The data utilized in this study were collected through questionnaires, combining online surveys and face-to-face interviews conducted in Beijing from December 1st to 31 December 2019. Participants provided confirmation and consent on the homepage of the online questionnaire prior to accessing the survey page. Similarly, the paper questionnaires for face-to-face surveys also included a confirmation and consent section at the beginning. Additionally, participants could discontinue the survey at any point during the process. The survey was randomly distributed across all 16 districts in Beijing (Figure 2). Given the focus on private car CO₂ emissions, we selected respondents aged between 18 and 70 with more than 1 year of driving experience to ensure data quality. To validate the collected data, logic consistency checks, and cross-validation questions were incorporated. For example, if a respondent reported a household monthly income lower than their individual monthly income, the questionnaire was marked as invalid and excluded from analysis. Face-to-face interviews were conducted at automobile sales service shops and shopping malls. When the respondents fill in the questionnaire, the interviewers stay aside and provide necessary instructions. For the online survey, the Wenjuanxing platform (https://www.wjx.cn/) was employed for data collection. Prerequisite questions are set to guarantee that the respondents own vehicles and in Beijing. A total of 600 questionnaires were distributed, with all collected. The response rate is 100%. Due to missing data, 176 questionnaires were excluded from analysis. Ultimately, 424 questionnaires are valid. The effective response rate is 70.67%.

The questionnaire contains information on four aspects as following: (1) the driver’s socio-demographic characteristics, including age, education level, family size, personal income, household income, whether there are elderly people and children in the family, and whether residence and workplace are in the same district, etc.; (2) the vehicle’s attributes, including engine displacement, vehicle age, total mileage, and fuel consumption, etc.; (3) the private car travel information of the driver’s household, including annual mileage, average daily mileage on weekdays and weekends respectively; and (4) eco-driving behaviors, encompassing most energy-saving driving behaviors in the green driving training proposed by the Ministry of Transport of the People’s Republic of China (2010), such as braking mode, accelerating habits, and air-conditioning usage, etc. The definitions and descriptions of all variables in the model are shown in Table 2.

Path analysis, a specific form of Structural Equation Model (SEM), was developed by Sewall Wright (Wright, 1921; Wright, 1934) and can be used to infer anticipated causal relationships among factors (Chen and Lei, 2017). Therefore, this paper investigates the effects of driver’s socio-demographics, travel characteristics, driving behaviors, and vehicle attributes on CO₂ emissions from private cars by establishing a path analysis model. Based on the theoretical analysis, the hypotheses of this research are proposed. The conceptual research framework is depicted in Figure 3. Existing research has shown that the driver’s socio-demographic characteristics affect vehicle selection preferences, travel distances, and travel-related CO₂ emissions. Travel distance, engine displacement, and fuel economy are likely to have significant impacts on travel-related CO₂ emissions. Additionally, these variables are likely to influence each other. For example, the driver’s family size and whether there are elderly people or children in the driver’s family are related. Family size would affect travel distance and frequency. Amos 28 is used in this study. In Amos 28 software, the observational variables are represented by rectangles. The error terms of variables are represented by circles. Arrows indicate causal relationship among variables, originating from the causal variables to outcome variables. The complete list of hypotheses is provided in the Supplementary Materials.

Hypotheses 1

The driver’s age will have a significant positive correlation with the driver’s eco-driving behaviours.

…

Hypotheses 39

The driving frequency will have a significant positive correlation with the vehicle’s annual CO₂ emissions.

Table 3 shows the path coefficients and significance of the conceptual model. Path analysis is conducted by removing the non-significant paths and estimating the residual coefficients using the standard regression path coefficients. P-values, indicating the significance of the path from independent to dependent variables, are used to select significant paths. Paths with p-value lower than 0.1 are considered statistically significant, while those with p-values higher than 0.1 are deemed non-significant and are removed. The framework and path coefficients of the adjusted model are shown in Figure 4 and Table 4.

Table 5 shows the goodness-of-fit statistics for the adjusted model and the corresponding acceptance criteria values. All indicators meet the fitted criteria, indicating a good fit of the model.

The results of the decomposition of the path coefficients of the model are shown in Table 6. The total influence in absolute value is, in this order, Mileage_y (0.934) > Engine (0.224) > FC (0.185) > Pop (0.078) > EK (0.032) > Income_p (0.018) > Eco (−0.010) > Edu_d (0.002) > Age_d (−0.001). The vehicle’s fuel consumption per 100 km (FC) and annual mileage (Mileage_y) exibited only direct effects, with no indirect effects, on the vehicle’s CO₂ emissions (E). It indicates that the impacts of FC and Mileage_y on E are not mediated by other variables. It is intuitive that these two variables, FC and Mileage_y, determine how much fuel is consumed and thus how much CO₂ is emitted. The results of the intermediary effect analysis revealed that the influences of the driver’s age (Age_d), education level (Edu_d), monthly personal income (Income_p), family size (Pop), whether there are elderly people or children in the household (EK) and eco-driving behaviors (Eco) on E are full mediation with no direct influences, indicating that these influencing factors influence E completely through mediator. The influences of Edu_d, Income_p, Pop and EK on E are positive, while the influences of Age_d and Eco are negative. This indicates that the driver’s socio-demographic characteristics and driving habits affect emissions only by altering intermediate variables in the model, such as vehicle choice, usage behavior, or fuel consumption. The vehicle engine displacement (Engine) has both a direct (0.069) and an indirect positive effect (0.155), resulting in a total effect (0.224) on E. The direct and indirect components suggest that vehicle engine displacement contributes to CO₂ emissions both directly through the vehicle’s inherent attributes and indirectly via associated fuel consumption patterns. Overall, these results underscore the importance of both vehicle-related and driver-related factors in shaping low-carbon private transportation. It emphasizes that interventions targeting only direct influencing factors (e.g. driving less or improving fuel efficiency) may be insufficient without considering upstream drivers like income, education, and driving habits, which influence emissions indirectly but substantially through the impact on vehicle choice and usage.

Figure 5 illustrates the influencing pathways of various factors on vehicle CO₂ emissions. As shown in Figure 5.1, the driver’s family size (Pop) indirectly affects CO₂ emissions (E) through the vehicle’s annual mileage (Mileage_y). An increase of one standard deviation in Pop results in a 0.083 standard deviation in Mileage_y, which increases E by 0.078 standard deviation. The results depicted in Figure 5.2 indicate that whether there are elderly people or children in the driver’s household (EK) indirectly influences E through Pop and Mileage_y. A one standard deviation increase in EK leads to a 0.414 standard deviation rise in Pop, causing a 0.034 standard deviation increase in Mileage_y, and subsequently a 0.032 standard deviation increase in E. Figure 5.3 demonstrates that the driver’s eco-driving behaviours (Eco) indirectly influence E through the fuel consumption per 100 km (FC). An increase of one standard deviation in Eco reduces FC by 0.056 standard deviation, resulting in a 0.010 standard deviation decrease in E. From Figure 5.4, the vehicle’s engine displacement (Engine) influences E both directly and indirectly via FC. A one standard deviation increases in Engine increase FC by 0.837 standard deviation, leading to a 0.155 standard deviation increase in E, with a direct effect of 0.069 standard deviation. Overall, a one standard deviation increase in Engine reaults in a total increase of 0.224 standard deviation in E, comprising direct (0.069) and indirect effects (0.155). Figure 5.5 shows that the driver’s personal income (Income_p) affects E through two indirect pathways. An increase of one standard deviation in Income_p raises Engine by 0.081 standard deviation, which further causes a 0.006 standard deviation increase in E directly and a 0.013 standard deviation increase indirectly through FC. As depicted in Figure 5.6, the driver’s education level (Edu_d) influences E through its effect on Income_p. A one standard deviation increase in Edu_d results in a 0.002 standard deviation increase in E. Figure 5.7 illustrates the influencing mechanisms of the driver’s age (Age_d) on E. An increase of one standard deviation in Age_d affects Edu_d and Eco by 0.113 and 0.115 standard deviations, respectively, leading to a 0.001 decrease in E.

This study advances existing research on the influencing factors and underlying mechanisms of transportation carbon emissions at the microscopic level in two aspects. First, it adopts a novel individual car-use perspective to analyze private car CO₂ emissions, going beyond traditional travel-based approaches that often combine various transport modes, making it difficult to isolate and analyze emissions from private vehicles. By integrating vehicle attributes, driver’s socio-demographics, travel characteristics, and driving habits into a path analysis model, this study reveals both the direct and indirect relationships among influencing factors. Second, the identification of high-emitting user profiles provides actionable insights for more targeted policy design, supporting more effective carbon reduction strategies in the private transportation sector.

Eco-driving, as individual vehicle usage behavior decisions including travel behavior and driving behavior, is a measure to improve fuel efficiency and reduce CO₂ emissions (Lois et al., 2019). Previous studies indicate that eco-driving has the potential to save 5%–10% energy consumption (Caban et al., 2019). The results of this study show that eco-driving behaviors, limited to driving behavior decisions, can save 5.6% of the vehicle’s energy consumption per 100 km and reduce 1% of the vehicle’s CO₂ emissions. Although a 1% reduction may seem modest at first glance, when aggregated across Beijing’s large and growing fleet of private vehicles, this reduction can lead to a significant impact on total transport-related emissions. As mentioned above, Beijing’s private vehicle ownership has increased continuously and is unlikely to decrease in the coming years. Additionally, it is not realistic for vehicle owners to completely give up private car use and switch to public transportation. In the short term, adopting eco-driving behaviors is an effective strategy to reduce CO₂ emissions from private vehicles in Beijing. Eco-driving training is the most direct way to promote eco-driving (Wu et al., 2018). Such training can enhance drivers’ awareness of eco-driving practices, facilitate the integration of eco-driving into their daily driving routines, improve their eco-driving experience and proficiency, and ultimately form eco-driving habits that reduce carbon emissions. Although Beijing has implemented policies to encourage eco-driving behaviors, such as the publication of the Manual of Automobile Energy Saving Driving and the promotion of eco-driving benefits through social media, radio and print media, eco-driving has not been formally incorporated into driver training and licensing examinations. Therefore, considering both the demonstrated emissions reduction potential and the scalability of behavioral change across the driver population, eco-driving should be formally integrated as a mandatory module of the initial and renewal process for driver’s license.

The results of this study show that elderly drivers are more inclined to adopt eco-driving behaviors to save fuel costs and further achieve carbon reduction. For drivers sensitive to fuel costs, integrating the use of mobile apps that monitor driving behaviors and provide eco-driving tips, along with monetary incentives for consistent eco-driving practices, is recommended. For drivers less sensitive to fuel or monetary costs, eco-driving policies can be designed from the aspects of convenience, social recognition and intrinsic motivation. From a convenience perspective, implementing priority parking for eco-friendly drivers and carpoolers can reduce parking difficulties. Developing eco-driving certification programs that award drivers with recognition for environmentally conscious driving habits, coupled with public acknowledgement programs such as “Eco-Driver of the Month”, can promote eco-driving by enhancing drivers’ social recognition and intrinsic motivation.

This study finds that the utilization of low energy-consumption vehicles could reduce the carbon emissions associated with increasing automobile ownership to a certain extent. Klöckner et al. (2013) also find that individuals who feel unable to reduce car use but have the willingness to reduce emissions tend to purchase and use low-emission cars. Identifying key socio-demographic characteristics to target specific groups is crutial for policymakers when developing effective policy interventions. The results of this study indicate that individuals with higher income and education levels exhibit lower preferences for low energy-consumption vehicles. The Beijing government should pay more attention to higher income and education groups when formulating and implementing policies to promote the adoption of low-emission vehicles. Regarding the income and education levels, the current literature presents inconsistent findings. For example, Chng et al. (2019) conducted research on socio-demographic profiles during car purchase reflections, and found that, in the UK, women, older people, non-white ethnic groups, urban residents are more likely to consider purchasing low-emission vehicles, including small-engine or environment-friendly cars and electric vehicles, while wealthier and more educated residents are less prone to. Conversely, Sovacool et al. (2018) assessed preferences for electric vehicles across Denmark, Finland, Iceland, Norway, and Sweden, and concluded that men with higher education in full-time employment are most likely to select low-emission cars. The different preferences for low-emission cars among higher income and education groups are attributed to a combination of economic capability, lifestyle preferences, social status, and environmental issues awareness (Buhmann and Criado, 2023; Knez et al., 2014). In the context of China, individuals with higher income and education levels may possess greater economic capability to purchase more expensive, higher-performance vehicles, which are often associated with higher energy consumption and carbon emissions. For many in this group, car ownership is not only a practical need but also a symbol of social status and personal success. Consequently, preferences may lean toward vehicles that reflect affluence and prestige rather than energy-efficiency and low-emission, which are sometimes perceived as budget or utilitarian options. Furthermore, although environmental awareness is increasing in China, it may not yet be the primary factor influencing car purchase decisions among consumers with high income and educational levels. Instead, lifestyle preferences, brand reputation, and perceived comfort or driving experience may weigh more heavily in their choices. Beijing has implemented stringent vehicle purchase control policies, including participation of fuel vehicles in the license plate lottery system and prioritization of new-energy vehicles for license plates. Based on current policies, designing low emission zones in key areas of Beijing–permitting only small-engine and new-energy cars–is an approach to further promote the switch to low-emission transportation.

This study finds that family size has positive effects on annual mileage and further indirectly affects private car CO₂ emissions positively. The standard coefficient of the total effect of family size on private car CO₂ emissions is 0.078, significantly less than 1, meaning that the per capita private car CO₂ emissions of larger families are lower. Larger families have more travel demands and shared mobility within the household, resulting in higher total CO₂ emissions and lower per capita CO₂ emissions. Xue and Yao (2022) observed similar positive effects of family size on private car CO₂ emissions using the Fifth Travel Survey of Beijing Inhabitants and land use data. As shown in Figure 6, over the past 3 decades, the household size in Beijing has decreased continuously. The increase of small-size families may reduce opportunities for shared car travel within households, potentially leading to increased carbon emissions. Ridesharing, drivers share empty seats via mobile applications (Shaheen and Cohen, 2021), can improve vehicle occupancy, reduce car usage, and lower carbon emissions (Lazarus et al., 2021; Si et al., 2022). However, numerous barriers, including economic, technological, business, behavioral, and regulatory obstacles, prevent the development and widespread adoption of ridesharing (Mitropoulos et al., 2021). To address economic barriers, implementing a transparent billing system with clear cost-sharing rules and dynamic pricing based on market supply and demand is suggested to reduce disputes among passengers and drivers. From the perspective of technology, using advanced algorithms to optimize passenger-driver matching and employing AI for intelligent route planning can minimize unnecessary time costs and detours. To overcome business barriers, establishing comprehensive feedback mechanisms to promptly address issues raised by users is essential. Behavioral barriers can be mitigated by enhancing identity verification, implementing real-time trip monitoring, and providing emergency alarm buttons and rapid response systems to ensure safety for both passengers and drivers. Additionally, further refinement of ridesharing-related laws and regulations is necessary.

This study extends the understanding of the factors and mechanisms influencing CO₂ emissions from an individual driving perspective. However, this study has some limitations. First, this study does not take the driver’s pro-environmental attitudes into consideration. We only introduce the driver’s eco-driving behaviors into the research framework and identify groups with better eco-driving behaviors and lower carbon emissions. Eco-driving training is an important way to improve eco-driving behaviors by improving the drivers’ awareness and attitudes towards eco-driving. The relationships between drivers’ attitudes towards eco-driving and their actual eco-driving behaviors should be further explored in future research. Secondly, electric vehicles (EVs) are not analyzed in this study. EVs are expected to significantly reduce CO₂ emissions from private vehicles, particularly when the electricity used is generated by low-carbon or zero-carbon sources (Chen et al., 2022; Choudhari et al., 2024). Due to the limited number of electric vehicles in Beijing in 2019, although the survey questionnaire was designed separately for fossil fuel vehicles and electric vehicles respectively, the sample size of electric vehicle data was insufficient for model analysis. Although the current penetration rate of EVs is still relatively low, support policies, technological advancements, and growing environmental awareness are expected to accelerate EV adoption, contributing to long-term emission reductions in the private transportation sector. Research on the influencing factors and mechanisms of the carbon footprint of EVs is also of great significance under rapid development of EVs. Finally, this study uses 2019 questionnaire data, providing a stable pre-pandemic baseline. Although China’s dual-carbon strategy began in 2021, the 2019 data remian relevant as many influencing factors, such as vehicle attributes, change gradually. Nonetheless, future research should update and expand the dataset to reflect recent developments and provide more targeted and timely policy recommendations.