The transition to sustainable transportation is a global imperative, with the widespread adoption of battery electric vehicles (BEVs) representing a critical pathway towards reducing carbon emissions and combating climate change. Supported by robust industrial policies and rapid technological innovation, markets like China have witnessed exponential growth in battery EV car sales, establishing the world’s largest consumer base for such vehicles. Understanding the complex decision-making mechanisms that drive actual consumer purchasing behavior, rather than mere stated intentions, is therefore of paramount importance for refining policy instruments, optimizing market offerings, and ensuring the sustainable electrification of the vehicle fleet.
Existing literature on green vehicle adoption has extensively employed theoretical frameworks such as the Theory of Planned Behavior (TPB) and the Technology Acceptance Model (TAM) to dissect the formation of purchase intention. These studies primarily focus on psychological constructs like environmental values, perceived usefulness, and social norms. However, a well-documented “intention-behavior gap” persists, particularly as the battery EV car market matures and consumer profiles diversify. Factors such as cognitive biases regarding product attributes (e.g., range expectations) can significantly weaken the translation of intention into actual purchase. This necessitates a shift in research paradigm towards modeling real purchase decisions using advanced analytical tools that can capture the nuanced interplay of multidimensional factors, including often-overlooked travel characteristics.
This study addresses key gaps in the literature by constructing a comprehensive analytical framework. We first model purchase intention using a psychological measurement model. Subsequently, we model the actual purchase decision—the choice between a battery EV car and an internal combustion engine vehicle (ICEV)—by employing an innovative ResLogit model. This model integrates a deep residual network (ResNet) with the classic Logit framework, overcoming the limitations of the restrictive Independence of Irrelevant Alternatives (IIA) assumption and providing superior predictive accuracy and interpretability compared to traditional discrete choice models. Crucially, our model incorporates a wide array of variables, including detailed travel behavior characteristics, to provide a holistic view of the determinants of battery EV car adoption. We further conduct elasticity and cross-effect analyses to quantify impact magnitudes and explore the dynamic relationship between the EV and ICEV markets.
Literature Review and Theoretical Framework
Research on the adoption of new energy vehicles has evolved significantly. Early studies heavily focused on the impact of policy incentives, such as purchase subsidies, tax exemptions, and infrastructure investment, using methods like elasticity analysis and Analytic Hierarchy Process (AHP) to determine variable weights. As technology advanced, attention shifted towards product-centric attributes like driving range, charging speed, and total cost of ownership, recognizing these as core decision factors for consumers considering a battery EV car.
The application of psychological theories has been instrumental. Studies based on TPB and TAM have consistently identified attitudes, perceived behavioral control, and subjective norms as key antecedents of purchase intention for a battery EV car. Perceived usefulness and ease of use also play significant mediating roles. More recently, machine learning techniques, such as random forest models, have been employed to rank feature importance, revealing non-linear interactions between policy tools and product attributes and enhancing prediction accuracy.
Despite these advances, notable gaps remain. First, there is a predominant focus on purchase intention rather than actual purchase behavior, overlooking the factors that bridge the intention-action gap. Second, the role of individuals’ travel patterns—such as daily travel distance, commute mode, and trip frequency—as crucial moderating variables linking product utility to real-world usage scenarios has not been comprehensively examined. An individual’s travel needs directly influence their sensitivity to a battery EV car’s range and charging requirements. Third, most discrete choice models fail to adequately capture the complex competitive and synergistic relationships between the emerging battery EV car market and the established ICEV market. This study aims to address these gaps by integrating travel characteristics into a sophisticated choice modeling framework capable of revealing such market dynamics.
Research Methodology and Data
Data Collection and Descriptive Statistics
To empirically investigate the purchasing behavior, survey data was collected from residents in a major metropolitan area. The questionnaire consisted of two main parts: the first measured latent psychological constructs using a 7-point Likert scale, and the second presented respondents with stated preference choice scenarios between vehicle types. After data cleaning and validation checks, 310 valid responses were obtained for analysis. Reliability and validity tests confirmed the data’s robustness for modeling.
The descriptive statistics of the sample are summarized in the table below. The sample was relatively balanced in gender, predominantly married, and consisted largely of three-person households. A significant majority had considerable driving experience (3-10 years). In terms of travel characteristics, most respondents reported moderate daily travel distances and frequencies, with public transport being a common choice for both commute and leisure trips.
| Variable | Category | Percentage (%) |
|---|---|---|
| Gender | Male | 47.1 |
| Female | 52.9 | |
| Marital Status | Married | 82.9 |
| Other | 17.1 | |
| Household Size | 2 persons | 22.6 |
| 3 persons | 57.7 | |
| 4+ persons | 19.7 | |
| Driving Experience | <3 years | 18.4 |
| 3-10 years | 63.5 | |
| >10 years | 18.1 | |
| Daily Travel Distance | <10 km | 15.8 |
| 10-30 km | 64.6 | |
| >30 km | 19.6 | |
| Primary Commute Mode | Public Transport | 28.4 |
| Private Car | 24.2 |
Modeling Purchase Intention: Structural Equation Model (SEM)
We first constructed a psychological model to quantify the key determinants of purchase intention for a battery EV car. The model integrates concepts from TPB and the Diffusion of Innovations theory. The framework posits that Perceived Innovation positively influences Perceived Usefulness and Perceived Ease of Use, which in turn shape Attitude. Simultaneously, Perceived Innovation affects Perceived Behavioral Control. Attitude, Social Influence, and Perceived Behavioral Control are hypothesized as direct antecedents of Purchase Intention. The structural model is represented by the following system of equations:
$$ \text{Perceived Usefulness} = \alpha_1 + \beta_1 \text{Perceived Innovation} + \zeta_1 $$
$$ \text{Perceived Ease of Use} = \alpha_2 + \beta_2 \text{Perceived Innovation} + \zeta_2 $$
$$ \text{Attitude} = \alpha_3 + \beta_3 \text{Perceived Usefulness} + \beta_4 \text{Perceived Ease of Use} + \zeta_3 $$
$$ \text{Perceived Behavioral Control} = \alpha_4 + \beta_5 \text{Perceived Innovation} + \zeta_4 $$
$$ \text{Purchase Intention} = \alpha_5 + \beta_6 \text{Attitude} + \beta_7 \text{Social Influence} + \beta_8 \text{Perceived Behavioral Control} + \zeta_5 $$
The model was estimated using maximum likelihood estimation. The fit indices, presented below, indicate a satisfactory model fit, with all values within acceptable thresholds.
| Fit Index | Obtained Value | Recommended Threshold | Assessment |
|---|---|---|---|
| Chi-Square/df | 1.833 | < 3.0 | Good |
| RMSEA | 0.052 | < 0.08 | Good |
| CFI | 0.938 | > 0.90 | Good |
| TLI | 0.926 | > 0.90 | Good |
The standardized path coefficients revealed that Attitude ($\beta_6 = 0.474, p<0.001$), Social Influence ($\beta_7 = 0.057, p<0.01$), and Perceived Behavioral Control ($\beta_8 = 0.059, p<0.01$) all had significant positive effects on Purchase Intention for a battery EV car. The most influential path to intention originated from Perceived Innovation through Perceived Usefulness/Ease of Use and Attitude, underscoring that consumers first evaluate the practical utility of the innovative technology before forming a favorable attitude and subsequent intention to purchase a battery EV car.
Modeling Purchase Behavior: The ResLogit Model
To model the actual choice between a battery EV car and an ICEV, we employ the ResLogit model. Traditional Multinomial Logit (MNL) models are limited by the IIA assumption, which restricts substitution patterns. The ResLogit framework overcomes this by integrating a deep residual network into the utility specification, allowing it to capture complex, non-linear relationships and correlations between alternatives.
In the ResLogit model, the utility $U_{int}$ that individual $n$ derives from choosing alternative $i$ in choice situation $t$ is given by:
$$ U_{int} = V_{int} + g_{int} + \epsilon_{int} $$
where $V_{int} = \boldsymbol{\beta}’ \mathbf{x}_{int}$ is the linear-in-parameters systematic utility, $\mathbf{x}_{int}$ is the vector of observed attributes (e.g., price, range, policy variables) and individual characteristics (e.g., travel patterns, demographics). The key innovation is the deep residual term $g_{int}$, which is generated by a ResNet architecture:
$$ \mathbf{h}^{(m)}_n = f^{(m)}(\mathbf{h}^{(m-1)}_n; \boldsymbol{\theta}^{(m)}) + \mathbf{h}^{(m-1)}_n, \quad \text{with} \quad \mathbf{h}^{(0)}_n = \mathbf{V}_n $$
$$ g_{int} = \boldsymbol{\omega}_i \cdot \mathbf{h}^{(M)}_n $$
Here, $f^{(m)}$ is a non-linear activation function at layer $m$, $\boldsymbol{\theta}^{(m)}$ are network parameters, and $\mathbf{h}^{(M)}_n$ is the final hidden layer representation. This residual network learns a complex function that captures unobserved heterogeneity and cross-alternative dependencies. The error term $\epsilon_{int}$ is assumed to be i.i.d. Gumbel distributed. The choice probability then becomes:
$$ P_{int} = \frac{\exp(V_{int} + g_{int})}{\sum_{j \in C} \exp(V_{jnt} + g_{jnt})} $$
This formulation allows the model to learn a rich representation that can reveal, for instance, how the utility of choosing a battery EV car is related to the utility of the ICEV alternative, going beyond the limitations of the standard Logit model.
We estimated both a traditional MNL model and the ResLogit model on our data. The input variables for the choice model included the latent Purchase Intention score from the SEM, socio-demographics, detailed travel characteristics (commute mode, frequency, distance), vehicle attributes (battery EV car price, range, charging time), policy variables (purchase tax, license plate policy), and charging infrastructure variables.
| Variable Category | Example Variables |
|---|---|
| Psychological Factor | Purchase Intention (from SEM) |
| Travel Characteristics | Daily Travel Distance, Commute Mode (Car/Public Transport), Trip Frequency |
| Vehicle Attributes | Battery EV Car Price, Driving Range, Charging Time; ICEV Price (reference) |
| Policy & Infrastructure | License Plate Policy (1 for favorable), Purchase Tax, Home Charger Availability, Charging Station Density |
| Socio-Demographics | Age, Income, Household Size, Driving Experience |
Analysis and Results
Model Estimation Results
The parameter estimates for key variables from both the traditional Logit and the ResLogit models are compared below. The ResLogit model showed superior overall fit, with a higher log-likelihood and lower prediction error rate.
| Variable | ResLogit Coefficient | Logit Coefficient |
|---|---|---|
| Constant | -0.197* | -0.141* |
| Psychological & Awareness | ||
| Purchase Intention | 0.327*** | 0.165*** |
| Prior Test Drive Experience | 1.102*** | 0.510*** |
| Travel Characteristics | ||
| Daily Travel Distance | -0.110** | -0.045** |
| Commute by Public Transport | -0.611*** | -0.496** |
| Commute by Private Car | 0.001 | -0.084 |
| Vehicle & Infrastructure | ||
| Battery EV Car Price | -0.122* | -0.228** |
| Battery EV Car Driving Range | 0.420** | 0.032 |
| Battery EV Car Charging Time | -0.378*** | -0.249*** |
| Home Charger Availability | 0.899*** | 0.396** |
| Policy Factors | ||
| Favorable License Plate Policy | 0.899*** | 0.396** |
| Purchase Tax | 0.206** | -0.059* |
| Demographics | ||
| Household Size | 0.001 | -0.057** |
| Driving Experience | -0.251*** | -0.135** |
| Model Fit | ||
| Log-Likelihood at Convergence | -292.53 | -298.69 |
| Prediction Error Rate | 9.4% | 14.7% |
* p<0.05, ** p<0.01, *** p<0.001
The negative constant in both models confirms the inherent market disadvantage for a battery EV car under neutral conditions. The ResLogit results provide several key insights:
- Psychological & Experience Factors: Purchase intention has a strong positive effect, validating its role as a key mediator. Direct experience (test drive) is an exceptionally powerful predictor of choosing a battery EV car.
- Travel Characteristics Matter: Longer daily travel distances and reliance on public transport for commute negatively impact the probability of choosing a battery EV car, reflecting persistent range and convenience concerns. The ResLogit model more accurately captured the nuanced role of car commuting compared to the Logit model.
- Product Attributes are Critical: The driving range of a battery EV car is a significant positive driver, while its price and charging time are significant negative factors. The strong positive coefficient for home charger availability highlights the importance of convenient charging.
- Policy is a Key Lever: A favorable license plate policy (e.g., dedicated “green plates” with access privileges) is one of the most potent positive influencers for battery EV car adoption.
- Demographics: Interestingly, more experienced drivers are less likely to choose a battery EV car, possibly due to ingrained habits or skepticism towards new technology.
Elasticity Analysis
To quantify the relative impact of changes in key variables, we computed direct elasticities. The elasticity $E_{P_i, x_k}$ measures the percentage change in the choice probability $P_i$ for the battery EV car given a 1% change in variable $x_k$:
$$ E_{P_i, x_k} = \frac{\partial P_i}{\partial x_k} \cdot \frac{x_k}{P_i} $$
The calculated elasticities for selected variables are summarized below:
| Variable | Elasticity Sign | Interpretation |
|---|---|---|
| Purchase Intention | Positive (+) | Increasing intention raises adoption probability. |
| Daily Travel Distance | Negative (-) | Longer distances reduce adoption probability. |
| Battery EV Car Price | Negative (-) | Price increases reduce adoption probability. |
| Battery EV Car Driving Range | Positive (+) | Range extension is a key motivator. |
| Battery EV Car Charging Time | Negative (-) | Faster charging promotes adoption. |
| Commute by Private Car | Positive (+) | Car commuters are more likely adopters. |
The elasticity curves are monotonic, indicating consistent directional effects. This analysis clearly shows that technological improvements (increasing range, decreasing charging time and price) and policy/psychological interventions (boosting intention, facilitating home charging) are effective strategies for promoting the battery EV car.
Cross-Effect Analysis: Revealing Market Synergy
A unique advantage of the ResLogit model is its ability to output a residual matrix that captures the cross-alternative dependencies learned by the deep network. This matrix, denoted $\mathbf{Q}$, reflects how the utility of one alternative is related to the utility of another. The elements of this matrix for the final residual layer are shown below:
$$ \mathbf{Q}^{(M)} = \begin{bmatrix} q_{EE} & q_{EI} \\ q_{IE} & q_{II} \end{bmatrix} $$
Where ‘E’ stands for battery EV car and ‘I’ for ICEV. In our estimated model, the matrix values were:
$$ \mathbf{Q}^{(M)} \approx \begin{bmatrix} 0.77 & 1.09 \\ 0.09 & 1.92 \end{bmatrix} $$
The off-diagonal elements $q_{EI}$ and $q_{IE}$ are both positive. This indicates a positive cross-effect or synergy between the markets. Specifically, a higher utility for the battery EV car is associated with a higher utility for the ICEV, and vice-versa. This seemingly counterintuitive result can be interpreted as a market expansion or reinforcement effect. As consumer interest and general utility for automobiles (stimulated by innovation, marketing, economic growth, or improved infrastructure) increase, it lifts the perceived utility of both vehicle types. The growth of the battery EV car market does not simply cannibalize the ICEV market in a zero-sum game; instead, it can stimulate overall automotive interest. However, the choice outcome is determined by the relative utility. The fact that $q_{EI} (1.09) > q_{IE} (0.09)$ suggests that growth in the ICEV market utility provides a stronger boost to the battery EV car’s utility than the reverse. This finding offers a novel theoretical perspective, suggesting that the development of the battery EV car market and the traditional automotive market can be mutually reinforcing during the transition phase.
Discussion and Implications
This study makes several contributions to the understanding of battery EV car adoption. By integrating a psychological intention model with a advanced discrete choice model of actual behavior, we bridge a critical gap in the literature. The empirical results underscore that while purchase intention is important, it is mediated and moderated by concrete factors related to practical use.
The significant role of travel characteristics highlights the need for a user-centric approach in policy and marketing. Potential buyers of a battery EV car assess the technology through the lens of their daily mobility patterns. Individuals with shorter, more predictable commutes and access to home charging are the natural early majority. For others, addressing range anxiety through infrastructure (ultra-fast public charging) and technology (longer-range batteries) remains paramount.
The ResLogit model proved to be a superior tool, not only in fit statistics but in providing more behaviorally plausible estimates, particularly for travel and demographic variables. Its ability to uncover the synergistic cross-effects between vehicle markets is a particularly valuable insight for industry strategists and policymakers. It suggests that policies supporting the overall automotive ecosystem (e.g., road infrastructure, consumer financing) can indirectly benefit the adoption of battery EV cars.
Conclusion and Policy Recommendations
This research provides a comprehensive analysis of the factors influencing the purchase of a battery EV car, moving beyond intention to focus on actual choice behavior. Using integrated SEM and ResLogit modeling, we identified purchase intention, driving range, license plate policies, private car commuting, and home charger availability as key positive drivers. Daily travel distance, purchase price, charging time, and reliance on public transport were identified as significant barriers.
Based on these findings, we propose the following multi-pronged strategy to accelerate the adoption of battery EV cars:
- Technology-Push: Continue R&D investments to lower costs, extend the driving range of every battery EV car, and reduce charging times. This directly addresses the top negative elasticities.
- Infrastructure-Pull: Accelerate the deployment of ubiquitous fast-charging networks and incentivize the installation of home/workplace chargers. This reduces “range anxiety” and integrates the battery EV car seamlessly into daily life.
- Targeted Demand-Side Policies: Maintain non-financial incentives like dedicated license plate privileges and access to bus lanes, as their impact is potent. Financial incentives should be carefully tailored and potentially shifted towards infrastructure investment and R&D.
- Consumer Engagement: Facilitate widespread test-drive opportunities and leverage social influence through early adopter testimonials. Marketing should emphasize the practical utility and compatibility of a battery EV car with specific travel patterns.
- Holistic Market Development: Recognize and foster the synergistic relationship between the EV and ICEV markets. Policies that improve the overall consumer automotive experience can create a rising tide that lifts both markets during the transition.
Future research could extend this framework to other geographical contexts, include a wider variety of powertrains (e.g., plug-in hybrids), and utilize longitudinal data to track the evolution of decision factors over time. Employing even more advanced neural architectures like Transformers could further enhance the model’s ability to interpret complex interactions. The journey to mainstream adoption of the battery EV car is multifaceted, requiring insights from sophisticated models that reflect the complexity of real-world consumer choice.