Understanding driving patterns of carbon emissions from the transport sector in China: evidence from an analysis of panel models

Abstract

China’s transport industry has made rapid progress, which has led to a great amount of carbon emissions. However, it is still unclear how carbon emissions from the transport sector are punctuated by shifts in underlying factors. This paper aims to examine the process of China’s carbon emissions from the transport sector as well as its major driving forces at the provincial level during the period of 2000 to 2015. We first estimate the carbon emissions from the 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 country, especially in the provinces of Chongqing and Inner Mongolia. However, the carbon emission intensity is decreasing in most provinces, except in Yunnan, Qinghai, Chongqing, Zhejiang, Heilongjiang, Jilin, Inner Mongolia, Henan and Anhui. We then quantify the effect of socioeconomic factors and their regional variations on carbon emissions using a panel model. The results show that the development of secondary industry is the most significant variable for carbon intensity at both the national and regional levels, while the effects of the other variables vary across regions. Among these factors, population density is the main factor of the increasing carbon emissions per capita from the transport sector for both the whole country 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 regions.

Graphical abstract

Appendix

Panel unit root tests

Using nonstationary variables to establish regression will lead to spurious regression. Therefore, it is significant to assess the variables with unit root tests (Wang et al. 2015). There are several unit root test methods used for panel data, including the Levin-Lin-Chu test (LLC) (Levin et al. 2002), Im-Pesaran-Shin test (IPS), ADF Fisher test (Dickey and Fuller 1979) and Fisher-PP test (Fisher 1945). The first three tests are commonly used in research (Gao and Zhu 2016; Xu and Zhang 2016; Zhang and Zhao 2014). The LLC test assumes that the variables have identical unit roots, so the autoregressive coefficient is the same across the cross sections, whereas the IPS test loosens the assumption of LLC test and allows variance across regions under the alternative hypothesis (Shuai et al. 2017). Therefore, the IPS test is used to examine the stationarity of the variables in this study. If the probabilities for the IPS test are less than 10%, we can reject the null hypothesis and consider the variables to be stationary.

Panel cointegration tests

After confirming that the variables are integrated of order one, the next step is to employ panel cointegration to identity whether a long-term relationship exists between the variables. The Pedroni test (Pedroni 2004), Kao test (Kao 1999) and Johansen (Johansen 1988) test are generally used in panel cointegration analyses. The Pedroni test constructs seven statistics to verify the cointegration relation among panel variables based on the regression residual of the cointegration equation, so it is applied in this study. Among these seven statistics, the group rho-statistic, group PP-statistic and group ADF statistic pool the regression residuals using the between-dimension approach. Four others, i.e., the panel v-statistic, panel r-statistic, panel PP-statistic and panel ADF statistic, pool the coefficients of autoregressive across different numbers by the within-dimension method (Ou et al. 2013). In addition, the panel ADF statistic is found to be more precise than the other six statistics (Örsal 2010). Thus, the panel ADF statistic is used to determine the goodness of fit. When the panel ADF statistic is less than 10%, it can be determined that there exists a long-run relationship between variables.