Understanding the Driving Patterns of the Carbon Emissions in Transport Sector in China: A Panel Data Analysis and Zoning Effect

China’s transportation industry has made rapid progress, which has led to a mass of carbon emissions. However, it is still unclear how the carbon emission from transport sector is punctuated by shifts in underlying drivers. This paper aims to examine the process of China’s carbon emissions from transport sector as well as its major driving forces during the period of 2000 to 2015 at the provincial level. We firstly estimate the carbon emissions from transport sector at the provincial level based on the fuel and electricity consumption using a top-down method. We find that the carbon emission per capita is steadily increasing across the nation, especially in the provinces of Chongqing and Inner Mongolia. However, the carbon emission intensity is decreasing in most provinces of China, except in Yunnan, Qinghai, Chongqing, Zhejiang, Heilongjiang, Jilin, Inner Mongolia, Henan and Anhui. We then quantify the effect of socio-economic factors and their regional variations on the carbon emissions using panel data model. The results show that the development of secondary industry is the most significant variable in both the entire nation level and the regional level, while the effects of the other variables vary across regions. Among these factors, population density is the main motivator of the increasing carbon emissions per capita from transport sector for both the whole nation and the western region, whereas the consumption level per capita of residents and the development of tertiary industry are the primary drivers of per capita carbon emissions for the eastern and central region.


Introduction
Transportation is essential to national economic development, because it provides carriers for product circulation.With prosperous economic development, the transportation industry has developed rapidly, with vast quantities of fossil fuel consumption and increasing negative influence on the environment.Transportation has become the second highest carbon emission industry and attributed approximately 23% of total carbon emission in 2013 in the world; these proportions will increase continuously due to higher levels of energy consumption and be projected to reach to 3206 million tons of standard oil equivalents in 2035 [1].In rapidly developing China, transport sector has undergone dramatic development.For example, according to China's National Bureau of Statistics, China's freight turnover and passenger turnover increased by average annual rate of 19.7% and 6.8% over the past 16 years, respectively.The energy consumption and carbon emissions grew with the boom of transportation.The average annual growth of energy consumption in transport sector is 14.67%, which is more than the total energy consumption growth rate, 12.03%.Furthermore, the transport carbon emissions increased from 174.7 million metric tons in 2000 to 754.3 million metric tons in 2015.Based on the forecast of the International Energy Agency and the study of Auffhammer and Carson [2], China's transport sector will accounts for more than one-third of the world's transport carbon emissions in 2035.
China has been a pioneer on the path of reducing carbon emissions since the government announced that China will sharply reduce its carbon emissions before 2020 in the Intended Nationally Determined Contributions in 2015.China's government is turning this commitment into effective action.They presented plans which required that, compared to 2005, carbon intensity must decrease by 40%~45% in 2020 and 60%~65% in 2030 by effectively controlling industrial carbon emissions.As a matter of fact, decreasing the carbon emissions from transport sector is the requirement of China's sustainable development strategies, and it will accelerate the sustainable development of transport worldwide.In the literatures, it is commonly acknowledged that the carbon emissions from transport are likely to grow with socio-economic development.Wang and Liu [3] examines the features and driving factors of carbon emission from commuter traffic in Beijing from 2000 to 2012, and find the per capita disposable income, vehicle-use intensity, population and transport capacity effects are the main drivers that increase carbon emissions.Loo and Li [4] trace the historical evolution and spatial disparity of carbon emissions from passenger transport in China from 1949 to 2009, and the result shows that the income growth is the principal factor leading to the growth of passenger transport carbon emissions and the main factor contributed to carbon emission reduction is the lower emission intensity supported by policies, although the effect is weak.The impact of freight transportation on carbon emissions has also attracted attention.By exploring the impacts of factors on the carbon emissions from road freight transportation in China from 1985 to 2007, Li et al [5] found that the economic development is the primary driving factor of carbon emissions, whereas the ton-kilometer per value added of industry and the market concentration level contribute significantly to reduce carbon emissions.Effects of fossil fuel share, fossil fuel intensity, and road freight transport intensity are all found responsible for carbon emissions [6].Researchers have examined the dynamic changes in total factor carbon emissions performance of China's transport sector [7], and used the cointegration method to examine the long-run relationship between carbon emissions and affecting factors, including urbanization rate, energy intensity, carbon emission intensity and economic activity in transport sector [8].Understanding the main driving forces of carbon emissions of transportation is important since that the transportation is a major source of carbon emissions in China [9].However, konwledges on the driving forces of carbon emissions in China's transport sector are insufficient for different regions.Particularly, due to China's imbalanced development, different regions are facing different challenges, which may lead to different carbon emissions patterns [10].Therefore, it's necessary to conduct a study of driving factor at the region level so that some control policies considering local realities can be provided for policymakers.This research distinguishes itself from previous studies in the following four aspects.First, although some studies used time-series data to examine the factors of carbon emissions from transport sector in China [8], the data used are stale, and many data for the 20th century are missing.Second, the previous studies [7] did not divide China into different regions when analyzing the carbon emissions from transport sector, which may result in bias in the driving patterns of the carbon emissions.The reason is that China has a wide geographical area, a large population, and a complex economy, with significant differences between regions.It may lead to a bias conclusion if we don't take the regional inequality into account.[11].Third, many studies ignored the carbon emissions from transport sector generated by using electricity and instead considered it to be clean energy, but we must recognize that most electricity in China is produced by thermal power plants, which also emit carbon dioxide [12].Fourth, the energy carbon emission intensity used in previous articles refers to the amount of CO2 produced by complete combustion of a unit of fuel, which reflects energy efficiency.It is influenced by many factors and has gradually increased with the progress of automobile manufacturing technology.Therefore, the use of a fix value to calculate the carbon emission intensity of energy may lead to inaccurate results.To overcome these deficiencies, this study first calculates China's provincial transport carbon emissions generated by fossil fuel and the electricity consumed by the transport industry over the past fifteen years and then examines the main driving forces of the carbon emissions, as represented by transport carbon emissions per capita (PCT) and carbon emission intensity (CI) using panel data models.Moreover, the zoning effects of the driving pattern are explored by dividing the whole nation into the western (Ningxia, Chongqing, Guizhou, Shaanxi, Qinghai, Sichuan, Xinjiang, Gansu, Guangxi and Yunnan), central (Shanxi, Inner Mongolia, Jilin, Heilongjiang, Anhui, Jiangxi, Henan, Hubei and Hunan) and eastern (Beijing, Tianjin, Hebei, Liaoning, Shanghai, Jiangsu, Zhejiang, Fujian, Shandong, Guangdong and Hainan) regions to process the model.

Estimation of carbon emissions from transport sector
Accurately measuring carbon emissions is the basis for analyzing the characteristics of the carbon emission from transport sector.For a mobile emissions source, there are two main approaches to calculating the carbon emissions.One is "bottom-up" method, which is an agent-based model calculating carbon emissions based mainly on the vehicle kilometers traveled, and it is widely used in the road transport sector [13,14].Although the method based on vehicle kilometers traveled can distinguish carbon emissions sources from different motor types and help clarify the underlying reason for carbon emissions, it also shows considerable uncertainty, especially in road transportation, because vehicle types, vehicle mileage, fuel types and road conditions will affect fuel consumption in different ways.The other one is a top-down method that measures the carbon emissions according to the consumed fuel.It has a distinct advantage in calculating the carbon emissions from China's transport sector [15].Because the production and supply of fuel in China are a state monopoly, the official data can be collected completely and conveniently.In conclusion, we use top-down method, which is a unified standard method published by the Intergovernmental Panel on Climate Change (IPCC) Guidelines [16-18], to estimate the carbon emissions from China's transport sector.This method involves three parameters: energy type, amount of energy consumed and carbon emissions factor.The average low-order calorific value, average carbon content on an energy basis and carbon oxidation rates are multiplied to obtain the carbon emission factors of fossil fuel.The formula for calculating the carbon emissions from fossil fuel is as follows: (1) where CF (kg CO2) denotes the carbon emissions from fossil energy consumption; i denotes the different types of fossil fuel (including coal, coke, crude oil, gasoline, kerosene, diesel, fuel oil, liquefied gas and natural gas); FC (kg) represents the consumption; and F (kg CO2/kg) is the carbon emissions coefficient; ALC (KJ/kg) is the average low-order calorific value; C (t/TJ) is the carbon content; and R is the carbon oxidation rates.
Carbon emission from electricity is calculated using: where CE (kg CO2) represents the carbon emissions from electricity consumption; EC (kg) represents the consumption of electricity; and EF (kg CO2/kg)is the carbon emissions factors of electricity.
The total carbon emission from transport sector is calculated as follows: where CT (kg) represents the total carbon emissions from transport sector in a province.

Panel data analysis
As an analysis method, the Kaya identical equation [19] is widely applied to investigate the factors that influence carbon emissions from transport sector by explaining the varying quality of carbon emissions.It is usually combined with decomposition methods, such as the logarithmic mean Divisia index (LMDI) methods [10] and the Fisher index method [20], to investigate the influencing factors.However, the Kaya identical equation is not suitable for explaining the existing quantity of emissions and instead ignores the historical factors.The Stochastic Impacts by Regression on Population Affluence and Technology (STIRPAT) model [21], an extension of the Impacts by Regression on Population Affluence and Technology (IPAT) model [22], is also a popular model for analyzing transport carbon emissions [10,23] because it can adopt any variables influencing environment in the model.There are other methods for investigating the carbon emissions from transport sector, such as the dynamic Vector Auto Regression Model [9] and the Tobit regression model [24].However, the multicollinearity of variables must be considered before using these regression models.Choosing an appropriate model is important because this choice can affect the interpretation of results.Panel data analysis is a method of investigating a regression relation in the spatial and temporal dimensions with many advantages.First, panel data model can reflect the individual heterogeneity in both dimension of space and time, this advantage prevent the biased results caused by time series data or cross-section data analysis.Second, panel data can provide more reliable estimation of parameter because it has more information, a wider range of variability and weaker multicollinearity.Third, panel data is more suitable for researching the process of dynamic regulation for it can associate experience and behavior at different times and locations.Fourth, the unit root test of panel data can solve the problem that nonstandard gradual distribution caused by the unit root test of time series.Researchers in many fields thus favor using panel data model to analyze the carbon emissions [25][26][27].This study employs the panel data model to explore the factors that affect carbon emissions related to transport sector.It is difficult to judge the size of a population when choosing the model types of panel data.Hence, there are two perspectives on choosing a model type: one is based on the analytical goal.When the analytical goal is to estimate the parameter and the sample in the model is not very big, the fixed effects model is better.When the error components of the model are to be analyzed, we choose the random effects model to determine whether there is a relationship between some explanatory variables and the individual effect in the model.The other goal is to judge the precondition of the model.The fixed effects model assumes that the heterogeneity term and the independent variables are correlated.In contrast, the random effects model assumes that the heterogeneity term and the independent variables are not correlated.In this study, we choose the latter method.The procedures for setting up the panel data model in this method are as follows: first, conduct unit root tests to validate the stationarity of variables at the provincial level; second, perform cointegration tests to judge whether there is a long-term relationship between the variables; and finally, establish the panel data model.
The panel data models can be divided into three types: pooled, fixed effects and random effects regression models.The model type is confirmed by the F-test and Hausman test.The process of model selection is shown in Figure 2.Under a certain significance level, if F<F0.05(N-1,NT-N-k), then the pooled panel model is employed; otherwise, the Hausman test should be conducted to choose the random effect panel model or the fixed effect panel model.When the probability of the Hausman test is less than 10%, the fixed effects panel model should be selected; otherwise, we should choose the random effects regression model [28].After confirming the model type, Seemingly Unrelated Regression method is employed to set up the regression equations, which can eliminate the effect from the cross sections heteroscedasticity and the autocorrelation of time series [29].
The formula of the F-test is: where SSEr and SSEu represent the residual sum of squares of the pooled regression model and fixed effects regression model, respectively; k is the number of public parameters; and N is the constraint conditions.Under a certain significance level, if F<F0.05(N-1,NT-N-k), then it is better to choose the pooled regression model; otherwise, the fixed effects regression model should be chosen.

Data
Table 1 shows the variable of the models.PCT is an important index for precisely describing the carbon emissions.In addition, to investigate the energy efficiency of the transportation industry, CI is proposed as an index to describe carbon emission intensity.CI is an explicit value which means carbon emission in unit output value of transport sector.A smaller PCT or CI value indicates a greater environmental benefit from unit transportation activity.Thus, the PCT and CI indices are considered dependent variables for examining the driving patterns.Then, the proportion of secondary and tertiary industries added value to GDP (SGDP and TGDP), passenger turnover (PT), freight turnover (FT), road network density (RD), population density (PD), consumption level of residents per capita (PCC), motor vehicle population (VP) and energy consumption structure (ES) are selected as the explanatory variables in the panel data models.ES is the ratio of clean fuel consumption to the total energy consumption, which measures the rationality of energy use in transport sector [30,31]; electricity, liquefied gas (LPG) and natural gas (CNG) are classified as clean fuels in this study.CI and PCT are the carbon emissions indexes; SGDP and TGDP are the industry structure indexes; PT, FT and ES represent the development of transport industry; RD and PD stand for the road facility and population concentration levels respectively; and PCC and VP show the social development levels.PCT and CI are calculated by formulas (5) and ( 6) respectively.
where P (10,000 people) and GDPT (billion RMB) donate the population and the GDP of transport industry in certain province, respectively.

Description of data
For better grasping the geographic distribution features of CI and PCT in different provinces and regions, all provinces are identified with different colors based on the total change rate in CI and PCT from 2000 to 2015, with the values of year 2000 as reference.The deeper the color is, the higher the change rate is.The overall trend is that the CI decrease with the time since the China's government decision was announced at the UN climate summit, i.e. to decrease CI by 2020, expect in some provinces, like Zhejiang, Chongqing, Yunnan, Qinghai, Heilongjiang, Jilin, Inner Mongolia, Henan and Anhui.In addition, we find that most of these provinces are concentrated in central regions (Figure 3).Across China, the PCT is steadily increasing due to the high rising transport energy consumption rate and the comparatively low population growth rate, especially in Qinghai, Chongqing, Inner Mongolia, Henan, Anhui and Jilin (Figure 4).
The trends of the explanatory variables of the three regions and the whole nation are shown in Figure 5.The figure indicates that the country shares the same trend in all three regions.From 2000 to 2015, the values of VP, PCC, RD and PD increase monotonically, TGDP and FT increase with fluctuations, and SGDP rises and then decreases in 2011.This resulted from the Chinese government implementing an industrial structure adjustment policy since the 1990s, and the intensity of this structure adjustment has increased gradually, especially since 2000, with the environment receiving more attention from the public.PT shows the same changed in trend with SGDP, and this change is caused by the policy allowing citizens to drive on the highway for free during holidays, as implemented in August 2012.This policy has decreased the number of people who travel by public transport and increased the number of self-driving tourists.ES has the opposite trend as PT: it decreases first and then increases because of the increasingly widespread use of new-energy automobiles and projects that aim to balance regional energy sources, such as the west-to-east gas transmission project.

Data source
The carbon emission factor of electricity production and fuel consumption are affected by fuel quentity and consumption technology level, which is relatively stable in a shorter period of time (i.e., several years).Besides, the annual data of China's electricity production and fuel consumption carbon emissions factor is unavailable.For these reasons, the standard value of carbon emission factor is employed in this research.Thus, the carbon content of fossil fuel, carbon oxidation rates (Table 2) and electricity carbon emission factors (Table 3) are derived from the Guidance for Compiling Provincial Greenhouse Gas Emission Lists (Trial).Fossil fuel consumption data, electricity consumption data and average low-order calorific values are collected from Energy Statistical Yearbook of China.The data of population, GDP of the transport industry and explanatory variables can be accessed from the Statistical Yearbook of China.

Unit root test
To avoid heteroskedasticity and non-stationarity phenomena, a natural logarithm transformation is conducted for some variables, including PD, PCC, RD, VP, PT and FT, before implementing the panel data analysis.Then, two different models are built.Model 1 depicts the relationship between transport carbon emissions and human activity, so we choose PCT as the dependent variable and PD, PCC, RD, VP and TGDP as independent variables.Model 2 is used to describe the effect on the carbon emissions from transport sector from the development of transport industry.Hence, CI is chosen as the dependent variable, and TGDP, SGDP, PT, FT where i=1,…, N for each province in the panel and t=1,…, 16 refers to the time period from 2000 to 2015. 0 and  , denote the constant terms and white noise respectively.
The prerequisite for Pedroni cointegration is that all variables in the models must be integrated of the order one.The IPS unit root test is conducted to confirm this state, and the results are reported in Table 4.The results show that except for the ES of the entire nation, all variables accept the null hypothesis of non-stationarity at a less than 10% level of significance at the original series.However, for the first-order differences, all variables are stationary at the 1% significance level.Based on this finding, we conclude that all variables are integrated with first-order differences, other than the ES of the entire nation.

Pedroni cointegration test
The Pedroni cointegration test is used to estimate the long-run relationship of the independent variables because all variables are stationary for the first-order difference.Table 5 shows the results of the seven test methods; of them, three accept the null hypothesis of no cointegration, and four reject it.In addition, the significance levels are different for these variables.According to the panel ADF-statistics, we further find that the independent variables of Model 1 and Model 2 all reject the null hypothesis of no cointegration at the 1% and 5% significance levels.The results of the Pedroni cointegration test obviously prove that regardless of the regions or models considered, the explanatory variables maintain a long-run relationship with the explained variable during the study period.

Driving patterns and zoning effects based on panel models
Model 1 and Model 2 are employed to estimate the driving patterns for both the entire nation and the three regions, respectively.The estimated results are given in Table 6 to Table 8.The null hypothesis of the F-test is that all models are based on pooled effects.As shown in Table 6, all of the F-values are higher than the critical value of the F-test at the 5% significance level, thus rejecting the null hypothesis of the pooled effects method.In addition, as indicated in Table 7, which shows the results of Hausman test, all of the P-values are less than the critical value at the 5% and 10% significance level, except for the western region in Model 2. This result indicates that the random effect model should be used for the western region in Model 2 and that the fixed effect model should be employed for the other Models.
Table 8 shows the regression results of Model 1, for which the fixed effect regression model is selected, and the R2 values are all greater than 0.9, which indicates good fitting.Further, it can be observed that the zoning effect of the carbon emissions from transport sector is prominent in China, which indicates that the factors have different effects on the carbon emissions from transport sector in different regions.Among the variables of Model 1, PD, PCC and TGDP have positive effects on PCT in all regions, but VP exerts a negative effect.In addition, the impacts of RD on PCT vary from region to region: it plays a driving role in the eastern region, plays a opposite role in the both entire nation and the western region, and has no significant influence in the central region.As expected, PD exerts a significantly positive influence on PCT, especially in the western region, with a coefficient of 5.468.The PD has continuously increased over the past sixteen years in China, which has put substantial pressure on both transportation and the PCT.The study of Wang et al [32] showed that for an increase of one inhabitant, the number of day trips will increase 2.64 persontimes.The western region is a vast territory with a sparse population which led to low transport intensity.Thus, the increasing demand of living material transportation and long-distance travel transportation with the expanded population scale will result in higher levels of carbon emissions.Therefore, it is unsurprising that growth in PD will lead to a corresponding increase in PCT.
PCC is found to have a positive influence on PCT in all three regions.Since 2000, with the overall increase of the consumption level in China, there has been an increasing tendency for high-end consumption, which represents choosing a high-energy-consumption trip mode, such as traveling by airplane, as the primary long distance trip mode.In developed regions, such as the eastern region, or in road-or railway-network-sparse regions, such as the western region, airplane travel is a superior choice.This explanation can be confirmed indirectly by data from the Statistical Bulletin for the Development of the civil aviation industry in 2016, which indicate that the handling capacity of airport passengers is highest (551 million people) in the eastern region, followed by the western region (301 million people).
TGDP is also found to contribute a positive effect on PCT.A booming tertiary industry leads to more demand on business and tourist transportation, particularly in road and air passenger transport.In terms of energy efficiency, rail and water transport are more efficient than road transport and air transport [4], thereby generating further growth in PCT.Furthermore, the central region has intensive road network while also having a less intensive high-speed railway than the eastern region, so travelers prefer road passenger transportation, which has led to higher carbon emissions due to the development of tertiary industry in the central region.
VP shows an inhibitory effect on PCT.With the rapid process of urbanization, more polycentric cities have appeared in China [33][34][35].Such a smart growth mode with a higher degree of mixed landuse and an intensive road network would significantly shorten the commuting distance [36].Furthermore, the vehicle oil consumption has been improved as the development of the technology.According to China's standard document named Fuel Consumption Limits for Heavy-duty Commercial Vehicle, the average fuel consumption of heavy-duty commercial vehicle produce in 2020 should decrease 15% than the same kind vehicle produced in 2015.As a result, an increasing VP contributes to some inhibitors in relation to PCT.A more stringent vehicle emission control policy and the polycentric urban development mode have caused the eastern region to have a superior ability to confront increasing VP compared with those of the other regions.Hence, the coefficient of the central region is the highest among the three regions.
RD influences PCT differently in various regions, but for the whole nation, it has an inhibitory action on PCT.The development of transportation infrastructure positively affects the carbon emissions in mega-cities, which are mainly concentrated in eastern China, but it has negative effects in medium-small cities [37].Road construction plays a vital role in national economic as a kind of traffic infrastructure, and the economic significantly influences PCT [5].The unbalanced development of transportation infrastructure ultimately results in different PCT in different regions.
Table 8 also shows the regression results of Model 2. There are only four independent variables in Model 2 for the whole nation: TGDP, SGDP, PT and FT, ES is not adopted because it rejects the null hypothesis of non-stationarity at 1% at the original series.Our result is in line with the studies of Guo et al [10]and Fan and Lei [20], which indicated that the ES has a great impact on the carbon emissions from transport sector.We also find that it is an important indicator for the carbon emissions from transport sector in both the eastern and central regions, with statistical significance at the 1% level.The individual fixed effects regression model is used for entire nation and for the central and eastern regions, but the random effects model is used for western region with an R2 value of the regression results of only 0.170.Nevertheless, some of the variables still have remarkable effects on the explained variable.In addition, the zoning effect also obviously exists.For example, except for the SGDP always significantly increasing the value of CI, the other variables affect CI differently in various regions.12 Secondary industry development maintains a significant positive role in relation to CI across China.

13
The eastern region is in the late industrial stage [38], so the transportation mode is also developing a 14 more convenient and higher CI, with the transformation of the product from bulky and low value-

27
FT has small impact but is highly statistically significant on CI in the eastern region.In terms of

37
Due to the geographic heterogeneity, ES inhibits the CI of transport sector in the eastern and 38 central regions, especially in the latter, with a coefficient of -2.766, but it shows statistical significance 39 in the western region.The western region has a high proportion of clean energy sources, but it also 40 has a complex natural geographical condition, which results in inefficient energy use [40].

Figure 2 .
Figure 2. Flowchart of model decision

Figure 3 .Figure 4 .Figure 5 .
Figure 3.The distribution of the CI change rate from 2000 to 2015

15
added to light-weight, deep processing and high value-added.In contrast, the other two region are 16 in the early-middle industrial stage, so the transportation mode is quite different from that of the 17 eastern region.Hence, the coefficient of SGDP of the eastern region is much higher than those of the 18 other regions.19Asshown in the results, PT is found to have inhibitory action on CI in the central and eastern 20 regions, but a galvanizing impact in the western region and the entire nation.The carbon emissions 21 of China's passenger transportation is related to the level of regional economic development and 22 natural geographic conditions [39].Compared with the western region, the better developed central 23 and eastern regions are more likely to get the support of advanced carbon-reduction technologies 24 and policies, which are significant for decreasing CI.The complicated natural geography in the west 25 also indirectly inhibits the implementation of carbon-reduction policies and extension of advanced 26 technologies.

28
freight transport structure, vessel transportation is the main transport form in the eastern region, but 29 it should not be ignored that there has been a sharp increas in freight volume in both airfreight and 30 road freight, which have lower energy intensities.Conversely, FT is an inhibitor of CI in the entire 31 nation and the central and western region.With the implementation of policies, such as the western 32 development campaign and the plan of rejuvenating old industrial bases in Northeast China, that 33 aim to improve the level of industrial agglomeration, the freight demand of unit industrial added 34 value has declined steadily and the empty-loading rate has also declined.Moreover, long-distance 35 freight transport has increasingly relied on more economic transport modes, such as water carriage, 36 rail transport and pipeline transport.

Table 2 .
Parameters used to calculate the carbon emissions of fossil fuels

Table 4 .
Results of the IPS unit root test 1The unit root tests of each variable are carried out with individual intercept.2***, ** and ** denote significance at 1%, 5% and 10%, respectively.

Table 6 .
F-test result for Model 1 and Model 2 1 There is only four independent variables in Model 2 of the whole nation: TGDP, SGDP, PT and FT.

Table 7 .
Hausman test result for Model 1 and Model 2 1 There is only four independent variables in Model 2 of the whole nation: TGDP, SGDP, PT and FT.

Table 8 .
Estimation parameters of the panel data models 1

Table 8 ,
we find that the impact of tertiary industries on the CI is positive only for the 5 eastern region; it is negactive for the whole nation and is statistically non-significant for the central 6 and western regions.Tertiary industry, which provides services for production and consumption, 7 exists on a larger scale in the eastern region of China than in other regions.The flourishing tourism, 8 catering, culture and sport industries strongly stimulate passenger transportation demand, drive the 9 development of transport industry and generate more carbon emissions.However, compared with 10 the impact of secondary industry development on transport CI, tertiary industry development has 11 much less influence in the eastern region and is even an inhibiting factor for the entire nation.