The transition towards low-carbon transportation is a global imperative, and the widespread adoption of Battery Electric Vehicles (BEVs or battery EV cars) is central to this shift, particularly in the world’s largest automotive market, China. National plans aim for new energy vehicles to constitute 20% of new car sales by 2025, with battery EV cars becoming the market mainstream by 2035. While national sales figures are promising, significant regional disparities persist. Notably, adoption rates in cold northern regions like Northeast China lag considerably behind warmer provinces. The harsh winter conditions in these areas pose unique challenges for battery EV car technology, affecting battery performance, driving range, and overall user experience. Understanding the nuanced decision-making processes of consumers in these challenging environments is therefore critical for targeted policy and product development. This research addresses this gap by proposing and implementing a novel analytical framework that synergizes text mining, a Panel Mixed Logit model, and an enhanced Unified Theory of Acceptance and Use of Technology (UTAUT) to dissect the vehicle attribute preferences and psychological mechanisms governing battery EV car selection in China’s cold regions.
Existing literature on electric vehicle adoption has extensively employed theoretical models like the Theory of Planned Behavior (TPB) and the Technology Acceptance Model (TAM). The Unified Theory of Acceptance and Use of Technology (UTAUT) consolidated these into a robust framework focusing on Performance Expectancy, Effort Expectancy, Social Influence, and Facilitating Conditions. However, these approaches often rely on a priori assumptions about influencing factors, potentially overlooking real-world, context-specific concerns voiced by consumers. Furthermore, while discrete choice models, particularly Mixed Logit models, have been effectively used to quantify preferences for vehicle attributes (e.g., price, range) and policy incentives, they frequently rely on static survey data. This study innovates by first employing text mining on authentic online consumer reviews from automotive websites in cold regions. This unstructured data reveals the genuine, often unanticipated, priorities and pain points of potential battery EV car buyers, such as acute concerns over winter range degradation and charging convenience in low temperatures. These mined insights directly inform the design of our stated preference experiment and the enhancement of the psychological model.
Our methodology is tripartite. First, we collected and analyzed tens of thousands of online reviews from consumers in cold provinces. Using Term Frequency-Inverse Document Frequency (TF-IDF) and Latent Dirichlet Allocation (LDA) topic modeling, we extracted key latent themes influencing purchase considerations. Second, based on these themes, we designed a discrete choice experiment (DCE) presenting respondents with repeated choices between a conventional internal combustion engine vehicle, a Plug-in Hybrid Electric Vehicle (PHEV), and a battery EV car, each with varying attribute levels (e.g., cold-weather range, purchase price, charging infrastructure density). The repeated choices for each respondent form a panel. We estimate a Panel Mixed Logit model, which accounts for both the panel structure of the data (correlated choices by the same individual) and preference heterogeneity across the population for specific attributes. The utility Unit for individual n choosing alternative i in choice task t is given by:
$$U_{nit} = V_{nit} + \epsilon_{nit} = \boldsymbol{x}_{nit}\boldsymbol{\beta}_n + \boldsymbol{w}_{nit}\boldsymbol{\alpha} + \boldsymbol{z}_{nit}\boldsymbol{\delta}_i + \epsilon_{nit}$$
where $\boldsymbol{x}_{nit}$ are attributes with random coefficients $\boldsymbol{\beta}_n$ (e.g., cost per 100km), $\boldsymbol{w}_{nit}$ are attributes with fixed coefficients $\boldsymbol{\alpha}$, $\boldsymbol{z}_{nit}$ are individual characteristics, and $\epsilon_{nit}$ is the i.i.d. extreme value error term. The choice probability is the integral of the standard logit probability over the distribution of the random parameters:
$$P_{nit} = \int L_{nit}(\boldsymbol{\beta}) f(\boldsymbol{\beta}|\boldsymbol{\theta}) d\boldsymbol{\beta}$$
Third, we augment the standard UTAUT framework by integrating psychological constructs identified as salient from the literature and our qualitative analysis, namely Personal Innovativeness (PI) and Perceived Risk (PR), alongside Environmental Awareness (EA). We test an improved structural equation model (SEM) to explain the behavioral intention (BI) to adopt a battery EV car:
$$BI = \gamma_1 PE + \gamma_2 EE + \gamma_3 SI + \gamma_4 FC + \gamma_5 PI + \gamma_6 EA + \gamma_7 PR + \epsilon$$
Data collection involved a web-based survey targeting residents of cold Chinese provinces (Heilongjiang, Jilin, Liaoning, Inner Mongolia, etc.), resulting in 410 valid respondents completing the DCE and psychological scale items.
The text mining analysis of online reviews from automotive websites successfully identified eight core themes. These were categorized into Vehicle Attributes and Psychological Attributes, as summarized in Table 1.
| Category | Theme | Interpreted Factor | Representative Keywords |
|---|---|---|---|
| Vehicle Attributes | Topic-1 | Financial Status | fuel consumption, cost-effectiveness, price, fuel-efficient, fuel, slightly expensive |
| Topic-2 | Vehicle Exterior | body, headlights, doors, sleek, workmanship, beautiful, wheelbase, color | |
| Topic-3 | Low-Temp Charging | charging, battery, charging pile, fast charge, time, slow charge | |
| Topic-4 | Low-Temp Range | winter, range, normally, battery level, cannot drive far, mileage, can drive | |
| Topic-5 | Vehicle Interior Space | trunk, space, very large, front row, back row, storage, interior, spacious, adequate, crowded | |
| Topic-6 | Driving Experience | highway, acceleration, comfortable, smooth, speed, powerful, push-back feeling | |
| Psychological Attributes | Topic-7 | Technological Innovation | technology, pure electric, intelligent, voice control, trendy, assistance |
| Topic-8 | Environmental Awareness | low-carbon, environmental protection, new energy, hybrid, green, pollution, carbon emissions |
The Panel Mixed Logit model was estimated using Hammersley Monte Carlo sampling. Key results are presented in Table 2. The model fit was satisfactory (McFadden Pseudo R² = 0.378).
| Variable | Coefficient (β) | Std. Error | z-value | P>|z| |
|---|---|---|---|---|
| Alternative Specific Constants (Base: Conventional Vehicle) | ||||
| ASC_PHEV | 0.395 | 0.468 | 0.84 | 0.398 |
| ASC_BEV | -2.478*** | 0.644 | -3.85 | 0.000 |
| Vehicle Attributes (Fixed Coefficients) | ||||
| Purchase Price | -0.925 | 0.786 | -1.18 | 0.239 |
| Cold-Weather Range (km) | 2.001*** | 0.001 | 1.36e+03 | 0.000 |
| Fastest Charging/Refueling Time (hrs) | -0.031*** | 0.006 | -5.18 | 0.000 |
| Public Charging Station Density | 2.542*** | 0.209 | 12.15 | 0.000 |
| Vehicle Attributes (Random Coefficients) | ||||
| Cost per 100 km (Mean) | -1.035*** | 0.007 | -143.7 | 0.000 |
| Cost per 100 km (Std. Dev.) | 0.022*** | 0.003 | 8.19 | 0.000 |
| Carbon Reduction Income (Mean) | 0.837** | 0.962 | 0.87 | 0.032 |
| Carbon Reduction Income (Std. Dev.) | 0.005 | 1.560 | 0.00 | 0.998 |
| Interaction: Individual Characteristics * PHEV Alternative | ||||
| Gender (Male=1) | 0.857*** | 0.108 | 7.93 | 0.000 |
| Age | 0.417*** | 0.084 | 4.99 | 0.000 |
| Education Level | 0.372*** | 0.060 | 6.19 | 0.000 |
| Annual Household Income | 0.125** | 0.051 | 2.46 | 0.014 |
| Annual Mileage | -0.727*** | 0.086 | -8.50 | 0.000 |
| Car Purchase Plan (Yes=1) | 0.489** | 0.193 | 2.54 | 0.011 |
| Interaction: Individual Characteristics * BEV Alternative | ||||
| Gender (Male=1) | 0.869*** | 0.140 | 6.21 | 0.000 |
| Age | 0.489*** | 0.106 | 4.62 | 0.000 |
| Education Level | 0.581*** | 0.077 | 7.58 | 0.000 |
| Annual Household Income | 0.586*** | 0.069 | 8.49 | 0.000 |
| Annual Mileage | -0.935*** | 0.120 | -7.78 | 0.000 |
| Car Purchase Plan (Yes=1) | 0.216 | 0.251 | 0.86 | 0.389 |
| Number of Observations: 7,209 | Log-Likelihood: -1557.50 | ***p<0.01, **p<0.05 | ||||
The results reveal several critical insights for the cold-region battery EV car market. Regarding vehicle attributes, cold-weather driving range has an exceptionally strong positive effect (β=2.001, p<0.01), underscoring “range anxiety” as the paramount technical concern. The operating cost per 100 km shows significant negative mean preference and heterogeneity, indicating sensitivity to energy costs, which are exacerbated in winter for a battery EV car. Interestingly, increasing the density of public charging stations has a very large positive coefficient (β=2.542, p<0.01), suggesting infrastructure is a more potent catalyst for adoption than marginally improving charging speed (β=-0.031, p<0.01). A market-based policy tool, personal carbon trading income, shows a positive and significant effect (β=0.837, p<0.05), highlighting its potential as a post-subsidy incentive.
Demographic analysis indicates that the typical early adopter profile for a battery EV car or PHEV in cold regions is male, younger, highly educated, and with higher household income. A crucial finding is the strong negative effect of annual mileage on the choice of both PHEVs and BEVs (with a larger magnitude for battery EV car), indicating that high-mileage drivers are the most range-anxious and hesitant to switch.
The enhanced UTAUT structural equation model demonstrated excellent fit (CFI=0.974, RMSEA=0.033) and explained 63% of the variance in Behavioral Intention (R²=0.63). All hypothesized paths were supported except for Environmental Awareness (EA), which was not significant in this cold-region context. The standardized path coefficients are shown in Figure 1, and hypothesis test results are summarized in Table 3.
| Hypothesis | Path | Std. Coefficient | t-value | Result |
|---|---|---|---|---|
| H1 | Effort Expectancy → Performance Expectancy | 0.561*** | 8.749 | Supported |
| H2 | Effort Expectancy → Behavioral Intention | 0.191* | 2.250 | Supported |
| H3 | Performance Expectancy → Behavioral Intention | 0.122* | 2.189 | Supported |
| H4 | Facilitating Conditions → Behavioral Intention | 0.182** | 3.039 | Supported |
| H5 | Social Influence → Behavioral Intention | 0.188** | 3.081 | Supported |
| H6 | Personal Innovativeness → Behavioral Intention | 0.165* | 2.156 | Supported |
| H7 | Personal Innovativeness → Perceived Risk | -0.560*** | -9.935 | Supported |
| H8 | Perceived Risk → Behavioral Intention | -0.116* | -2.237 | Supported |
| H9 | Environmental Awareness → Behavioral Intention | -0.091 | 1.374 | Not Supported |
| ***p<0.001, **p<0.01, *p<0.05 | ||||
The psychological model reveals that Personal Innovativeness (PI) is a key driver, not only directly increasing intention to adopt a battery EV car but also indirectly by significantly reducing Perceived Risk (PR). This suggests that innovative consumers are more open to new technology like a battery EV car and view its associated risks (financial, performance, safety) as lower. The non-significant role of Environmental Awareness contradicts some studies in milder climates, emphasizing that in harsh cold regions, practical concerns about performance, cost, and infrastructure outweigh altruistic environmental motives for most consumers when considering a battery EV car.
Based on these integrated findings, multi-stakeholder policy recommendations are proposed. For Battery EV Car Manufacturers and Related Enterprises, strategies must address the core cold-region痛点: invest heavily in battery thermal management systems to mitigate winter range loss, collaborate on charging infrastructure (especially DC fast chargers), and offer differentiated pricing/leasing models with strong battery warranties to alleviate perceived risk. Targeting high-mileage drivers with tailored packages is essential.
For Government Bodies, policy should focus on deploying charging infrastructure in residential and commercial areas, implementing time-of-use electricity pricing to encourage off-peak charging, and pioneering pilot programs for personal carbon trading to provide a sustainable market-based incentive for choosing a battery EV car. Cross-departmental coordination is needed to align energy, transport, and environmental policies.
For Automotive Vertical Websites, platforms should enhance information presentation by featuring “winter-range” tests and real-user cold-weather experiences prominently. Structuring reviews and community forums around key identified concerns (charging speed in cold, range accuracy) can better inform potential battery EV car buyers and build confidence through peer testimonials.
In conclusion, this study provides a comprehensive, empirically grounded analysis of battery EV car selection behavior in China’s challenging cold regions. By integrating text mining, advanced choice modeling, and psychological theory, we move beyond generic adoption models to uncover the specific interplay of technical limitations (winter range, infrastructure), economic factors (operating cost, carbon income), and consumer psychology (innovativeness vs. risk perception). The findings underscore that accelerating battery EV car adoption in these areas requires a concerted effort: technologically robust vehicles designed for extreme climates, strategically deployed and affordable infrastructure, targeted policies that address real consumer economics and anxieties, and improved market intelligence and communication. This holistic approach is vital for achieving an equitable and nationwide transition to electric mobility, ensuring that consumers in cold regions are not left behind in the shift towards sustainable transportation.