of Energy Efficiency and Renewable Energy, 2023). In parallel, governments have launched major initiatives to scale up charging infrastructure, including the U.S. federal government's $7.5 billion investment in national charging infrastructure under the Bipartisan Infrastructure Law (IEA, 2023a; The White House, 2023). As a result, empirical evidence increasingly suggests that the technical driving ranges of contemporary PEVs can meet the daily mobility demands of most urban drivers (Brancaccio and Deflorio, 2023; Langbroek et al., 2019; Melliger et al., 2018). However, technological adequacy does not necessarily translate into perceived reliability. Even when average daily travel distances fall well within battery limits, concerns about charging accessibility, station availability, waiting times, and the risk of being stranded may persist. Thus, consumer hesitation may reflect not only perceived battery range limitations but also broader anxieties regarding charging infrastructure sufficiency and dependability. This distinction suggests that adoption barriers are shaped by a composite perception of electric driving range and charging security, rather than by range limitations alone.

Despite significant technological and infrastructural advancements, persistent hesitation continues to be framed in the literature primarily as range anxiety¹ (Li et al., 2017; Munshi et al., 2022; Singh et al., 2020). Originally conceptualized by Griffin (1990), range anxiety refers to the fear of depleting a vehicle's battery before reaching a charging opportunity, even when objective travel requirements are likely to be met. While this concept captures an important psychological dimension of EV adoption, it often implicitly conflates concerns about battery capacity with concerns about charging access and reliability. As charging infrastructure becomes more central to the EV ecosystem, it is increasingly important to recognize that perceived vulnerability may arise not only from the vehicle's technical range but also from uncertainties surrounding charging availability and system dependability. Yet, this broader psychological dimension remains underexplored in existing EV adoption research, particularly within smart and human-centered mobility planning contexts.

Existing studies on EV adoption behavior—whether relies on psychological or economic modeling frameworks—have primarily focused on technical barriers such as battery capacity and the spatial availability of charging infrastructure, while paying comparatively less attention to the underlying psychological perceptions that shape how these attributes are interpreted by consumers (see Table 1). Although technical adequacy has improved substantially, perceived vulnerability may persist even when objective performance meets daily mobility needs. A limited number of studies explicitly incorporate range-related perceptions into behavioral models. For example, Lane et al. (2018) introduce a binary variable capturing drivers’ range concerns. However, such simplified treatments do not fully capture the multidimensional and subjective nature of perceived vulnerability associated with electrified mobility. In particular, concerns about electric driving range and charging accessibility are often intertwined in individuals’ decision-making processes.

Table 1.
A summary of studies applying discrete choice models to explore EV adoption behavior.

Methodology

Fuel type Driving range and charging

alternative infrastructure factor

Joint vehicle

Study, location Choice model CV HEV PHEV BEV decisions Technical Psychological

Beggs et al. (1981), the United States Ordered logit √ √ √

Brownstone et al. (1996), the United States Nested logit √ √ √ √

Maness and Cirillo (2012), the United States Mixed logit √ √ √ √ √

Hoen and Koetse (2014), Netherlands Mixed logit √ √ √

Hackbarth and Madlener (2016), Germany Latent class √ √ √ √ √

Smith et al. (2017), Australia Best-worst √ √ √

Cirillo et al. (2017), the United States Mixed logit √ √ √ √ √

Liao et al. (2018), Netherlands Latent class √ √ √ √

Wolbertus et al. (2018), Netherlands Mixed logit √ √ √ √

Kwon et al. (2018), Korea Multinomial logit √ √

Lane et al. (2018), the United States Multinomial probit and binary logit √ √ √ √ √

Leard (2018), the United States Mixed logit √ √ √ √ √

Liao et al. (2019), Netherlands Mixed logit and hybrid choice √ √ √ √

Qian et al. (2019), China Mixed logit √ √ √ √

Ghasri et al. (2019), Australia Hybrid choice √ √

Li et al. (2020), China Mixed logit & latent class √ √ √

Guerra and Daziano (2020), the United States Mixed logit &latent class √ √ √

Bansal et al. (2021), India Hybrid choice √ √ √

Khan et al. (2021), Canada Latent class √ √ √ √ √

Jia and Chen (2021), the United States Mixed logit √ √ √ √ √

Munshi et al. (2022), India Binary logit √ √

Hamed et al. (2023), Jordan Random parameters ordered probit √ √

Present study, the United States Nested logit with a latent construct √ √ √ √ √ √

	Methodology
Beggs et al. (1981), the United States	Ordered logit	√			√		√
Brownstone et al. (1996), the United States	Nested logit	√			√	√	√
Maness and Cirillo (2012), the United States	Mixed logit	√	√		√	√	√
Hoen and Koetse (2014), Netherlands	Mixed logit	√			√		√
Hackbarth and Madlener (2016), Germany	Latent class	√	√	√	√		√
Smith et al. (2017), Australia	Best-worst	√			√		√
Cirillo et al. (2017), the United States	Mixed logit	√	√		√	√	√
Liao et al. (2018), Netherlands	Latent class	√		√	√		√
Wolbertus et al. (2018), Netherlands	Mixed logit	√		√	√		√
Kwon et al. (2018), Korea	Multinomial logit				√		√
Lane et al. (2018), the United States	Multinomial probit and binary logit	√	√	√	√			√
Leard (2018), the United States	Mixed logit	√	√	√	√		√
Liao et al. (2019), Netherlands	Mixed logit and hybrid choice	√		√	√		√
Qian et al. (2019), China	Mixed logit	√		√	√		√
Ghasri et al. (2019), Australia	Hybrid choice				√		√
Li et al. (2020), China	Mixed logit & latent class	√			√		√
Guerra and Daziano (2020), the United States	Mixed logit &latent class	√			√		√
Bansal et al. (2021), India	Hybrid choice	√			√		√
Khan et al. (2021), Canada	Latent class	√	√	√	√		√
Jia and Chen (2021), the United States	Mixed logit	√	√	√	√		√
Munshi et al. (2022), India	Binary logit				√		√
Hamed et al. (2023), Jordan	Random parameters ordered probit				√		√
Present study, the United States	Nested logit with a latent construct	√	√	√	√	√		√

Note: “Joint Vehicle Decisions” refers to models that simultaneously model fuel type and vehicle transaction decisions (e.g., add and trade). CV = conventional gasoline/diesel vehicle; HEV = Hybrid electric vehicle; PHEV = plug-in hybrid electric vehicle; BEV = battery electric vehicle; PEV = PHEV or BEV.

Addressing this gap, the first contribution of the present study is to examine whether and how perceived range and charging anxiety influences EV adoption behavior. To this end, we introduce a latent psychological construct that captures individuals’ self-reported apprehensions related to electric driving range limitations, charging infrastructure availability, and the risk of being stranded. This construct is developed using confirmatory factor analysis (CFA) based on individuals’ responses to a set of attitudinal indicators. The estimated latent factor is then integrated into an economic-based discrete choice model to assess its behavioral impact, thereby bridging psychological theory with empirically grounded urban mobility modeling.

The second contribution of this study lies in the novel structure of our choice model, which is a two-level nested logit (NL) framework that jointly captures both the vehicle transaction decision and the choice of vehicle fuel type, including various EVs. Specifically, the upper level of the nesting structure models an individual's vehicle transaction decision among four alternatives including no transaction, sell, trade, and add. Simultaneously, the lower level models the fuel type of the acquired vehicle—whether traded-for or added—selected from a choice set including CV, HEV, PHEV, and BEV. This joint modeling framework enables a richer behavioral characterization by distinguishing between individuals willing to replace an existing vehicle with an EV and those seeking to add an EV to their household fleet. Such insights are critical for smart urban mobility planning, where household-level decisions directly inform infrastructure investments, policy incentives, and decarbonization strategies. For example, Zhu (2025) proposes a practical smart-city planning methodology that integrates urban planning principles with digital infrastructure deployment by emphasizing coordinated governance schemes, scenario-based design, and alignment of digital and physical spaces. This highlights that our distinction between adding versus replacing EVs can directly guide city planning decisions, such as where to target charging infrastructure and incentives.

The remainder of the paper is structured as follows. Section 2 reviews the relevant literature on EV adoption, with particular emphasis on range and charging anxiety, and vehicle transaction behavior. Section 3 outlines the modeling framework, including the latent construct formulation and NL structure. Section 4 describes the empirical dataset and provides summary statistics. Estimation results and key behavioral insights are presented in Section 5. Finally, Section 6 concludes the paper with a summary of findings and directions for future research.

2. Literature review

To contextualize the present study, this section reviews the existing literature on EV adoption with a focus on electric driving range and charging accessibility as both technical attributes and psychological perceptions, as well as on the joint decision-making processes surrounding vehicle acquisition. Section 2.1 surveys the treatment of electric driving range and charging infrastructure in terms of vehicle specifications, spatial availability, and system performance. Section 2.2 highlights the behavioral literature on perceived range and charging anxiety as psychological barriers to electrified mobility. Finally, Section 2.3 reviews research on the joint modeling of vehicle transaction type and fuel type choice.

2.1. Technical aspects of electric driving range and charging infrastructure

A substantial body of literature examines the technical dimension of electric mobility, particularly battery driving range and charging infrastructure availability, as core attributes shaping PEV adoption behavior at the disaggregate level² (e.g., individual decision-makers). More broadly, EV adoption studies treat this technical dimension as the central system attribute shaping consumer decisions. These studies typically adopt either psychological frameworks (such as the Theory of Planned Behavior, the Theory of Reasoned Action, and the Diffusion of Innovation) or economic frameworks, most notably discrete choice models. Psychological approaches associate an individual's behavioral intention to accept, adopt, or use EV merely with latent (unobserved) subjective constructs such as attitudes, perceptions, emotions, and symbolic meanings, commonly referred to as taste heterogeneity (see a review in Rezvani et al., 2015 and examples of empirical studies in Haustein and Jensen, 2018; Moon, 2021). Within this stream, electric driving range is often treated as a salient attribute influencing perceived usefulness and risk. However, charging accessibility and infrastructure adequacy are increasingly recognized as complementary technical determinants of adoption, as they jointly define the operational feasibility of electric mobility.

In contrast, economic models advance behavioral explanations by accounting for both latent and observed heterogeneity. The latter reflects differences across measurable individual characteristics (such as income, education, and household vehicle ownership) which can significantly influence adoption decisions. This dual capability makes discrete choice models especially valuable for analyzing policy impacts and quantifying trade-offs in EV adoption behavior (see Liao et al., 2017 for a comprehensive review). Within the economic-based literature, discrete choice models—also the methodological focus of the present study—are commonly estimated using datasets collected through stated preference (SP) experiments, where respondents are asked to make hypothetical choices among predefined vehicle alternatives. SP methods are particularly advantageous in emerging markets such as EV adoption, where actual consumer behavior may be difficult to observe due to limited market penetration. In contrast, revealed preference (RP) datasets capture real-world choices made from available alternatives. While RP data is often considered more behaviorally realistic, it is scarce in the context of EV adoption—especially in the United States—because PEVs only entered the market in 2010, following the earlier introduction of HEVs in 2000 (Alliance for Automotive Innovation, 2023). Moreover, EVs still represent less than 2% of the total U.S. vehicle stock. As a result, relatively few U.S.-based studies utilize RP data to examine EV adoption behavior. Notable exceptions include Javid and Nejat (2017), Nazari et al. (2018), Nazari et al. (2019), and Nazari et al. (2023), though these works do not explicitly investigate the role of driving range in EV adoption behavior.

Empirical studies that apply discrete choice models—closely aligned with the present research—are summarized in Table 1, which highlights key attributes such as study location, modeling approach, and treatment of the driving range factor. One of the earliest contributions is by Beggs et al. (1981), who estimate an ordered logit model using data from nine U.S. cities in which individuals rank-ordered their preferences among 16 vehicle types. Their analysis reveals a significant negative influence of limited driving range on the demand for BEVs. Another foundational study is that of Brownstone et al. (1996), who model the joint decision-making process involving both vehicle transaction and fuel type choice for one- and two-vehicle households. Using a NL model estimated on California-based survey data, they identify vehicle driving range as a salient determinant in households’ decisions to adopt EVs.

Subsequent studies continue to emphasize the influence of driving range on preferences for EV technologies. Hoen and Koetse (2014) examine Dutch consumers’ preferences for alternative fuel vehicles—including CV, BEV, and fuel cell vehicles—using a mixed logit model. Their findings indicate that strong negative sentiment toward BEVs is primarily driven by concerns over limited driving range and long refueling times. Interestingly, individuals with higher vehicle-miles of travel exhibit lower interest in alternative fuel vehicles, despite expressing a higher willingness-to-pay for increased driving range. More recently, Bansal et al. (2021) employ a hybrid choice model to analyze SP data collected in India, estimating individuals’ preferences between CV and BEV. Their results reveal a substantial willingness-to-pay, ranging from USD $7 to $40 per additional kilometer of BEV driving range (based on a baseline of 200 km). In another India-based study, Munshi et al. (2022) estimate a binary logit model to examine BEV adoption among middle-income working adults, identifying acceptable driving range as a key determinant in adoption decisions.

The above review provides robust evidence that technical aspects of electric mobility (e.g., maximum mileage per charge, charging duration, and infrastructure availability) play a critical role in shaping preferences for PEVs across various geographic and socio-economic contexts. Discrete choice models, particularly those estimated on SP datasets, have shown effective in quantifying trade-offs and capturing heterogeneity in adoption behavior. However, while numerous studies incorporated driving range and charging availability as an objective vehicle attribute, few examine how these attributes are subjectively perceived and behaviorally internalized. In most empirical applications, range is treated as a technical constraint rather than as a perceived vulnerability shaped by individual attitudes, risk perceptions, and emotional responses. Moreover, battery range and charging accessibility are often analyzed separately, despite their behavioral interdependence in shaping perceptions of vehicle reliability and mobility security. This distinction suggests that the barrier to EV adoption may not lie solely in technical range limitations, but in a broader psychological construct encompassing both electric driving range and charging security. Accordingly, there is a need for a more behaviorally nuanced modeling approach that explicitly captures perceived range and charging anxiety as a latent construct. By integrating these psychological perceptions into a discrete choice framework, the present study advances the behavioral realism of EV adoption analysis and bridges the gap between technical system attributes and consumer-level decision-making.

2.2. Psychological dimensions of electric driving range and charging accessibility

Advancements in the battery technology (Sanguesa et al., 2021) have significantly extended the driving range of recently introduced BEV models, with several surpassing 200 miles in the 2020 U.S. vehicle market (US EPA, Office of Energy Efficiency and Renewable Energy, 2023). In parallel, policy interventions aimed at expanding publicly available PEV charging infrastructure represent a complementary strategy for mitigating perceived limitations associated with electric mobility (IEA, 2023a; The White House, 2023). While improvements in battery capacity address objective range constraints, charging infrastructure expansion targets perceived accessibility and reliability concerns that shape user confidence. For example, Bonges and Lusk (2016) examine the influence of the strategic design and placement of parking spaces, and charging stations on consumer perceptions. Their findings demonstrate that infrastructure deployment can reduce psychological resistance when implemented in a coordinated and visible manner, thereby enhancing both perceived and actual charging availability. These results highlight that consumer apprehension is not solely driven by battery capacity but also by the perceived adequacy, visibility, and dependability of charging networks.

A growing body of literature examining travel patterns across diverse population segments concludes that the driving range of currently affordable PEVs is sufficient to meet the daily mobility needs of most drivers. For instance, Melliger et al. (2018) develop a simulation-based model using national travel survey data from Switzerland and Finland to assess the adequacy of BEV driving range. Their findings show that BEVs available in the 2016 vehicle market could accommodate 85–90% of all national trip distances, with this share increasing to 99% through the adoption of high-range BEVs and supportive public charging infrastructure policies. Similarly, Langbroek et al. (2019) analyze travel behavior on the island of Gotland, Sweden, where a robust charging network and EV rental programs create an environment well-prepared for EV use. By comparing the route choices of CV and BEV renters, the study finds that driving distance was not a limiting factor for either group. Moreover, renting an EV led to increased user knowledge and more positive attitudes toward EV adoption.

Despite the aforementioned advancements in driving range, which are shown to be sufficient for most daily travel needs, recent studies continue to report that range limitations remain a key barrier to PEV adoption (e.g., Munshi et al., 2022). This disconnect is largely attributed to range anxiety, a psychological phenomenon characterized by a disproportionate fear of battery depletion that is not justified by actual travel requirements (Griffin, 1990). This interpretation is supported by a growing body of literature. For instance, Li et al. (2017), in their review of studies relied in psychological theories and methods, identify driving range as a psychological factor that significantly shapes behavioral intentions to adopt, purchase, or use EVs. A similar review study by Singh et al. (2020) categorizes driving range as both a technological and psychological barrier, reinforcing its dual nature. Range anxiety itself is influenced by experiential factors such as first-time EV use, familiarity with vehicle capabilities, and prior exposure to charging infrastructure (Franke et al., 2016; Rauh et al., 2017). For a comprehensive overview of empirical studies on BEV range satisfaction, range anxiety, and PHEV electric driving range utilization, readers are referred to the systematic review by Daramy-Williams et al. (2019).

Using choice models, Lane et al. (2018) investigate the effect of range concern, which is defined as an explanatory binary variable taking the value 1 for those who are concerned about PEV driving range and 0 otherwise, on EV adoption behavior. For the empirical analysis, the authors estimate two binary logit models to explain the choice of PHEV and BEV versus other fuel types, as well as a multinomial probit model with a choice alternative set including CV, HEV, PHEV, and BEV. The study finds that range concern is among the most influential perceived vehicle performance attributes and concludes that addressing range anxiety is essential for promoting widespread BEV adoption in the United States.

In summary, while recent technological advancements have substantially improved electric driving range and expanded charging infrastructure networks, these developments alone have not fully addressed public reluctance toward PEV adoption. A growing body of research confirms that range and charging anxiety, which is a psychological construct rooted in perceived limitations rather than actual travel needs, remains a critical barrier, particularly for BEVs. However, as the EV ecosystem evolves, this apprehension increasingly reflects not only concerns about battery range but also uncertainties surrounding charging accessibility, infrastructure reliability, and the risk of being stranded. Empirical findings indicate that such perceptions are shaped by subjective experiences, such as prior EV use and familiarity with charging systems, and that they exert measurable influence on adoption decisions. Despite this behavioral relevance, most existing studies either treat range and charging concern as a simple binary variable or omit it altogether from formal behavioral models. This underscores the need for more rigorous modeling approaches that incorporate latent psychological constructs to capture the nuanced ways in which range and charging anxiety influences decision-making. The present study addresses this gap by introducing a latent construct capturing perceived range and charging anxiety and embedding it within a NL model of vehicle transaction and fuel type choice. In doing so, the study advances a more behaviorally realistic representation of EV adoption within the context of smart, human-centered urban mobility planning.

2.3. Joint modeling of vehicle transaction type and electric vehicle adoption behavior

Choice-based studies on EV adoption typically analyze the decision-making process at the individual level, where an individual selects a preferred vehicle fuel type (such as CV, HEV, PHEV, and BEV) from a set of alternatives. This decision represents one dimension of the broader vehicle ownership problem (see Anowar et al., 2014 for a review), which may be jointly determined alongside other dimensions, such as vehicle transaction type. The rationale behind joint decision making is that an individual's decision on choosing, for instance, a BEV as vehicle fuel type may be related to whether the BEV will be traded for an existing vehicle of the individual or will be added to his/her vehicle inventory. Ignoring this interdependence risks misrepresenting both the behavioral structure and policy sensitivity of EV adoption decisions.

A review of the relevant literature presented in Table 1 reveals that, despite offering valuable insights, there exist a limited number of studies on joint modeling of vehicle transaction and fuel type decisions. An example is the study by Maness and Cirillo (2012) who estimate a mixed logit model to examine individuals’ vehicle transaction type (no-transaction, sell, and add) and fuel type (CV, HEV, and BEV) using hypothetical six-year market scenarios intended to reflect evolving vehicle dynamics in the United States. Similarly, Cirillo et al. (2017) analyze individuals’ choice from an alternative set encompassing retaining current vehicle, buying gasoline vehicle, buying HEV, and buying BEV. To consider random taste heterogeneity, they estimate a mixed logit model which also captures dynamics of HEV and BEV adoption behavior as reflected by a web-based database in a nine-year hypothetical time window in the United States. While offering valuable insights, these studies do not consider comprehensive sets of vehicle transaction types and the full spectrum of EV technologies, such as HEV, PHEV, and BEV. The present study contributes to this gap by modeling a more detailed joint decision structure that simultaneously captures individuals’ choices over fuel type and transaction type, thereby offering a more behaviorally realistic and policy-relevant framework for analyzing EV adoption.

To jointly estimate multiple vehicle decisions in a choice-based modeling framework,³ Mohammadian and Miller (2003a) and Mohammadian and Miller (2003b) suggest a multilevel NL model, where each level corresponds to a separate but interrelated decision. Building on this framework, Nazari et al. (2018) examine Americans’ vehicle transaction and fuel type decisions by estimating a three-level NL model, wherein the first level gives the vehicle transaction choice among no-transaction, sell, trade, and add. The second level determines fuel type of the added and traded-for vehicles between CV and PEV, and the third level gives the type of CV from the related alternatives including gasoline, diesel, and hybrid vehicles. While innovative, their analysis relies on revealed preference (RP) data collected from two geographically distinct regions due to RP data scarcity, raising potential concerns about spatial consistency, though the authors address this through spatial transferability verification. Moreover, the datasets were collected in 2012–2013, a period likely reflective of innovators and early adopters rather than mainstream users. Moreover, their model does not differentiate between PHEVs and BEVs, instead grouping them under a single PEV category. To address these limitations, the present study extends the work of Nazari et al. (2018) by estimating a two-level NL model using a single, consistent SP dataset collected in 2019 (described in Section 4), while also considering a more inclusive set of transaction alternatives and distinguishing between PHEV and BEV adoption behavior.

3. Methodology

Figure 1 shows the proposed two-level NL modeling framework integrating a latent psychological construct explaining perceived range and charging anxiety. The model jointly represents vehicle transaction and fuel type decisions of households (as the decision-making unit) within a unified behavioral structure. At the first level, households select among four mutually exclusive vehicle transaction alternatives including no transaction (retain the existing vehicle), sell an existing vehicle without replacement, trade an existing vehicle for another vehicle, and add a new vehicle to the existing fleet. The fuel type of the vehicles decided to be acquired, either as traded-for or added vehicles, is determined at the second level of the NL model simultaneously. The fuel type options include CV, HEV, PHEV, and BEV. The combination of transaction type and fuel type yields a total of 10 distinct alternatives, as shown in the left and middle panels of Figure 1. This hierarchical structure enables the model to explicitly distinguish between fleet expansion decisions (add) and vehicle replacement decisions (trade), while simultaneously differentiating across electrification levels (HEV, PHEV, and BEV). Such differentiation is essential for identifying whether electrification occurs through substitution of CVs or through incremental growth of the household fleet. By modeling these dimensions jointly, the framework provides a more behaviorally realistic representation of EV adoption dynamics.

Figure 1.

Nested logit model of joint vehicle transaction and fuel type choice (10 alternatives) with a latent construct for perceived range and charging anxiety.

The latent construct, perceived range and charging anxiety, is estimated through CFA and incorporated into the utility functions of the structural NL model. This integration allows the analysis to capture the influence of underlying psychological perceptions, which is related to electric driving range limitations and charging accessibility, on discrete vehicle transaction and fuel type choices.

Note that defining the choice alternatives at the vehicle level is consistent with prior studies on, for instance, vehicle transaction choice (e.g., Mohammadian and Miller, 2003a; Roorda et al., 2009), and vehicle transaction and fuel type choice (e.g., Cirillo et al., 2017; Nazari et al., 2018). Another approach, which is defining the choice alternatives at the household (or individual) level, raises the issue of multiple-choice observations for those who make multiple vehicle transaction choices, for instance, selling an existing vehicle and at the same time trading an existing vehicle for a BEV. Defining the alternatives at the vehicle level avoids these issues and allows for a cleaner, one-to-one mapping between each modeled decision and each vehicle acquisition event.

The remainder of this section concisely presents the formal specification of the well-known NL (Williams, 1977) for vehicle transaction and fuel type choice along with the CFA used to estimate the latent construct and the bootstrap procedure employed to assess statistical robustness. The NL model is formulated over a set of I alternatives $[A l t_{1}, \dots, A l t_{I}] \forall i \in {1, 2, \dots, I = 10}$ , corresponding to the combinations of vehicle transaction type (no-transaction, sell, trade, add) and fuel type (CV, HEV, PHEV, BEV), as shown in the left side of Figure 1. For each observation of the sample dataset indexed as $q \in {1, 2, \dots, Q = 3, 536}$ , which is suppressed hereafter for brevity, an alternative is selected by the corresponding individual to maximize his/her utility. The utility of the $i$ ^th alternative $(U_{A l t_{i}})$ , which is expressed in equation (1), is as a function of the vector of exogenous variables $(x_{i})$ and expected value of the latent construct $(\hat{x} *)$ via the corresponding coefficients denoted by vector $α_{i}$ and $β$ , respectively. This relationship is visually represented in Figure 1 by solid arrows connecting both the latent construct and the vector of exogenous variables to the utility of choice alternatives. The unobserved random error component $(ε_{i})$ is assumed to follow an independently and identically distributed (i.i.d.) extreme value Type I distribution, with no endogeneity bias, i.e., $E (ε_{i} | x_{i}) = 0$ and $E (ε_{i} | \hat{x} *) = 0$ .

U_{A l t_{i}} = x_{i} α_{i} + \hat{x} * β + ε_{i} \forall i \in {1, 2, \dots, I}

(1)

The latent construct

x *

is estimated separately using CFA based on the sample dataset of 1230 individual decision-makers, indexed by

n \in {1, 2, \dots, N}

. For brevity, the index n is suppressed in subsequent expressions. The CFA model is specified in equation (2), which links the latent construct

x *

to a set of observed measurement indicators

l_{g}

through the associated vector of factor loadings

γ_{g}

. This relationship is visually represented in Figure 1 by the dashed arrow connecting the latent construct to its the measurement indicators (bottom-right panel). The model also incorporates a normally distributed random error term

η_{g} \sim [0, Σ]

, which captures the effects of unobserved influences, where

σ

denotes the covariance matrix. The CFA yields standardized factor loadings and individual-level latent scores

\hat{x} *

, which are subsequently incorporated into the structural NL model in equation (1). This two-step approach allows the behavioral impact of underlying psychological perceptions to be quantified within the discrete choice framework.

l_{g} = x * γ_{g} + η_{g} \forall g \in {1, 2, \dots, G}

(2)

To derive the likelihood function of the NL model, the set of I alternatives is grouped into J mutually exclusive nests, denoted as

[(A l t_{1 | 1}, \dots, A l t_{I_{1} | 1}), (A l t_{1 | 2}, \dots, A l t_{I_{2} | 2}), \dots, (A l t_{1 | J}, \dots, A l t_{I_{J} | J})]

, where each subscript

i | j

represents the

i

th alternative within

j

th nest. In this study, the nests correspond to the four vehicle transaction types (i.e., no-transaction, sell, trade, and add); hence,

J = 4

. The probability of choosing the

i

^th alternative from the

j

^th nest,

P_{i j}

, is the product of the conditional probability of choosing alternative i given nest j,

P_{i | j}

, and the marginal probability of selecting nest j,

P_{j}

, as shown in equation (3). These components are further defined in equations (4) and (5), respectively. The marginal probability

P_{j}

depends on the inclusive value (IV) of the nest, calculated using equation (6), and scaled by a nest-specific log-sum coefficient

γ_{j}

. Finally, the full sample log-likelihood, expressed in equation (7), is maximized over choice observations (i.e.,

q \in {1, 2, \dots, Q}

) via maximum likelihood estimation to obtain the model parameters.

\begin{aligned} P_{i j} = P_{i | j} \times P_{j} \end{aligned}

(3)

\begin{aligned} P_{i | j} = \frac{\exp (x_{i} α_{i} + \hat{x} * β)}{\sum_{l = 1}^{I_{j}} \exp (x_{i} α_{i} + \hat{x} * β)} \end{aligned}

(4)

\begin{aligned} P_{j} = \frac{\exp [γ_{j} I V_{j}]}{\sum_{k = 1}^{J} \exp [γ_{k} I V_{k}]} \end{aligned}

(5)

\begin{aligned} I V_{j} = \ln (\sum_{i = 1}^{I_{j}} \exp (x_{i} α_{i} + \hat{x} * β)) \end{aligned}

(6)

\begin{aligned} \log L = \sum_{q = 1}^{Q} (\ln P_{i j, q}) = \sum_{q = 1}^{Q} \ln (P_{i | j, q} \times P_{j, q}) \end{aligned}

(7)

To enhance statistical robustness, a nonparametric bootstrap procedure is applied to both the CFA and the NL components of the model. Resampling with replacement is conducted across the observations, and the models are re-estimated across replications to generate empirical distributions of the parameter estimates. Bootstrap-based standard errors (SEs), t-statistics, and percentile confidence intervals (CIs) are then computed from these distributions. This approach accounts for sampling variability across the full modeling framework, reduces reliance on asymptotic assumptions, and provides more reliable inference for the latent construct parameters and inclusive value coefficients central to the study's conclusions.

4. Data

The NL model is empirically estimated on a sample dataset from the 2019 California Vehicle Survey conducted by the California Energy Commission (2019). The survey design and sampling strategy are consistent with earlier waves administered between 2015 and 2017, as documented in Fowler et al. (2018). The dataset used in this study consists of 3536 SP choice observations regarding future vehicle transaction and fuel type decisions, provided by 1230 households. Each household is represented by one individual respondent. Descriptive statistics for the individual- and vehicle-level variables are respectively presented in Sections 4.1 and 4.2.

4.1. Individual-level variables

Table 2 presents the statistical distribution of individual-level socio-economic characteristics and travel patterns, which serve as exogenous variables in the model. Household structure is captured by the number of adults (age ≥ 18), with a sample mean of 2.172 (SD = 0.856), and the presence of children, which is reported by 40.57% of respondents. Household annual income is categorized into three groups, including low (< $75 K), medium ($75 K ≤ < $150 K), and high (≥ $150 K), comprising 39.43%, 39.02%, and 21.55% of the sample, respectively. Regarding ethnicity, 13.58% of respondents identify as Hispanic or Latino. Residential type is observed in four categories, with the majority (70.16%) living in single-family detached houses. Lastly, travel pattern is reflected by the use of ridesharing services, such as Uber and Lyft, which is observed for 24.31% of respondents.

Table 2.
Statistical distribution of individual attributes (sample size = 1230).

Exogenous variable Category # observations Share (%)

Socio-Economic Characteristic

Household structure

Number of adults (age ≥ 18) M = 2.172, SD = 0.856 — —

Presence of child(ren) Yes 499 40.57

No 731 59.43

Household annual income level Low (<$75 K) 485 39.43

Medium ($75 K ≤ < $150 K) 480 39.02

High (≥$150 K) 265 21.55

Ethnicity Hispanic or Latino 167 13.58

Non-Hispanic or Latino 1063 86.42

Residential type Single family detached house 863 70.16

Single family attached house 124 10.08

Apartment 222 18.05

Other 21 1.71

Travel pattern

Using ridesharing services Yes 299 24.31

No 931 75.69

Exogenous variable	Category	# observations	Share (%)
Socio-Economic Characteristic
Household structure
Number of adults (age ≥ 18)	M = 2.172, SD = 0.856	—	—
Presence of child(ren)	Yes	499	40.57
	No	731	59.43
Household annual income level	Low (<$75 K)	485	39.43
	Medium ($75 K ≤ < $150 K)	480	39.02
	High (≥$150 K)	265	21.55
Ethnicity	Hispanic or Latino	167	13.58
	Non-Hispanic or Latino	1063	86.42
Residential type	Single family detached house	863	70.16
	Single family attached house	124	10.08
	Apartment	222	18.05
	Other	21	1.71
Travel pattern
Using ridesharing services	Yes	299	24.31
	No	931	75.69

In addition to socio-economic and travel pattern variables, respondents were asked five questions regarding their perceived concerns about various aspects of electric driving range of PEVs. These responses form the basis for constructing the latent variable representing perceived range anxiety. Figure 2 displays the average concern levels across the sample (N = 1230) for each of the five measurement indicators. As shown, concerns associated with BEVs tend to be slightly higher than those related to PHEVs, aligning with expectations. For example, the average concern regarding the “lack of charging infrastructure outside of home” is greater for BEVs than for PHEVs. Additionally, the “fear of getting stranded with BEV,” which is specific to BEVs, registers a sample average of 0.382.

Figure 2.

Average response values for measurement indicators of plug-in electric vehicle range anxiety (sample size = 1230 individuals).

4.2. Vehicle-level variables

Table 3 presents the statistical distribution of the vehicle-related variables included in the model. The first variable is the model outcome, which captures the choice of future vehicle transaction and fuel type decision made by the 1230 sampled individuals. These decisions yield a total of 3536 SP observations. The distribution of the ten alternatives in the choice set is shown in Table 3, with the highest share attributed to “trading for a CV” at 35.27%, followed by “adding a CV” at 22.17%. Options involving PHEVs and BEVs account for smaller proportions of the observed choices. It is worth mentioning that while these proportions are relatively modest, they reflect the actual market penetration of electrified vehicles at the time of data collection (2019) and are therefore behaviorally realistic. Nevertheless, smaller cell sizes may reduce statistical precision for these alternatives. To ensure robust inference, we employ bootstrap-based SEs and CIs in the estimation results, thereby mitigating concerns related to limited alternative shares (as Sections 3 and 5 for methodological details and empirical analysis).

Table 3.
Statistical distribution of vehicle attributes (sample size = 3536).

Variable Category # observations Share (%)

Outcome variable

Vehicle transaction and fuel types No transaction—alternative 1 245 6.93

Sell—alternative 2 66 1.87

Trade, CV—alternative 3 1247 35.27

Trade, HEV—alternative 4 499 14.11

Trade, PHEV—alternative 5 146 4.13

Trade, BEV—alternative 6 103 2.91

Add, CV—alternative 7 784 22.17

Add, HEV—alternative 8 292 8.26

Add, PHEV—alternative 9 83 2.35

Add, BEV—alternative 10 71 2.01

Exogenous variable

Vehicle age before transaction (year) Mean = 2.701; SD = 4.762 (not applicable for “add” alternative) — —

Vehicle ownership type before transaction Owned 2132 60.29

Leased 174 4.92

Not applicable (i.e., “add” alternative) 1230 34.79

Variable	Category	# observations	Share (%)
Outcome variable
Vehicle transaction and fuel types	No transaction—alternative 1	245	6.93
	Sell—alternative 2	66	1.87
	Trade, CV—alternative 3	1247	35.27
	Trade, HEV—alternative 4	499	14.11
	Trade, PHEV—alternative 5	146	4.13
	Trade, BEV—alternative 6	103	2.91
	Add, CV—alternative 7	784	22.17
	Add, HEV—alternative 8	292	8.26
	Add, PHEV—alternative 9	83	2.35
	Add, BEV—alternative 10	71	2.01
Exogenous variable
Vehicle age before transaction (year)	Mean = 2.701; SD = 4.762 (not applicable for “add” alternative)	—	—
Vehicle ownership type before transaction	Owned	2132	60.29
	Leased	174	4.92
	Not applicable (i.e., “add” alternative)	1230	34.79

Note: CV = conventional gasoline/diesel vehicle; HEV = hybrid electric vehicle; PHEV = plug-in hybrid electric vehicle; BEV = battery electric vehicle.

In addition to the outcome variable, two exogenous vehicle-level variables are considered. The first is vehicle age prior to the transaction decision, which has a mean of 2.701 years and a standard deviation of 4.762 years. This variable is not applicable for the “add” alternatives, where a new vehicle is introduced without replacing an existing one. The second exogenous variable is vehicle ownership type before the transaction, which distinguishes between owned (60.29% of cases) and leased vehicles (4.92%). For 34.79% of the observations—those corresponding to “add” decisions—ownership type is not applicable.

5. Results

5.1. Confirmatory factor analysis of range and charging anxiety indicators

The latent construct, termed perceived electric range and charging anxiety, captures the individuals’ concerns about limited electric driving range and the availability of public charging infrastructure for PEVs. The construct is measured using five observed indicators (see Section 4.1 for descriptive statistics). The strength of association between the latent construct and each indicator—expressed as standardized factor loadings—is estimated through CFA using data from 1230 respondents. The estimation results are reported in Table 4. To enhance statistical robustness, bootstrap resampling with 500 replications is applied at the respondent level when estimating the CFA model. All reported SEs and CIs for the factor loadings are derived from the empirical bootstrap distributions. This approach reduces reliance on normality assumptions for the measurement errors and provides distribution-free inference for the factor loadings. To ensure model identification, the loadings are normalized to lie within [−1, 1], which are all found to be statistically significant at a 95% confidence interval.

Table 4.
Confirmatory factor analysis of the perceived range and charging anxiety construct (N = 1230).

Measurement indicator Standardized loading t-statistics

Limited driving range on electric battery of PHEV 0.556 11.17

Limited driving range on electric battery of BEV 0.525 12.27

Lack of PHEV charging infrastructure outside of home 0.399 15.28

Lack of BEV charging infrastructure outside of home 0.439 14.28

Fear of getting stranded with BEV 0.255 7.14

Measurement indicator	Standardized loading	t-statistics
Limited driving range on electric battery of PHEV	0.556	11.17
Limited driving range on electric battery of BEV	0.525	12.27
Lack of PHEV charging infrastructure outside of home	0.399	15.28
Lack of BEV charging infrastructure outside of home	0.439	14.28
Fear of getting stranded with BEV	0.255	7.14

Note: PHEV = plug-in hybrid electric vehicle; BEV = battery electric vehicle. Model fit indices: GFI = 0.993; AGFI = 0.966; SRMR = 0.029; RMSEA = 0.007.

The CFA model demonstrates excellent goodness-of-fit based on multiple fit indices. The Goodness-of-Fit Index (GFI = 0.993) and the adjusted GFI (AGFI = 0.966) both exceed the recommended threshold of 0.90 (Gao et al., 2017). The standardized root-mean-square residual (SRMR = 0.029) and the root-mean-square error of approximation (RMSEA = 0.007) also satisfy the conventional acceptance criteria of 0.05 or lower, as suggested by Byrne (2016) for SRMR, and Steiger (1990) and Browne and Cudeck (1992) for RMEA. In addition, the model chi-square is statistically significant at a 95% confidence interval (Golob, 2003). These indicators overall confirm the adequacy of the measurement model and support the validity of the latent construct used in the structural NL analysis.

The standardized factor loadings reveal meaningful variation in the strength of association between the latent construct and individual indicators. Perceived limitations in electric driving range for PHEVs (0.556) and BEVs (0.525) exhibit the strongest loadings, suggesting that perceived battery range constraints constitute the primary dimension of the construct. Concerns regarding charging infrastructure availability for BEVs (0.439) and PHEVs (0.399) also load positively and significantly, though at moderately lower magnitudes. The indicator capturing fear of being stranded with a BEV shows the weakest, yet still statistically significant, loading (0.255). It is worth noting that three of the five indicators reflect charging infrastructure availability rather than driving range per se. This pattern indicates that the latent construct extends beyond pure range anxiety and encompasses a broader psychological perception that integrates concerns about both electric driving distance and charging accessibility. Accordingly, the construct is more accurately interpreted as range and charging anxiety, representing a composite perception of technological and infrastructural constraints associated with PEV use.

The ordering of the factor loadings suggests that perceived battery range limitations remain the dominant component of the construct, while charging infrastructure concerns constitute a meaningful secondary dimension. This distinction has important policy implications. If the behavioral effects identified in the subsequent structural model are partially driven by infrastructure-related perceptions, policy interventions aimed at expanding and improving public charging networks may be as critical as advances in battery technology and information campaigns designed to reduce misconceptions about electric range.

5.2. Nested logit model of vehicle transaction and fuel type choice

The estimation results of the NL model are reported in Table 5. The model captures the joint decision-making process of vehicle transaction type (i.e., no transaction, sell, trade, and add) and fuel type (i.e., CV, HEV, PHEV, and BEV). All reported SEs and CIs are obtained using 500 bootstrap replications, providing robust inference for the nonlinear specification. The overall model fit is acceptable, with a log-likelihood ratio-based goodness-of-fit statistic $(ρ^{2} = 1 - L L (β) / L L (0))$ of 0.233 and McFadden's adjusted rho-squared $(ρ_{c}^{2} = 1 - L L (β) / L L (c))$ of 0.165. These values fall within the typical range indicating reasonable explanatory power for behavioral choice models.

Table 5.
Estimation results of the nested logit model for vehicle transaction and fuel type choice.

Exogenous variable No-transaction Trade

CV HEV PHEV BEV

Coeff. SE Coeff. SE Coeff. SE Coeff. SE Coeff. SE

(t-stat) [95% CI] (t-stat) [95% CI] (t-stat) [95% CI] (t-stat) [95% CI] (t-stat) [95% CI]

Constant — — 1.325 0.075 — — — — — —

(17.58) [1.18, 1.47]

Socio-Economic Characteristics

Household structure

# Adults (age ≥ 18) — — — — — — −1.009 0.164 −1.473 0.239

(−6.16) [−1.35, −0.69] (−6.16) [−2.00, −1.02]

With child(ren) (age < 18) — — — — — — — — −0.284 0.187

(−1.51) [−0.66, 0.07]

Household annual income

High income (≥ $150 K) — — — — 0.480 0.116 0.451 0.187 1.151 0.181

(4.15) [0.26, 0.70] (2.41) [0.04, 0.80] (3.36) [0.81, 1.49]

Ethnicity

Hispanic or Latino — — 1.473 0.187 [1.11, 1.86] 1.495 0.229 1.093 0.320 — —

(7.87) (6.52) [1.05, 1.99] (3.42) [0.37, 1.65]

Residential type

Apartment — — — — 0.457 0.133 — — — —

(3.44) [0.19, 0.69]

Travel Pattern

Used ridesharing services — — — — 0.521 0.116 — — — —

(4.51) [0.30, 0.73]

Vehicle Attributes

Log of vehicle age (year) — — 0.717 0.077 0.648 0.092 0.411 0.120 0.299 0.142

(9.33) [0.57, 0.88] (7.01) [0.48, 0.83] (3.44) [0.17, 0.64] (2.10) [−0.02, 0.55]

Vehicle ownership

Leased — — 2.749 0.442 3.185 0.467 3.698 0.483 3.454 0.502

(6.22) [2.11, 3.81] (6.82) [2.48, 4.27] (7.66) [2.98, 4.97] (6.88) [2.65, 4.59]

Latent construct

Perceived range and charging — — — — −1.049 0.486

anxiety — — — — (−2.16) [−2.06, −0.09]

Inclusive value (IV) — — 0.943 0.049

(19.34) [0.853, 1.04]

Sell Add

CV HEV PHEV BEV

Coeff.(t-stat) SE[95% CI] Coeff.(t-stat) SE[95% CI] Coeff.(t-stat) SE[95% CI] Coeff.(t-stat) SE[95% CI] Coeff.(t-stat) SE[95% CI]

Constant −1.695 0.166 1.320 0.088 — — — — — —

(−10.20) [−2.03, −1.38] (14.92) [1.17, 1.49]

Socio-economic characteristics

Household structure

# Adults (age ≥ 18) — — — — — — −1.46 0.203 −1.436 0.243

(−7.20) [−1.88, −1.11] (−5.91) [−2.03, −1.04]

With child(ren) (age < 18) — — — — — — — — −0.458 0.240

(−1.91) [−0.91, 0.03]

Household annual income

High income (≥ $150 K) — — — — — — 0.612 0.230 0.814 0.236

(2.66) [0.13, 1.03] (3.46) [0.32, 1.24]

Residential type

Apartment — — — — 0.621 0.157 — — — —

(3.97) [0.30, 0.91]

Travel pattern

Used ridesharing services — — — — 0.548 0.143 — — — —

(3.83) [0.28, 0.81]

Vehicle attributes

Log of vehicle age (year) 0.888 0.133 — — — — — — — —

(6.68) [0.60, 1.13]

Latent construct

Perceived range and charging anxiety — — — — −1.303 0.588

— — — — (−2.218) [−2.42, −0.09]

Inclusive value (IV) 1.000 0.926 0.057

(16.11) [0.81, 1.04]

Note: CV = conventional gasoline/diesel vehicle; HEV = hybrid electric vehicle; PHEV = plug-in hybrid electric vehicle; BEV = battery electric vehicle; SE = standard error; CI = confidence interval; base alternative = no-transaction; number of observations = 3536; log-likelihood at convergence $(L L (β))$ = −6246.485; log-likelihood at zero $(L L (0))$ = −8141.941; log-likelihood at constants $(L L (c))$ = −7484.911.

The validity of the two-level nesting structure—where level 1 represents the vehicle transaction decision and level 2 captures the fuel type choice for traded and added vehicles—is supported by the estimated IV parameters. Specifically, the IV parameters for the trade nest (0.943; SE = 0.049; 95% CI [0.853, 1.04]) and the add nest (0.926; SE = 0.057; 95% CI [0.81, 1.04]) fall between 0 and 1, and are statistically significant at the 95% confidence level based on bootstrap inference. For identification, the IV parameters for the degenerate no-transaction and sell nests are fixed at 1.000. Moreover, the IV estimates are closer to 1 than to 0, indicating a moderate degree of correlation between the upper-level transaction decision and the lower-level fuel type choice. This suggests that although transaction type and fuel type decisions are interrelated, the correlation structure does not collapse to a fully independent multinomial logit model. The statistically significant IV parameters, along with their bootstrap confidence intervals excluding zero, provide empirical support for the nested specification over a simple multinomial logit framework.

Using the estimated parameters, the choice probabilities for all ten alternatives are simulated over the sample. As illustrated in Figure 3, the predicted shares (gray bars) closely replicate the observed market shares (pink bars), further demonstrating the model's strong performance in reproducing real-world behavioral patterns.

Figure 3.

Observed versus estimated shares of vehicle transaction and fuel type alternatives replicated by the estimated nested logit model (sample size = 3536 choices).

The alternative-specific constants capture the baseline utility associated with each option after controlling for observed explanatory variables. As shown in Table 5, the constant for the CV alternative in the add nest is positive and statistically significant at the 95% confidence level based on bootstrap inference (SE = 0.088; 95% CI [1.17, 1.49]). In contrast, the constant for the sell alternative is negative and statistically significant (SE = 0.166; 95% CI [−2.03, −1.38]). These results indicate that, holding all observed characteristics constant, households exhibit an intrinsic preference toward adding a CV to their fleet, while selling a vehicle is associated with a baseline disutility. The positive constant for CV suggests that, independent of socio-economic and vehicle attributes, conventional vehicles remain the reference or default option in household fleet decisions. Conversely, the negative sell constant reflects a general reluctance to reduce fleet size, consistent with evidence that vehicle ownership provides flexibility and insurance against travel uncertainty. In contrast, the constant terms associated with EV alternatives (HEV, PHEV, and BEV) are statistically insignificant once observed attributes and the latent perception are controlled for. This suggests that there is no residual systematic preference for or against EVs beyond the effects captured by socio-economic characteristics, vehicle attributes, and perceived range and charging anxiety. Rather than reflecting an omitted variable bias, the absence of significant EV constants indicates that preferences for these alternatives are largely explained by the included explanatory variables, supporting the adequacy of the model specification.

The subsequent sections interpret the effects of the exogenous variables, the majority of which are statistically significant at the 95% confidence level based on bootstrap SEs and percentile CIs.

5.2.1. Effects of socio-economic characteristics

Household structure is represented by two variables: (a) the number of adults aged 18 or older residing in the household and (b) a binary indicator capturing the presence of at least one child (age < 18). The number of adults exhibits a negative effect on several PEV-related alternatives. Specifically, the coefficient is negative for both PHEV and BEV alternatives within the trade and add nests. For example, in the trade nest, the coefficient for BEV equals −1.473 (SE = 0.239; 95% CI [−2.00, −1.02]), while in the add nest the corresponding coefficient equals −1.436 (SE = 0.243; 95% CI [−2.03, −1.04]). These results suggest that households with more adult members are systematically less inclined to electrify their fleet. One plausible explanation is that larger adult households face more complex and heterogeneous mobility needs, increasing the perceived value of fuel flexibility and refueling convenience. Multiadult decision-making may also introduce greater consensus constraints and risk aversion, favoring CV technologies over alternatives perceived as infrastructure-dependent.

The presence of children further reduces the utility of BEV alternatives. In the trade nest, the coefficient for BEV is negative (−0.284; SE = 0.187), though only marginally significant, while in the add nest the effect is stronger (−0.458; SE = 0.240; 95% CI [−0.91, 0.03]), indicating a consistent downward shift in BEV utility. Although some confidence intervals approach zero, the direction of the effect remains stable across specifications, suggesting that households with children exhibit greater hesitation toward BEV adoption. This behavioral pattern may reflect heightened sensitivity to travel reliability when transporting children, particularly for time-sensitive trips such as school drop-offs and emergency travel. Similar behavioral patterns are observed in prior research, including a recent study (Singh et al., 2020) and an empirical analysis using California data (Nazari et al., 2019).

Household income enters the model as a dummy variable equal to 1 for high–income households (annual income ≥ $150 K). The results show that individuals in this group exhibit a greater propensity to adopt BEVs, followed by PHEVs and traded–for HEVs. This preference ordering is intuitive given the higher purchase and ownership costs associated with EVs, particularly BEVs, which also align with evidence from Liao et al. (2017) and an empirical study by Nazari et al. (2018). High–income households are better positioned to absorb these additional costs, making them more likely early movers in electrified mobility.

Ethnicity is found to be a significant factor, captured through a binary variable equal to 1 for individuals identifying as Hispanic or Latino. The coefficient remains significant after controlling for household income, household composition, leasing status, and other socio-economic characteristics, suggesting that the observed association is not solely attributable to income differences. This variable positively influences the utility of three traded-for alternatives, namely, HEV, CV, and PHEV, with coefficients decreasing in magnitude across the alternatives. This pattern suggests that Hispanic or Latino individuals exhibit the strongest preference for traded HEVs, followed by CVs and PHEVs, highlighting a modest but discernible tendency toward electrified options that maintain conventional fueling flexibility or familiarity. While the model does not permit causal interpretation, several potential mechanisms may underlie this association, including differences in household vehicle usage patterns, residential location, cultural attitudes toward emerging vehicle technologies, or exposure to peer adoption networks. These interpretations remain speculative, and further research incorporating interaction effects, spatial disaggregation, or qualitative insights would be necessary to disentangle the behavioral drivers underlying this observed preference pattern.

Residential type enters the model as a binary variable equal to 1 for apartment dwellers. The results indicate a positive and significant effect of apartment residence on the utility of HEV-related alternatives. This finding suggests that individuals living in apartments are more inclined to choose HEVs over PEVs. A likely explanation lies in the limited availability of private charging infrastructure in multiunit housing. Unlike residents of single-family homes, apartment dwellers may face practical barriers to overnight charging, reducing the feasibility of BEV or PHEV adoption. These findings suggest that apartment dwellers are more inclined toward HEVs relative to other fuel types. Notably, no significant positive effect is observed for BEV or PHEV alternatives. This pattern is consistent with the practical constraints associated with multiunit housing, where access to private overnight charging is often limited. Unlike residents of single-family homes, apartment dwellers may face uncertainty regarding charging availability, installation feasibility, and shared infrastructure costs. In such settings, HEVs, which require no external charging, offer partial electrification benefits without infrastructure dependence.

Hardman et al. (2018) and US DOE, Office of Energy Efficiency and Renewable Energy (2023) report similar patterns, and Nazari et al. (2018) emphasize the role of built-environment constraints in shaping EV adoption decisions. Supporting this perspective, Garcia Blàzquez et al. (2026) show that in Spanish multiunit residential buildings, charger availability and building-level power/load management dramatically influence charging convenience and cost. These results underscore that reliable and affordable home-charging access constitutes a critical enabling condition for PEV adoption. Similarly, Conte et al. (2025) show that in Winterthur, Switzerland, integrating EV charging with community solar and smart load management reduced peak electricity loads by approximately 30% while maximizing renewable energy use. Their findings illustrate that coordinated energy–mobility integration and demand management can simultaneously ease grid constraints and improve charging reliability, thereby creating more favorable conditions for PEV adoption.

These integrated findings highlight that residential charging infrastructure and demand management are key drivers of PEV adoption patterns. Accordingly, our findings suggest that targeted sustainable policies are needed to overcome charging barriers for apartment dwellers. For example, cities could require EV-ready parking with shared chargers in multiunit buildings. Utilities could complement these efforts by offering multifamily electricity rates and managed-charging programs to handle increased demand without costly grid upgrades. Outreach initiatives could further inform residents about these available charging resources and smart-charging options. By directly addressing the built-environment constraints identified in the present study, such measures would make EV adoption more feasible for apartment residents.

5.2.2. Effects of travel pattern

Travel pattern is captured through a binary variable indicating whether an individual uses ridesharing services such as Uber and Lyft. This variable enters the utility equations for the HEV-related alternatives with positive coefficients, suggesting a greater likelihood of HEV adoption among ridesharing users. Notably, the coefficient is slightly larger for the “add HEV” alternative than for “trade HEV,” indicating a stronger inclination toward increasing household vehicle holdings rather than replacing existing vehicles. This result carries important policy implications so as contrary to the assumption that increased use of shared mobility services leads to vehicle shedding, the findings suggest that ridesharing users may actually contribute to increased vehicle ownership, particularly through the addition of HEVs. One possible interpretation is that frequent exposure to mobility services heightens awareness of transportation costs, fuel efficiency, and environmental considerations, thereby motivating households to complement shared trips with a more efficient private vehicle. However, this behavioral pattern implies that policies promoting ridesharing as a standalone strategy for reducing private vehicle ownership may not automatically achieve reductions in vehicle stock.

From a sustainability perspective, these findings suggest that shared mobility initiatives should not be implemented in isolation but rather coordinated with broader vehicle electrification and travel demand management strategies. For instance, cities could align ridesharing incentives with measures that discourage net vehicle additions, such as parking pricing reforms, congestion pricing schemes, and vehicle registration fees that escalate with household fleet size. Such policies would help ensure that the environmental gains associated with hybrid adoption are not offset by growth in total vehicle ownership. In addition, incentive structures could be designed to favor vehicle replacement over fleet expansion. Targeted rebates or tax credits conditioned on the scrappage of older internal combustion engine vehicles, rather than unconditional purchase incentives, would better align electrification goals with reductions in overall vehicle stock. The results overall indicate that shared mobility adoption does not inherently substitute for private vehicle ownership and may, under certain conditions, coexist with fleet expansion. Sustainable transportation policies should therefore explicitly consider these behavioral interdependencies to prevent unintended increases in total vehicle holdings, even when the marginal additions are more fuel-efficient hybrids.

5.2.3. Effects of vehicle attributes

Vehicle age is incorporated in the model in logarithmic form and enters the utility equations for the sell alternative and all four trade alternatives (i.e., trade-for CV, HEV, PHEV, and BEV), all with positive coefficients. This indicates that, as the age of an existing vehicle increases, the likelihood of selling or trading it for another vehicle, particularly for a CV, also increases. This trend aligns with both intuitive expectations and previous empirical studies (Nazari et al., 2018). Importantly, the logarithmic specification suggests a diminishing marginal effect of vehicle age so that the impact of aging is more pronounced for relatively newer vehicles than for much older ones. Among the trade options, the strength of this effect decreases from CV to BEV, highlighting that as vehicles age, households are more inclined to replace them with traditional technologies (i.e., CVs) rather than leap directly to BEVs.

From a sustainability perspective, these dynamics highlight the importance of aligning electrification policies with natural replacement cycles. Because households appear most responsive during specific replacement windows, targeted and time-sensitive incentives that escalate with vehicle age could redirect replacement trajectories toward low-emission technologies. Examples of such incentives are enhanced rebates for trading aging internal combustion vehicles for BEVs, scrappage programs tied explicitly to electrified replacements, and emissions-based registration fees. Absent such strategic interventions, the routine aging and turnover of vehicles may inadvertently perpetuate the dominance of CVs, slowing the pace of fleet decarbonization.

Vehicle ownership type is included in the model as a dummy variable equal to 1 for leased vehicles. This variable enters the utility equations for all four traded-for fuel types, indicating a strong association between leasing and the likelihood of vehicle replacement. Specifically, individuals leasing their current vehicle are significantly more inclined to trade it in, with a descending preference for PHEV, BEV, HEV, and CV, as reflected in the respective coefficient magnitudes. This finding is consistent with prior empirical studies that highlight the positive impact of leasing on BEV adoption in California (Nazari et al., 2019) and Gotland, Sweden (Langbroek et al., 2019), as well as on the broader trend of trading in a vehicle for HEV, PHEV, or BEV in the U.S. context (Nazari et al., 2023).

One potential explanation is that leasing attracts individuals who value access to newer vehicle technologies, lower maintenance costs, and reduced concerns about depreciation or trade-in value, benefits that align well with the characteristics of EVs. Moreover, because leased vehicles typically come with mileage limits, this trend may reflect a match between the leasing model and lower vehicle-miles traveled. In this light, individuals with more moderate travel demands may find EVs, particularly PHEVs and BEVs, especially appealing when leasing is an option. From a policy perspective, these dynamics suggest that electrification strategies should extend beyond traditional purchase-based incentives. Lease-specific tax credits, rebates structured to reduce monthly payments, and manufacturer- or utility-supported leasing programs could meaningfully lower adoption barriers. Moreover, strengthening the pipeline from leased vehicles to the secondary market would expand the availability of lower-cost used BEVs, thereby broadening access across income groups and accelerating diffusion throughout the vehicle fleet.

5.2.4. Effects of latent construct

The latent construct capturing perceived range and charging anxiety is included in the model and is found to significantly influence only the BEV-related alternatives, both for added and traded-for options, with a negative sign. The bootstrap-based standard errors and 95% confidence intervals confirm the robustness of these effects, as the intervals for both BEV coefficients exclude zero. This indicates that individuals with higher levels of perceived range and charging anxiety are significantly less likely to choose BEVs, regardless of whether the decision involves replacing an existing vehicle or expanding the household fleet. In contrast, the latent construct does not significantly influence PHEV alternatives. This distinction is behaviorally meaningful. As established in the CFA (Section 5.1), the construct captures a composite psychological dimension encompassing concerns about limited electric driving range, charging infrastructure availability, and the risk of being stranded. Because PHEVs retain an internal combustion engine as a backup power source, they effectively mitigate both range and charging accessibility concerns. The absence of a significant effect for PHEVs therefore suggests that these vehicles are perceived as a lower-risk electrification pathway.

These findings are consistent with prior literature verifying that range-related perceptions primarily constrain BEV adoption rather than PHEV uptake Lane et al. (2018) identify range anxiety as a pivotal factor influencing BEV, but not PHEV, adoption in the United States, and Jia and Chen (2021) similarly report that battery range perceptions significantly affect BEV preferences alone. Furthermore, the magnitude of the negative coefficient is somewhat larger for the add-BEV alternative than for the trade-BEV alternative, indicating that anxiety-related disutility may be more pronounced when households consider expanding their fleet with a BEV rather than replacing an existing vehicle. This pattern suggests that range and charging anxiety acts as a stronger deterrent to incremental electrification (fleet expansion) than to substitution-based electrification (replacement).

The effect of this latent factor on BEV alternatives carries important policy implications. Because this construct reflects concerns about limited range, inadequate charging infrastructure, and the risk of being stranded, the findings indicate that psychological and informational barriers remain central to BEV adoption. The absence of a significant effect for PHEVs further confirms the interpretation that backup internal combustion capability reduces perceived vulnerability, positioning PHEVs as a transitional technology for risk-averse households. Accordingly, policies promoting full electrification needs to address not only technological performance but also consumer perceptions of reliability and charging security. Infrastructure strategies should therefore emphasize reliability, visibility, and usability in addition to spatial expansion. While increasing fast-charging capacity—particularly along corridors and in dense urban areas—is essential, complementary measures such as real-time charger availability systems, standardized payment platforms, and high uptime standards are equally critical for building user confidence. In residentially constrained areas, public–private partnerships can support high-power shared charging hubs to reduce perceived vulnerability.

In parallel, information and experiential interventions are needed to recalibrate perceptions. Communicating real-world range performance and total cost advantages, alongside offering short-term BEV trials or fleet demonstration programs, can narrow the gap between perceived and actual vehicle capability. Finally, because range and charging anxiety imposes a stronger disutility on adding a BEV than on replacing an existing vehicle, incentive design should prioritize replacement-based adoption, such as enhanced rebates tied to scrappage, rather than encouraging fleet expansion. This can be concluded that range and charging anxiety functions as a behavioral barrier rather than a purely technical constraint. Effective sustainable transportation policy needs to therefore combine infrastructure reliability, targeted information, experiential exposure, and well-calibrated incentives to systematically reduce perceived risk and accelerate BEV diffusion.

6. Concluding remarks

EVs are a cornerstone of decarbonization strategies aimed at achieving net-zero emissions by replacing conventional fossil-fueled vehicles (CVs). Despite substantial advancements in battery technology and significant public investment in charging infrastructure, the market penetration of plug-in electric vehicles (PEVs) remains below expectations. While limited electric driving range is often cited as a primary obstacle, accumulating evidence suggests that the barrier is not purely technological but also psychological. Concerns regarding battery sufficiency, charging accessibility, and the risk of being stranded persist even when objective travel needs are adequately met. To address this behavioral dimension, the present study develops a two-level nested logit (NL) model incorporating a latent construct capturing perceived range and charging anxiety. This construct reflects individuals’ subjective concerns about electric driving range, charging infrastructure availability, and charging reliability. The modeling framework jointly represents vehicle transaction decision (no transaction, sell, trade, and add) and fuel type choice (CV, HEV, PHEV, and BEV), thereby distinguishing between households that replace an existing vehicle and those that expand their fleet. Using a dataset from California, the model provides a behaviorally rich characterization of EV adoption pathways across transaction types and fuel technologies.

The empirical results reveal that perceived range and charging anxiety significantly reduces the likelihood of BEV adoption, with a stronger deterrent effect observed for vehicle addition relative to replacement. In contrast, the latent construct does not exert a statistically significant influence on PHEV adoption, suggesting that the presence of a backup internal combustion engine mitigates perceived range and charging concerns. This asymmetry indicates that the psychological barrier operates primarily in contexts where full electrification is required, reinforcing the transitional role of PHEVs for risk-averse consumers.

Beyond psychological factors, adoption patterns vary systematically across socio-economic and household characteristics. High-income households (≥$150 K annually) exhibit a stronger propensity toward PEV adoption, particularly BEVs. Households with a greater number of adults are less inclined to adopt PEVs, potentially reflecting greater coordination complexity or heterogeneous mobility needs. The presence of children reduces the likelihood of BEV adoption, especially as an added vehicle. HEV adoption is positively associated with ridesharing usage, suggesting that familiarity with emerging mobility services may complement hybrid adoption rather than substitute for private vehicle ownership. Moreover, leasing status plays a substantial role in shaping replacement behavior so that individuals leasing vehicles, particularly older ones, are significantly more likely to trade them for electrified options, with especially strong effects for PHEVs. Leasing may reduce perceived financial and technological risks, thereby facilitating experimentation with newer vehicle technologies.

These findings overall highlight that EV adoption is shaped by a combination of structural constraints and psychological perceptions. Policies aimed at accelerating electrification need to therefore address not only technological performance and infrastructure deployment but also perceived charging security and reliability. Interventions such as enhanced charging visibility, reliability standards, lease-based incentives, and targeted outreach to specific demographic segments may prove particularly effective in reducing psychological resistance and promoting broader BEV diffusion.

To address limitations related to both modeling structure and data, we suggest four avenues for future research. First, the empirical analysis is based on data collected in California, a state characterized by relatively high EV policy support, dense charging infrastructure, and strong pro-environmental cultural attitudes. These contextual factors may limit the external validity of the findings. In regions with less developed charging networks, lower policy incentives, and colder climates that reduce effective battery performance, perceived range and charging anxiety may manifest differently, potentially with greater intensity. Conversely, in highly mature EV markets with extensive infrastructure familiarity, the psychological barrier identified in this study may be attenuated. Future research needs to therefore replicate the proposed framework across diverse geographic contexts to examine how infrastructure maturity, climate conditions, and policy environments moderate the behavioral impact of perceived range and charging anxiety.

Second, the dataset was collected in 2019, prior to several significant shifts in the U.S. EV market. Since then, EV market share has increased substantially, mainstream BEV models commonly exceed 300 miles of range, and public charging networks have expanded. Importantly, prior research indicates that direct EV exposure, infrastructure familiarity, and social normalization reduce range-related concerns. These experiential factors have likely intensified in recent years. This raises an important question regarding the temporal stability of the psychological barrier identified in this study, namely, whether perceived range and charging anxiety represents a persistent structural constraint or a context-dependent perception that evolves with market maturity. We argue that while the magnitude of the effect may shift over time, the underlying behavioral mechanism remains relevant. Even as objective performance improves, perceived reliability, charging visibility, and experiential familiarity continue to shape consumer decision-making. However, the intensity and distribution of this latent construct may differ in post-2019 contexts. Longitudinal or repeated cross-sectional analyses are therefore essential to assess whether perceived range and charging anxiety has diminished, persisted, or transformed in response to technological progress and social normalization. Such analyses would allow researchers to recalibrate the behavioral parameters and evaluate the continued applicability of targeted policy interventions.

Third, while the survey captures respondents’ expected vehicle transaction and fuel-type decisions under realistic market conditions, it is based on stated intentions rather than revealed purchase behavior. As with all stated preference data, hypothetical bias cannot be entirely ruled out, and stated intentions may not translate perfectly into actual market outcomes. Accordingly, the findings should be interpreted as reflecting prospective behavioral tendencies rather than observed transactions. Future research could strengthen external validity by integrating retrospective panel data, combining stated and revealed preference sources, and tracking actual vehicle acquisition decisions over time. Such efforts would allow for validation of the behavioral mechanisms identified here and further assess the stability of perceived range and charging anxiety in shaping real-world EV adoption.

Fourth, future research could extend the present framework by estimating a fully simultaneous integrated choice and latent variable model, in which the measurement and choice components are jointly estimated within a single likelihood function. Such an approach would explicitly account for the covariance between the latent psychological construct and the choice process, thereby eliminating concerns related to sequential estimation and generated regressor bias. A hybrid choice specification could also allow for richer behavioral structures, including interactions between latent attitudes and observable attributes, as well as heterogeneity in psychological effects across socio-demographic groups. While computationally more demanding, particularly in multilevel nested settings, an integrated choice and latent variable extension would provide a useful robustness check and deepen the understanding of how perceived range and charging anxiety jointly evolves with vehicle transaction and fuel-type decisions.

Overall, while the present study provides evidence that perceived range and charging anxiety constitutes a meaningful behavioral barrier in the examined context, continued empirical validation across space and time is necessary to better understand how this psychological construct evolves alongside technological and infrastructural progress.

Footnotes

Acknowledgments

The authors are grateful to the California Energy Commission for database provision.

ORCID iD

Fatemeh Nazari

Author contributions

Fatemeh Nazari: conceptualization, formal analysis, methodology, data curation, software, and writing—original draft. Abolfazl (Kouros) Mohammadian: conceptualization, methodology, supervision, and writing—review and editing.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the US National Science Foundation (Grant Number 2112650).

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Notes

References

Alliance for Automotive Innovation (2023) U.S. light-duty advanced technology vehicle (ATV) sales (2011-2022). Retrieved from: https://www.autosinnovate.org/resources/electric-vehicle-sales-dashboard.

Anowar

Eluru

Miranda-Moreno

(2014) Alternative modeling approaches used for examining automobile ownership: A comprehensive review. Transport Reviews 34(4): 441–473.

Bansal

Kumar

Raj

, et al. (2021) Willingness to pay and attitudinal preferences of Indian consumers for electric vehicles. Energy Economics 100: 105340.

Beggs

Cardell

Hausman

(1981) Assessing the potential demand for electric cars. Journal of Econometrics 17(1): 1–19.

Bonges

HA III

Lusk

(2016) Addressing electric vehicle (EV) sales and range anxiety through parking layout, policy and regulation. Transportation Research Part A: Policy and Practice 83: 63–73.

Borges

CEL

Aranda

JAS

Barbosa

JLV

(2024) Analysis of human mobility behavior in smart cities and smart environments: A systematic literature review and taxonomy. Journal of Smart Cities and Society 3: 127–162.

Boudet

(2019) Public perceptions of and responses to new energy technologies. Nature Energy 4(6): 446–455.

Brancaccio

Deflorio

(2023) Extracting travel patterns from floating car data to identify electric mobility needs: A case study in a metropolitan area. International Journal of Sustainable Transportation 17(2): 181–197.

Browne

Cudeck

(1992) Alternative ways of assessing model fit. Sociological Methods & Research 21(2): 230–258.

10.

Brownstone

Bunch

Golob

, et al. (1996) A transactions choice model for forecasting demand for alternative-fuel vehicles. Research in Transportation Economics 4: 87–129.

11.

Byrne

(2016) Structural equation modeling with AMOS: Basic concepts, applications, and programming, 3rd Edition. New York, NY, USA: Routledge. https://doi.org/10.4324/9781315757421 .

12.

California Energy Commission (2019) 2019 California vehicle survey. Retrieved from: https://www.nrel.gov/tsdc. National Renewable Energy Laboratory, Transportation Secure Data Center.

13.

Cirillo

Liu

Maness

(2017) A time-dependent stated preference approach to measuring vehicle type preferences and market elasticity of conventional and green vehicles. Transportation Research Part A: Policy and Practice 100: 294–310.

14.

Conte

Carabias

Kienast

, et al. (2025) Positive energy district evolution: A test case for winterthur’s hard community. Journal of Smart Cities and Society 4: 152–165.

15.

Costa

Horta

Correia

, et al. (2021) Diffusion of electric vehicles in Brazil from the stakeholders’ perspective. International Journal of Sustainable Transportation 15(11): 865–878.

16.

Daramy-Williams

Anable

Grant-Muller

(2019) A systematic review of the evidence on plug-in electric vehicle user experience. Transportation Research Part D: Transport and Environment 71: 22–36.

17.

Egbue

Long

(2012) Barriers to widespread adoption of electric vehicles: An analysis of consumer attitudes and perceptions. Energy Policy 48(Journal Article): 717–729.

18.

European Environment Agency (EEA) (2025) Health impacts of exposure to noise from transport. Retrieved from: https://www.eea.europa.eu/ims/health-impacts-of-exposure-to-1.

19.

Fowler

Cherry

Adler

, et al. (2018) 2015–2017 California Vehicle Survey, prepared for California Energy Commission. Retrieved from https://www.nrel.gov/transportation/secure-transportation-data/assets/pdfs/cec_2015-2017_california_vehicle_survey_report.pdf (CEC-200-2018-006).

20.

Franke

Rauh

Günther

, et al. (2016) Which factors can protect against range stress in everyday usage of battery electric vehicles? Toward enhancing sustainability of electric mobility systems. Human Factors: The Journal of the Human Factors and Ergonomics Society 58(1): 13–26.

21.

Gao

Rasouli

Timmermans

, et al. (2017) Understanding the relationship between travel satisfaction and subjective well-being considering the role of personality traits: A structural equation model. Transportation Research Part F: Traffic Psychology and Behaviour 49: 110–123.

22.

Garcia Blàzquez

Villafafila-Robles

Codina-Escolar

(2026) Comparative analysis of electric vehicle charging infrastructure in Spanish multi-unit residential buildings. Journal of Smart Cities and Society 5: 3–21.

23.

Ghasri

Ardeshiri

Rashidi

(2019) Perception towards electric vehicles and the impact on consumers’ preference. Transportation Research Part D: Transport and Environment 77: 271–291.

24.

Gnann

Stephens

Lin

, et al. (2018) What drives the market for plug-in electric vehicles? A review of international PEV market diffusion models. Renewable and Sustainable Energy Reviews 93: 158–164.

25.

Golob

(2003) Structural equation modeling for travel behavior research. Transportation Research Part B: Methodological 37(1): 1–25.

26.

Griffin

(1990) Clinical methods: The history, physical, and laboratory examinations, anxiety (chapter 202). Retrieved from: https://www.ncbi.nlm.nih.gov/books/nbk315/# (3rd Edition).

27.

Guerra

Daziano

(2020) Electric vehicles and residential parking in an urban environment: Results from a stated preference experiment. Transportation Research Part D: Transport and Environment 79: 102222.

28.

Hackbarth

Madlener

(2016) Willingness-to-pay for alternative fuel vehicle characteristics: A stated choice study for Germany. Transportation Research Part A: Policy and Practice 85: 89–111.

29.

Hamed

Mustafa

Al-Sharif

, et al. (2023) Modeling the households’ satisfaction level with the first electric vehicle and the time until the purchase of the second electric vehicle. International Journal of Sustainable Transportation 17(1): 52–64.

30.

Hardman

Jenn

Tal

, et al. (2018) A review of consumer preferences of and interactions with electric vehicle charging infrastructure. Transportation Research Part D: Transport and Environment 62: 508–523.

31.

Haustein

Jensen

(2018) Factors of electric vehicle adoption: A comparison of conventional and electric car users based on an extended theory of planned behavior. International Journal of Sustainable Transportation 12(7): 484–496.

32.

Higueras-Castillo

Guillén

Herrera

L-J

, et al. (2021) Adoption of electric vehicles: Which factors are really important? International Journal of Sustainable Transportation 15(10): 799–813.

33.

Hoen

Koetse

(2014) A choice experiment on alternative fuel vehicle preferences of private car owners in The Netherlands. Transportation Research Part A: Policy and Practice 61(Journal Article): 199–215.

34.

Hossein Rashidi

Mohammadian

(Kouros) (2016) Application of a nested trivariate copula structure in a competing duration hazard-based vehicle transaction decision model. Transportmetrica A: Transport Science 12(6): 550–567.

35.

Hossein Rashidi

Mohammadian

(Kouros) and Koppelman

(2011) Modeling interdependencies between vehicle transaction, residential relocation and job change. Transportation 38(6): 909–932.

36.

International Energy Agency (IEA) (2023a) Transport: Improving the sustainability of passenger and freight transport. Retrieved from: https://www.iea.org/topics/transport.

37.

International Energy Agency (IEA) (2023b) Global EV outlook 2023: Securing supplies for an electric future. Retrieved from: https://www.iea.org/reports/global-ev-outlook-2023.

38.

Javid

Nejat

(2017) A comprehensive model of regional electric vehicle adoption and penetration. Transport Policy 54: 30–42.

39.

Jia

Chen

(2021) Are Individuals’ stated preferences for electric vehicles (EVs) consistent with real-world EV ownership patterns? Transportation Research Part D: Transport and Environment 93: 102728.

40.

Kester

de Rubens

Sovacool

, et al. (2019) Public perceptions of electric vehicles and vehicle-to-grid (V2G): Insights from a nordic focus group study. Transportation Research Part D: Transport and Environment 74: 277–293.

41.

Khan

Maoh

Dimatulac

(2021) The demand for electrification in Canadian fleets: A latent class modeling approach. Transportation Research Part D: Transport and Environment 90: 102653.

42.

Kwon

Son

Jang

(2018) Evaluation of incentive policies for electric vehicles: An experimental study on Jeju Island. Transportation Research Part A: Policy and Practice 116(Journal Article): 404–412.

43.

Lane

Dumortier

Carley

, et al. (2018) All plug-in electric vehicles are not the same: Predictors of preference for a plug-in hybrid versus a battery-electric vehicle. Transportation Research Part D: Transport and Environment 65: 1–13.

44.

Langbroek

Cebecauer

Malmsten

, et al. (2019) Electric vehicle rental and electric vehicle adoption. Research in Transportation Economics 73: 72–82.

45.

Leard

(2018) Consumer inattention and the demand for vehicle fuel cost savings. Journal of Choice Modelling 29: 1–16.

46.

Wang

Chen

, et al. (2020) Consumer preferences for battery electric vehicles: A choice experimental survey in China. Transportation Research Part D: Transport and Environment 78: 102185.

47.

Long

Chen

, et al. (2017) A review of factors influencing consumer intentions to adopt battery electric vehicles. Renewable and Sustainable Energy Reviews 78(Journal Article): 318–328.

48.

Liao

Molin

van Wee

(2017) Consumer preferences for electric vehicles: A literature review. Transport Reviews 37(3): 252–275.

49.

Liao

Molin

Timmermans

, et al. (2018) The impact of business models on electric vehicle adoption: A latent transition analysis approach. Transportation Research Part A: Policy and Practice 116(Journal Article): 531–546.

50.

Liao

Molin

Timmermans

, et al. (2019) Consumer preferences for business models in electric vehicle adoption. Transport Policy 73: 12–24.

51.

Maness

Cirillo

(2012) Measuring future vehicle preferences: Stated preference survey approach with dynamic attributes and multiyear time frame. Transportation Research Record: Journal of the Transportation Research Board 2285(1): 100–109.

52.

Melliger

van Vliet

Liimatainen

(2018) Anxiety vs reality – sufficiency of battery electric vehicle range in Switzerland and Finland. Transportation Research Part D: Transport and Environment 65: 101–115.

53.

Mohammadian

(Kouros) and Miller

(2003a) Dynamic modeling of household automobile transactions. Transportation Research Record: Journal of the Transportation Research Board 1831(1): 98–105.

54.

Mohammadian

(Kouros) and Miller

(2003b) Empirical investigation of household vehicle type choice decisions. Transportation Research Record: Journal of the Transportation Research Board 1854: 99–106.

55.

Moon

S-J

(2021) Effect of consumer environmental propensity and innovative propensity on intention to purchase electric vehicles: Applying an extended theory of planned behavior. International Journal of Sustainable Transportation 16(11): 1032–1046.

56.

Munshi

Dhar

Painuly

(2022) Understanding barriers to electric vehicle adoption for personal mobility: A case study of middle income in-service residents in Hyderabad city, India. Energy Policy 167: 112956.

57.

Nazari

Mohammadian

Stephens

(2018) Dynamic household vehicle decision modeling considering plug-in electric vehicles. Transportation Research Record: Journal of the Transportation Research Board 2672(49): 1–10. 10.1177/0361198118796925.

58.

Nazari

Mohammadian

Stephens

(2019) Modeling electric vehicle adoption considering a latent travel pattern construct and charging infrastructure. Transportation Research Part D: Transport and Environment 72(Journal Article): 65–82.

59.

Nazari

Noruzoliaee

Mohammadian

(Kouros) (2023) Electric vehicle adoption behavior and vehicle transaction decision: Estimating an integrated choice model with latent variables on a retrospective vehicle survey. Transportation Research Record: Journal of the Transportation Research Board 2678(4): 1–20. 10.1177/03611981231184875.

60.

Noel

de Rubens

Sovacool

, et al. (2019) Fear and loathing of electric vehicles: The reactionary rhetoric of range anxiety. Energy Research & Social Science 48: 96–107.

61.

Qian

Grisolía

Soopramanien

(2019) The impact of service and government-policy attributes on consumer preferences for electric vehicles in China. Transportation Research Part A: Policy and Practice 122: 70–84.

62.

Rauh

Franke

Krems

(2017) First-time experience of critical range situations in BEV use and the positive effect of coping information. Transportation Research Part F: Traffic Psychology and Behaviour 44: 30–41.

63.

Rezvani

Jansson

Bodin

(2015) Advances in consumer electric vehicle adoption research: A review and research agenda. Transportation Research Part D: Transport and Environment 34(Journal Article): 122–136.

64.

Roorda

Carrasco

Miller

(2009) An integrated model of vehicle transactions, activity scheduling, and mode choice. Transportation Research Part B: Methodological 43(2): 217–229.

65.

Sanguesa

Torres-Sanz

Garrido

, et al. (2021) A review on electric vehicles: Technologies and challenges. Smart Cities 4(1): 372–404.

66.

Santos

Behrendt

Maconi

, et al. (2010) Part I: Externalities and economic policies in road transport. Research in Transportation Economics 28(1): 2–45.

67.

Singh

Vaibhav

(2020) A review and simple meta-analysis of factors influencing adoption of electric vehicles. Transportation Research Part D: Transport and Environment 86: 102436.

68.

Smith

Olaru

Jabeen

, et al. (2017) Electric vehicles adoption: Environmental enthusiast bias in discrete choice models. Transportation Research Part D: Transport and Environment 51: 290–303.

69.

Steiger

(1990) Structural model evaluation and modification: An interval estimation approach. Multivariate Behavioral Research 25(2): 173–180.

70.

The White House (2023) Fact sheet: Biden-Harris administration announces new standards and major progress for a made-in-America national network of electric vehicle chargers. https://www.whitehouse.gov/briefing-room/statements-releases/2023/02/15/fact-sheet-biden-harris-administration-announces-new-standards-and-major-progress-for-a-made-in-america-national-network-of-electric-vehicle-chargers/.

71.

U.S. Department of Energy (US DOE), Office of Energy Efficiency and Renewable Energy (2023) Charging electric vehicles at home. https://afdc.energy.gov/fuels/electricity_charging_home.html.

72.

U.S. Environmental Protection Agency (US EPA), Office of Energy Efficiency and Renewable Energy (2023) Fuel economy guide, model year 2020.

73.

Williams

(1977) On the formation of travel demand models and economic evaluation measures of user benefit. Environment and Planning A 9(3): 285–344.

74.

Wolbertus

Kroesen

van den Hoed

, et al. (2018) Policy effects on charging behaviour of electric vehicle owners and on purchase intentions of prospective owners: Natural and stated choice experiments. Transportation Research Part D: Transport and Environment 62: 283–297.

75.

Yamamoto

Kitamura

Kimura

(1999) Competing-risks-duration model of household vehicle transactions with indicators of changes in explanatory variables. Transportation Research Record: Journal of the Transportation Research Board 1676: 116–123.

76.

Yamamoto

Madre

J-L

Kitamura

(2004) An analysis of the effects of French vehicle inspection program and grant for scrappage on household vehicle transaction. Transportation Research Part B: Methodological 38(10): 905–926.

77.

Zhu

(2025) Toward a practical methodology for smart city planning and design: Key elements, issues, and directions. Journal of Smart Cities and Society 4: 100–120.

	Methodology
		Fuel type					Driving range and charging
		alternative					infrastructure factor
						Joint vehicle
Study, location	Choice model	CV	HEV	PHEV	BEV	decisions	Technical	Psychological
Beggs et al. (1981), the United States	Ordered logit	√			√		√
Brownstone et al. (1996), the United States	Nested logit	√			√	√	√
Maness and Cirillo (2012), the United States	Mixed logit	√	√		√	√	√
Hoen and Koetse (2014), Netherlands	Mixed logit	√			√		√
Hackbarth and Madlener (2016), Germany	Latent class	√	√	√	√		√
Smith et al. (2017), Australia	Best-worst	√			√		√
Cirillo et al. (2017), the United States	Mixed logit	√	√		√	√	√
Liao et al. (2018), Netherlands	Latent class	√		√	√		√
Wolbertus et al. (2018), Netherlands	Mixed logit	√		√	√		√
Kwon et al. (2018), Korea	Multinomial logit				√		√
Lane et al. (2018), the United States	Multinomial probit and binary logit	√	√	√	√			√
Leard (2018), the United States	Mixed logit	√	√	√	√		√
Liao et al. (2019), Netherlands	Mixed logit and hybrid choice	√		√	√		√
Qian et al. (2019), China	Mixed logit	√		√	√		√
Ghasri et al. (2019), Australia	Hybrid choice				√		√
Li et al. (2020), China	Mixed logit & latent class	√			√		√
Guerra and Daziano (2020), the United States	Mixed logit &latent class	√			√		√
Bansal et al. (2021), India	Hybrid choice	√			√		√
Khan et al. (2021), Canada	Latent class	√	√	√	√		√
Jia and Chen (2021), the United States	Mixed logit	√	√	√	√		√
Munshi et al. (2022), India	Binary logit				√		√
Hamed et al. (2023), Jordan	Random parameters ordered probit				√		√
Present study, the United States	Nested logit with a latent construct	√	√	√	√	√		√

Exogenous variable	No-transaction		Trade
	No-transaction		CV		HEV		PHEV		BEV
	Coeff.	SE	Coeff.	SE	Coeff.	SE	Coeff.	SE	Coeff.	SE
	(t-stat)	[95% CI]	(t-stat)	[95% CI]	(t-stat)	[95% CI]	(t-stat)	[95% CI]	(t-stat)	[95% CI]
Constant	—	—	1.325	0.075	—	—	—	—	—	—
Constant			(17.58)	[1.18, 1.47]
Socio-Economic Characteristics
Household structure
# Adults (age ≥ 18)	—	—	—	—	—	—	−1.009	0.164	−1.473	0.239
							(−6.16)	[−1.35, −0.69]	(−6.16)	[−2.00, −1.02]
With child(ren) (age < 18)	—	—	—	—	—	—	—	—	−0.284	0.187
									(−1.51)	[−0.66, 0.07]
Household annual income
High income (≥ $150 K)	—	—	—	—	0.480	0.116	0.451	0.187	1.151	0.181
					(4.15)	[0.26, 0.70]	(2.41)	[0.04, 0.80]	(3.36)	[0.81, 1.49]
Ethnicity
Hispanic or Latino	—	—	1.473	0.187 [1.11, 1.86]	1.495	0.229	1.093	0.320	—	—
			(7.87)		(6.52)	[1.05, 1.99]	(3.42)	[0.37, 1.65]
Residential type
Apartment	—	—	—	—	0.457	0.133	—	—	—	—
					(3.44)	[0.19, 0.69]
Travel Pattern
Used ridesharing services	—	—	—	—	0.521	0.116	—	—	—	—
					(4.51)	[0.30, 0.73]
Vehicle Attributes
Log of vehicle age (year)	—	—	0.717	0.077	0.648	0.092	0.411	0.120	0.299	0.142
			(9.33)	[0.57, 0.88]	(7.01)	[0.48, 0.83]	(3.44)	[0.17, 0.64]	(2.10)	[−0.02, 0.55]
Vehicle ownership
Leased	—	—	2.749	0.442	3.185	0.467	3.698	0.483	3.454	0.502
			(6.22)	[2.11, 3.81]	(6.82)	[2.48, 4.27]	(7.66)	[2.98, 4.97]	(6.88)	[2.65, 4.59]

Latent construct
Perceived range and charging					—	—	—	—	−1.049	0.486
anxiety	—	—	—	—					(−2.16)	[−2.06, −0.09]
Inclusive value (IV)	—	—	0.943	0.049
			(19.34)	[0.853, 1.04]
	Sell		Add
			CV		HEV		PHEV		BEV
	Coeff.(t-stat)	SE[95% CI]	Coeff.(t-stat)	SE[95% CI]	Coeff.(t-stat)	SE[95% CI]	Coeff.(t-stat)	SE[95% CI]	Coeff.(t-stat)	SE[95% CI]
Constant	−1.695	0.166	1.320	0.088	—	—	—	—	—	—
	(−10.20)	[−2.03, −1.38]	(14.92)	[1.17, 1.49]
Socio-economic characteristics
Household structure
# Adults (age ≥ 18)	—	—	—	—	—	—	−1.46	0.203	−1.436	0.243
							(−7.20)	[−1.88, −1.11]	(−5.91)	[−2.03, −1.04]
With child(ren) (age < 18)	—	—	—	—	—	—	—	—	−0.458	0.240
With child(ren) (age < 18)
									(−1.91)	[−0.91, 0.03]
Household annual income
High income (≥ $150 K)	—	—	—	—	—	—	0.612	0.230	0.814	0.236
							(2.66)	[0.13, 1.03]	(3.46)	[0.32, 1.24]
Residential type
Apartment	—	—	—	—	0.621	0.157	—	—	—	—
					(3.97)	[0.30, 0.91]
Travel pattern
Used ridesharing services	—	—	—	—	0.548	0.143	—	—	—	—
					(3.83)	[0.28, 0.81]
Vehicle attributes
Log of vehicle age (year)	0.888	0.133	—	—	—	—	—	—	—	—
	(6.68)	[0.60, 1.13]
Latent construct
Perceived range and charging anxiety					—	—	—	—	−1.303	0.588
	—	—	—	—					(−2.218)	[−2.42, −0.09]
Inclusive value (IV)	1.000		0.926	0.057
			(16.11)	[0.81, 1.04]