Our aim in this study is to investigate the role of school enrollment at various levels in the formation of the EKC in China. To get estimates of the variables, we employ the ARDL model. ARDL has various advantages over other techniques, but the foremost advantage of this technique is that it performs well even if the variables are a mixture of I(0) and I(1). However, we can’t include the variables of I(2) in the analysis. Therefore, to confirm that none of the included variables is I(2), we employ two unit root tests, one without a structural break and another with a structural break. Table 2 provides the findings of both the unit root tests. From the results, we can deduce that most of the variables are I(1) and some are I(0) by using either of the tests; however, none of the variables in the analysis is I(2). These findings give us a positive signal that we can apply the ARDL model. Another important thing that needs to be decided before starting our formal discussion is selecting the appropriate lag. Our data is annual, and we apply a maximum of two lags and to choose the correct number of lags, we use Akaike Information Criterion (AIC).
Table 2
| Unit root without break test | Unit root with break test |
| I(0) | I(1) | Decision | I(0) | Break date | I(1) | Break date | Decision |
CO2 | -0.898 | -4.689*** | I(1) | -5.025*** | 2002 | | | I(0) |
PE | -2.654* | | I(0) | -4.452** | 2001 | | | I(0) |
SE | -0.325 | -4.965*** | I(1) | -4.332 | 2006 | | | I(0) |
TE | -0.356 | -2.887* | I(1) | -0.897 | 2012 | -6.231 | 2014 | I(1) |
AE | -1.789 | -7.654 | I(1) | -3.323 | 2018 | -6.689 | 2002 | I(1) |
GDP | -0.875 | -2.635* | I(1) | -3.201 | 2002 | -4.356* | 2007 | I(1) |
EC | -0.456 | -3.678* | I(1) | -4.821** | 2002 | | | I(0) |
Internet | -0.623 | -2.658* | I(1) | 5.398*** | 2006 | | | I(0) |
GE | -2.671* | | I(0) | -5.164 | 2015 | | | I(0) |
Note: ***p<0.01; **p<0.05; and *p<0.10 |
We have used four different proxies for education for empirical analysis: enrollment at primary, secondary, and tertiary levels. Moreover, we also include the average total years of schooling to get the accumulative impact of education on the CO2 emissions. First, we discuss the short-run estimates, and after that, we discuss the ones in the long run.
Table 3 illustrates the short-run and long-run results of the ARDL model. In short run, the estimated coefficient of D(PE) is positively significant, and the estimate of D(PE2) is negatively significant. Similarly, the estimate attached to D(SE) is positive, and the estimate attached to D(SE2) is negative and significant. Then the estimate attached to D(TE) is positively significant, whereas the estimate attached to D(TE2) is negatively significant. Lastly, the estimates attached D(AE) is positive at first lag, and the estimate attached to D(AE2) is negatively significant at first lag. These findings confer that increased enrollment at all levels does increase the CO2 emissions; however, at the later stage, increased enrollment at all levels decreases the CO2 emissions. The estimated coefficients of D(GDP) are positively significant in all models, and the estimates attached to D(GDP2) are negatively significant, confirming the presence of EKC in the short run. Given the importance of long-run results, we now pay attention to the long-run estimates provided in Table 3. The validity of the long-run results depends on the confirmation of cointegration between them. To that end, we rely on two tests of cointegration, i.e., F-test and ECMt−1 and the results of both the tests are provided in Table 3, which confirm that cointegration exists among the CO2, SE, PE, TE, AE, GDP, EC, Internet, and GE.
The long-run estimates attached to PE, SE, TE, and AE are positively significant, or more precisely, we can say that a 1% rise in primary, secondary, tertiary, and aggregate enrollment increases the CO2 emissions by 1.570%, 1.218%, 0.008%, and 0.580%. Conversely, the estimates attached to PE2, SE2, TE2, and AE2 are negatively significant or in terms of elasticity, we can say that a 1% rise in PE2, SE2, TE2, and AE2 causes the CO2 emissions to decline by 2.058%, 0.171%, 0.023%, and 0.580%. Generally, our findings imply that an increase in educational activities at all levels increases the CO2 emissions at the early stages; however, later on education help reduces the CO2 emissions. In other words, the relationship between CO2 emissions and all levels of education follows an inverted U-shaped path, implying that CO2 emissions increase in the early parts of increased educational activities and decline at the later stages. As the demand for education in the country increases, on one side, more educational infrastructure such as schools, colleges, universities, and hostels are required; on the other side, demand for transportation facilities, laundry, dry cleaning, and saloon services will also increase (Becken et al. 2001, 2003; Gossling, 2002). As a result, the energy demand rises, which is a primary driver of CO2 emissions. However, the positive effects of education on the environment may come later once the education sector starts producing more trained, skilled, capable and efficient human resources that can replace the more energy-intensive inputs in the production process and ultimately reduce CO2 emissions.
Social and economic activities performed by humans cause emissions, and education is an important source that can positively alter human behavior (Jian et al. 2021). Moreover, education help raise the technical skills and capabilities that will improve human efficiency in all walks of life and contribute to economic development (Usman et al. 2021). Experts of economics strongly agree that human capital is essential for the economic growth of a country in the long run. They also agree that human capital is a by-product of formal education, trained and experienced labor, research and development, which are fundamental parts of the inputs used in the production function (Barro, 1991). Most developing economies have replaced their labor-intensive production techniques with human-capital intensive ones and achieved the economic goals with a more clean environment. However, the evidence suggests that education may pollute the environment in developing and emerging economies due to energy-intensive infrastructure and low economic development (Mahalik, 2021). The energy consumed by educational activities at various levels may differ at different stages of economic development (Inglesi-Lotz and Morales, 2017), which may form an inverted U-shaped relationship between education and CO2 emissions.
As far as the long-run relationship between economic development and environmental quality is concerned, we can see that the estimates attached to GDP are positive but insignificant in most models. However, the estimates attached to GDP2 are negatively significant in all models confirming the presence of an inverted U-shaped relationship between economic development and CO2 emissions in China. Such an inverted U-shape relationship is known as the EKC of Grossman and Kreuger (1995), which implies that the early part of economic growth pollutes the environment and later improves it. This result also implies that China is heading towards sustainable development, i.e., achieving economic growth without polluting the environment further.
The estimate of the control variable of EC suggests that increased energy consumption causes the CO2 emissions to rise; however, the estimates of Internet and GE are significant and negative, implying that increased internet subscriptions and government expenditures cause the CO2 emissions to decline in China.
Finally, to confirm the efficiency of our estimates, some diagnostic tests are also outlined in Table 3. Firstly, Langrage Multiplier (LM) test confirms that our residuals are free from first-order serial correlation. Secondly, the Ramsey RESET test confirms that no misspecification is found in our model. Thirdly, Breusch Pagan (BP) test approves that the variance of the error terms is homoscedastic. Finally, the CUSUM and CUSUMSQ confirm the parametric stability of the models where ‘U’ represents the stable parameters and ‘US’ represents the unstable parameters.
Table 3
ARDL estimates of CO2 emissions
| Primary education | | Secondary education | | Tertiary education | | Average education | |
| Coefficient | t-Stat | Coefficient | t-Stat | Coefficient | t-Stat | Coefficient | t-Stat |
Short-run | | | | | | | | |
D(PE) | 2.618*** | 2.763 | | | | | | |
D(PE(-1)) | -1.561 | 1.482 | | | | | | |
D(PE2) | -3.229*** | 2.782 | | | | | | |
D(PE2(-1)) | 2.021 | 1.432 | | | | | | |
D(SE) | | | 3.371 | 0.435 | | | | |
D(SE(-1)) | | | 9.660* | 1.832 | | | | |
D(SE2) | | | -0.515 | 0.540 | | | | |
D(SE2(-1)) | | -1.210* | 1.835 | | | | |
D(TE) | | | | | 0.010* | 1.772 | | |
D(TE(-1)) | | | | | 0.003 | 1.043 | | |
D(TE2) | | | | | -0.039* | 1.885 | | |
D(AE) | | | | | | | 0.429 | 0.783 |
D(AE(-1)) | | | | | | | 2.449*** | 3.070 |
D(AE2) | | | | | | | 0.032 | 0.853 |
D(AE2(-1)) | | | | | | -0.177*** | 3.019 |
D(GDP) | 1.153** | 2.133 | 1.894*** | 3.406 | 1.045** | 1.989 | 1.825*** | 3.032 |
D(GDP(-1)) | 0.479 | 0.822 | | | 0.567 | 1.172 | 0.866 | 1.578 |
D(GDP2) | 1.376* | 1.760 | 2.169** | 2.879*** | 2.310 | 3.512*** | 1.289** | 1.981 |
D(EC) | 0.859*** | 2.783 | 1.255*** | 5.846 | 1.105*** | 5.135 | 1.527*** | 5.892 |
D(EC(-1)) | 0.447 | 1.444 | | | | | -0.429 | 1.275 |
D(Internet) | -0.002 | 0.510 | -0.003* | 1.828 | -0.005* | 1.843 | -0.011** | 2.150 |
D(Internet(-1)) | 0.008*** | 2.756 | | | -0.009** | 2.414 | -0.019* | 1.752 |
D(GE) | 0.008 | 0.466 | -0.013 | 1.070 | -0.004 | 0.426 | 0.233 | 0.123 |
D(GE(-1)) | 0.036* | 1.699 | | | | | | |
Long-run | | | | | | | | |
PE | 1.570** | 2.182 | | | | | | |
PE2 | -2.058** | 2.164 | | | | | | |
SE | | | 1.218* | 1.755 | | | | |
SE2 | | | -0.171** | 1.986 | | | | |
TE | | | | | -0.008*** | 2.983 | | |
TE2 | | | | | 0.023* | 1.921 | | |
AE | | | | | | | 0.580*** | 3.436 |
AE2 | | | | | | | -0.039*** | 3.266 |
GDP | 0.270 | 0.196 | 0.353 | 0.318 | 0.002 | 0.022 | 0.386** | 2.330 |
GDP2 | -0.879** | 2.065 | -0.456* | 1.867 | -0.231* | 1.661 | -0.776** | 2.101 |
EC | 1.194*** | 5.733 | 1.431*** | 3.599 | 1.174*** | 9.466 | 1.568*** | 4.218 |
Internet | 0.001 | 0.118 | -0.002* | 1.835 | -0.001* | 1.880 | -0.002* | 1.952 |
GE | -0.002 | 0.123 | -0.019*** | 2.823 | -0.017*** | 3.181 | -0.012* | 1.840 |
C | -5.114 | 2.071 | 3.334*** | 2.301 | 7.299 | 2.245 | 5.401*** | 7.866 |
Diagnostics | | | | | | | | |
F-test | 13.12 | | 12.10 | | 13.12 | | 9.789 | |
ECM(-1) | -0.471*** | 6.729 | -0.592*** | 9.563 | -0.652*** | 9.521 | -0.578*** | 7.307 |
LM | 1.298 | | 1.023 | | 0.398 | | 1.785 | |
BP | 0.325 | | 0.875 | | 0.795 | | 1.035 | |
RESET | 0.689 | | 1.487 | | 0.980 | | 2.033 | |
CUSUM | S | | S | | S | | S | |
CUSUM-sq | S | | S | | S | | S | |
Note: ***p<0.01; **p<0.05; and *p<0.10 |