The Impact of Human Activities on Carbon Dioxide Emission in the Asian Countries from a Spatial Econometric Perspective
Tubagus Ajie Rahmansyah, S2139235
Master Thesis Economics, Rijksuniversiteit Groningen Supervisor: Prof. dr. J. Paul Elhorst
August 2012
Abstract:
This paper provides a spatial-econometric analysis of the role of multiple factors in determining environmental degradation in the Asian countries. The idea is coming from the IPAT equation constructed by Ehrlich and Holdren (1971) and also Environmental Kuznet Curve (EKC) theory by Kuznets (1995). The IPAT equation describes the multiplicative contribution of Population (P), Affluence or economic activities (A) and Technology (T) to human impact (I) on the environment, while the EKC theory explains a development-environment relationship. By applying a spatial panel data estimation method for 35 Asian countries from period 1995 to 2008, this paper finds that ignoring spatial interaction effects will lead to biased estimation result. This paper also finds evidences of the existence of an inverted U-shaped EKC. The last finding from this paper is that all explanatory variables are statistically significance influence the amount of CO2 emission in the atmosphere. Moreover, economic activities and technology become the most important factors in determining CO2 emission since they produce direct and indirect effects.
1 1. Introduction
One of the main environmental issues and major topics during international conferences is CO2 emission and the global warming problem. This issue turns to attract worldwide attention due to its tremendous negative impacts toward the environment and also creates potential hazard toward humankind directly or indirectly. Therefore, becoming the global concerns, it requires comprehensive and integrated actions from countries worldwide in tackling this problem.
The current level of CO2 emission in the atmosphere has increased the earth's temperature for its trapping heat and light from the sun, which is sent back to earth. Based on the report from Intergovernmental Panel on Climate Change (IPCC, 2001), global air temperature has risen 0.6 degrees Celsius (1 degree Fahrenheit) since1861. In addition, The World Bank (2012) pointed out that from year 2002 to 2008, the world’s CO2 emission increased from 4.1 to 4.8 metric tons per capita.
There are three main factors that are the subject of study in the field of economics related to the increase of CO2 emission. These factors are economic activities, population, and technology. Economists use these three factors to investigate the role of human impact towards the environmental condition.
2 Figure 1: The Environmental Kuznets Curve: a development-environment relationship, source Panayotou (1993)
The second factor, population, escalates CO2 emission in the atmosphere through human activities such as respiration, consumption, the loss of CO2 sequestration resources such as forest areas, which are cleared for settlements; and also burning of fossil fuels. The last factor, technology, suppresses CO2 emission through efficiency on the production and consumption process.
These three main factors are generally analyzed for their role in the CO2 emission based on the IPAT equation developed by Ehrlich and Holdren (1971). The form of the equation is as follows:
I = P A T [1]
Here I is Human Impact on the Environment, P is Population, A is Affluence or economic activities per person, and T is Technology. Dietz and Rosa (1997) slightly modify the IPAT equation and applied the reformulation to the anthropogenic sources of CO2 emissions. The modification made by them is known as the Stochastic IPAT model.
Many economic studies related to CO2 emission using the stochastic IPAT model show significant effects of the economic activities (in terms of GDP per capita), population and the technology to the increase of CO2 emission. However, as mentioned by Videras (2012), most of these studies estimate this model using linear regression and assume that the observations are spatially independent. This assumption is incorrect since the data itself is already spatial
3 in nature. Therefore by ignoring spatial interaction effects, it will lead to biased estimation results.
To provide better insight about correlation between the economic activities, population and technology to the CO2 emission, spatial panel estimation is required. This estimation method will not only tackle omitted variable bias problem (ignoring spatial interaction effect), it will also provide more informative results that include more variation and less collinearity among variables as explained by Elhorst (2010a).
With regard to the explanation above, this paper is aimed to study the human impact relation to the environmental condition especially in terms of economic activities, population and technology toward the CO2 emission in Asian countries using spatial econometric analysis.
There are two main issues to be analyzed further in this paper. These include: (i) do interaction effects have significant influence to the relationship of economic activities, population and technology with the CO2 emission? and (ii) is there any evidence of an inverted U-shaped EKC?
The structure of this paper is as follows: Section 2 gives existing literature related to human impact relation to the environmental condition, section 3 gives the method that is used to analyze the relation, section 4 explains the data that is used, section 5 explains the result of the analysis, and section 6 gives the general conclusion of the analysis.
2. Literature Review
4 producing cannot compensate the growth of population. This point of view makes his perspective as the pessimistic one. He believed that scarcity and limitation will make population growth as the cause of starvation, poverty, dieses, environmental problems, etc.
The Neoclassical theory of Solow and Swan (1956) puts population growth in the same position as the classic theory does. Population growth becomes the burden for economic development. The idea that leads them to this assumption is related to the relationship among the factor of production and it’s characteristic. Based on Neoclassical theory, economy can be accelerated through the use of capital and labour. On the other hand, in the long run, capital accumulation faces the law of diminishing return. Therefore, although the use of capital and labour increase the speed of economic growth will decrease. If economic growth cannot compete with population growth, then the same problem that face by the classic theory will happen.
As the growth theory evolves, position of population growth to the economic growth switch over into a strengthening factor. Romer (1996) explains clearly in his book “Advanced Macroeconomics” about this assumption on the endogenous growth theory. This growth theory, called the new growth theory, puts technological progress and human capital as parts of production factors. It implies that to increase and speed up output production, an economy can ease the use of capital (natural resource) by using high technology and high quality and quantity of human resource on production processes. Thus, the speed of economic growth can compete with the speed of population growth. Implicitly, this also explains that through high technology and high quantity and quality human resources, problems related to natural resources (environmental problem on CO2 emission) related to the increase of economic growth can be attenuated.
5 One of the earliest studies using EKC hypothesis is made by Grossman and Krueger (1994). They try to discover a relationship between economic activity and environmental quality for a variety of pollutants (Candia, 2003). This study clarifies that as an economy grows pollutants increase until a certain point then decreases as economy growing further. Graphically, the result from their study follows an inverse-U-shaped.
A recent study by Dhanda, Adrangi, and Chatrath (2005) about the linkage between GDP and Emission, analyses the EKC in more general perspective. Based on their study, economic activities (GDP) contributes to the increase of CO2 emission without concerning the difference of income level. Another finding from their study is that nations at the lowest stage of economic development have the highest rate of emission. Then, as their level of economic development increase, the rate of CO2 emission is stable then reaches the lowest rate as they achieve their highest level of economic development. These findings also give positive support to the EKC hypothesis as Canadia (2003) does.
To describe the relationship between population and CO2 emission, Onozaki (2008) conducted a study about the impact of population growth to the CO2 emission. Based on Onozaki, population gives great impact to the increase CO2 emission by direct and indirect human activities. In addition, population growth especially in the developing countries is a critical factor for manipulation of global CO2 increase.
To get better insight into the performance of human impact to the environmental condition, some studies have been conducted by including all of the main factors that derive the CO2 emission. The first study was done by Dietz and Rosa (1997). They analyze human impact based on an equation that was constructed by Ehrlich and Holdren (1974) which is known as the IPAT equation (see equation [1]).
6 is applicable to a wide variety of impacts (including GHG emissions). Then they did an assessment of the model’s overall fit to an appropriate data base by reformulating the model in stochastic version as follow:
Ii = aPib Aic Tid ei [2]
Here I, P, A, and T stay the same but they use the subscript i to denote that the quantities were across observation units. As the data set of IPAT equation in form of panel data become available and demand of information on analysis become high, a study on IPAT equation using panel data analysis turn out to be important. Shi (2001) use equation [2] as the base of the model to analyze the relationship between human impact and global carbon dioxide emission. He modified the equation [2] into:
ln Iit = a + b1(lnPit) + b2(lnAit) + b3(lnTit) + eit [3]
In equation [3] I, P, A, and T are defined similarly as in equation [1], but instead of using cross sectional data analysis as it was used by Dietz and Rosa (1997) here he used panel data analysis. One of the aims of using this analytic method was to address the issue of whether the impact of population growth on emission could vary across countries with different income levels of economic development. Then, to prove the existence of EKC on the data set, he used another specification with polynomial term of affluence in the model. The model transforms into:
ln Iit = a + b1(lnPit) + b2(lnAit) + b3(lnAit)2 + b4(lnTit) + eit [4]
This transformation is in line with a publication by Grossman and Krueger (1994). This publication has become the groundwork for most of the research regarding to EKC analysis (Candia, 2003).
7 interaction effects as policy recommendation. However, he used cross sectional data analysis so that analyzing the impact of the change of the data as the time changes cannot be done.
Although some studies related to the human impact on the CO2 emission have been done, but most of them are constructed based on solely impact either relationship between population and emission or economic activities and emission. Even if there are some studies using IPAT equation as Videras (2012) and Shi (2001) did, but they use cross sectional data analysis or without spatial interaction effects. Moreover, as believed by Videras (2012), we cannot depend on global model to describe the relationship on the IPAT equation. Therefore, a study using IPAT model using spatial panel data analysis become necessary to give better perspective on the human impact on the environment especially population, economic activities, and technology to the increase of CO2 emission in Asian countries.
3. Methodology
To analyze the IPAT relationship and the existence of EKC, this paper will use equation [4] as the base model and then transform it by including spatial interaction effects in the model. This change leads us to spatial panel model estimation. There are three important parts of this estimation: model specification, construction of W Matrix, and estimation procedure.
3.1. Model Specification
8 [5] 𝑢𝑖𝑡 = 𝜆 𝑤𝑖𝑗𝑢𝑖𝑡+ 𝜀𝑖𝑡 𝑁 𝑗=1 = + + =1 + + + =1
where i is an index for the cross-sectional dimension (i=1,2,...,N) and t is an index for the time dimension (t=1,2,...,T). y is the dependent variable, x represents the independent variables with associated parameter β, u is the spatially autocorrelated error term, and ε is the error term with zero mean and variance σ2 . ρ is the spatial autoregressive coefficient vector for spatially lagged dependent variable, θj is a spatial autoregressive coefficient for one of
spatially lagged independent variables, and λ is the spatial autocorrelation coefficient in the error term. W is an N x N nonnegative matrix describing the arrangement of the unit in the sample and wij is an element of a spatial weights
matrix W. μ capture spatial specific effects and η capture time-period specific effects. Both, μ and η, are optional.
However, as shown by Manski (1993), including all types of the interaction effects in the model would cause identification problems in the interaction parameters. Therefore he suggests that one of the interaction effects should be excluded. LeSage and Pace (2009) recommend a solution for this problem by including all interaction effects except the spatially autocorrelated error term. By doing this, estimation will only suffer a loss of efficiency. On the contrary, ignoring spatial dependence in the dependent variable and/or in the independent variables leads to omitted variable bias problem (Greene, 2005).
There are three kinds of models that spatial econometricians mainly used in their studies. These models are spatial lag model, spatial error model, and spatial Durbin model. The two former models employ one type of interaction effect while the latter model uses two types of interaction effects.
9 𝑦𝑖𝑡 = 𝜌 𝑤𝑖𝑗𝑦𝑗𝑡+ 𝑥𝑖𝑡𝛽 + 𝜇𝑖 + 𝜂𝑡+ 𝜀𝑖𝑡 𝑁 𝑗=1 𝑢𝑖𝑡 = 𝜆 𝑤𝑖𝑗𝑢𝑖𝑡 + 𝜀𝑖𝑡 𝑁 𝑗=1 𝑦𝑖𝑡 = 𝜌 𝑤𝑖𝑗𝑦𝑗𝑡+ 𝑥𝑖𝑡𝛽 + Ɵ𝑗𝑤𝑖𝑗 𝑁 𝑗=1 𝑥𝑖𝑗𝑡+ 𝜇𝑖+ 𝜂𝑡+ 𝜀𝑖𝑡 𝑁 𝑗=1 [6]
where variable wijyjt reflectsthe interaction effect among dependent variables.
Spatial error model includes the spatial autoregressive process in the error term of the model. This model describes correlated effect, where similar unobserved environmental characteristics result in similar behaviour. The spatial error model takes the form:
= + + +
[7]
where variable wijuit describes the interaction effect among error term.
The last model, the spatial Durbin model, includes the spatial lagged dependent variable and the spatial lagged independent variables in the model. Therefore this model not only assumes that dependent variables from neighboring units is interdependent but also posits exogenous interaction effects, where the dependent variable of a spatial unit depends on the explanatory variables of its neighboring units. The spatial Durbin model takes the form:
[8]
where variable wijxijt represents the exogenous interaction effects.
11 3.2. Construction of W Matrix
Anselin (2006) defines the "W" matrix as a weighted average of neighboring values, where the neighbors are specified through the use of a so-called spatial weights matrix. Leenders (2002), as explained by Elhorst (2010b), mentions that choosing the right specification of weights matrix is important since the value and the significance level of the interaction parameters depend on the specification of the W matrix.
Elhorst (2010b) pointed out that there are 4 (four) spatial weights matrices that are often used in empirical research in spatial econometric analysis. The first weighted matrix is p-order binary contiguity matrices. The second is an inverse distance matrix. The third is a q-nearest neighbor matrix. The last weighted matrix is a block diagonal matrix where each block represents a group of spatial units that interact with each other but not with observations in other groups. In this paper, the weighted matrix is based on first-order binary contiguity matrices.
The choice of using first-order binary contiguity matrices is based on two reasons. The first reason is related to closeness of a country to its’ neighbour (share a common border). This condition might play a large role in stimulating cross-country mimicking behavior. Moreover, as our dependence variable is CO2 emission (pollutant gas), It is obvious that CO2 emission from one country might give direct effect to its’ neighbouring countries related to spill over effect.
12 Figure 3: Map of Asia Continent, source World Atlas
3.3. Estimation Procedures
To do estimation, first we should conduct testing procedures to choose the best spatial econometric model. Then we choose the right panel model estimation. This paper will adopt testing procedures suggested by Elhorst (2010b) which is known as the specific-to-general approach to choose the best spatial model specification.
13 However if the result of the (robust) LM test inclined neither spatial lag nor spatial error model, then we should conduct significance tests on ρ (spatial lag model) and λ (spatial error model) using spatial lag and spatial error model estimation respectively. If both, ρ and λ, are significant then we should conduct the exact procedure as if the (robust) LM test inclined to spatial lag, spatial error, or spatial lag and spatial error model. Contrary, if ρ and λ are not significant then we should conduct an estimation on OLS model with (selection of) WX variables. If we cannot reject null hypothesis that there are no exogenous interaction effect (θ = 0) then OLS model estimation become our choice, but if we reject the null hypothesis (θ = 0) then we should do another test using spatial Durbin model estimation. If the result from this estimation suggest that we cannot reject hull hypothesis (Ho:ρ=0) model with (selection of) WX variables suffices. Contrary if we reject the null hypothesis (Ho:ρ=0) than spatial Durbin model is the best choice. All these testing procedure are illustrated in Figure 4.
After we find the appropriate spatial model then we estimate the model using fixed effect and random effect estimation. To choose which one is more appropriate, we conduct a Hausman test.
Finally after we choose the appropriate spatial model and estimation method, we can calculate direct and indirect effects of the IPAT equation. Using this calculation, we analyze the relationship between all of our explanatory variables and explained variable. This analysis will lead us to conclusions whether analysis on IPAT equation using spatial econometric technic is the best option, what factors that will increase or decrease CO2 emission in the atmosphere, and is there any strong evidence of an inverted U-shaped EKC.
14 Figure 4: Flow chart of testing procedure
OLS Estimation
LM Test
Spatial Lag & Spatial Error Model
Spatial Lag or Spatial Error Model
Neither Spatial Lag or Spatial Error Model
- Spatial Lag Model - Spatial Error Model - Spatial Lag and Spatial
Error Model
Spatial Durbin Model Estimation Spatial Lag Estimation
Spatial Error Estimation
ρ or λ
Insignificant Significant ρ or λ
OLS Estimation with (selextion of) WX
Variables
Ho: θ=0, Not Rejected
OLS Model
Model with (selection of) WX Variables
Ho: ρ =0, Not Rejected
Ho: θ=0, Rejected Spatial Durbin Estimation
Ho: ρ =0, Rejected
Ho: ρ =0 and Ho:θ+ρβ=0 Rejected
LR and/or Wald Test (robust) LM Test
Ho: θ =0 Not Rejected
Spatial Lag Model
if (robust) LM Test Spatial Lag
Spatial Error Model
if (robust) LM Test Spatial error Ho: θ+ρβ =0 Not Rejected
15 4. Data
The variables in this study include data on CO2 emissions (I), Population (P), Economics Activities (A), and Technology (T). Table 1 provides an overview of the variables and their definitions and indicators, as well as the sources of the data.
Table 1: Overview of the variables included in the study and their definitions, indicators, and the source of the data
No Variable Definition Indicator Source
1 CO2
Emission (I)
Chemical compound that consist of 2 atoms, carbon and oxygen that is emitted from human activities such as production process, breathing, etc or from natural activities such as plant assimilation, burnt tress, etc.
CO2 emissions (metric tons per capita)
World Bank WDI
2 Population (P)
Number of people who live in the same area or region in certain time. Total Population World Bank WDI 3 Economics Activities (A)
Represents the average consumption of each person in the population (country)
GDP per capita (US$ at constant 2000 prices) World Bank WDI 4 Technology (T)
Represents how resources intensive the production of affluence is; how much environmental impact is involved in creating,
16 All variables are taken from the World Bank World Development Indicators (WDI) database. From 41 countries in Asia continent, the obtainability of balanced data on CO2 emission, total population, GDP per capita, and GDP per unit of energy use limits the possible sample to 35 countries. The list of the countries that are included in this paper is in Appendix A.
The sample period for the annual data is 14 years from 1995 until 2008. The data has both time-series and cross-sectional properties (i.e. panel data). This panel data structure allows a broader range of issues to be addressed in comparison to data that solely contains time-series or cross-sectional properties. The amount of available data is simply larger. Additionally, by combining time-series and cross-sectional data the number of degrees of freedom, and thereby the power of the statistical tests, can be increased. General overviews of all of the data that are used in this paper are presented in Appendix B.
From those four figures in the Appendix B, it is clear that the data show an increasing trend in all the variables considered. This result gives preliminary conclusion that human impact has significant influence to the CO2 emission.
5. Results
There are three main sections of the results in this paper. First is the determination of appropriate specific fixed effects (spatial specific effects μ and/or time-period specific effects η). Second is spatial model specification. Third is the analysis of effects estimation.
5.1 Determination of Specific Fixed Effect
17 hypothesis that there are no spatial fixed effects (µi)(LR test: 1411.9130, with 35 degrees of freedom [df], p < 0.0000). However we fail to reject the null hypothesis that there are no time-period fixed effects ( t) (LR test: 8.5779, with14 degrees of freedom [df], p < 0.8571). Therefore we only include spatial specific effect (μ) into our model in the analysis below.
Table 2: Estimation results of IPAT equation using panel data models without
spatial interaction effects
LR test spatial fixed effects: (1411.9130, with 35 degrees of freedom [df], p < 0.0000) LR test time-period fixed effects: (8.5779, with14 degrees of freedom [df], p < 0.8571) LM test: spatial lag model
5.2 Spatial Model Specification
In this section we use the results of (classical) LM tests, (robust) LM test, Wald tests, LR tests, and Hausman test to choose the right spatial model
Determinants (1) (2) (3) (4) Pooled OLS Spatial fixed effects Time-period fixed effects Spatial and time-period fixed effect Ln(P) 0.955 (77.19) 0.935 (16.08) 0.955 (77.26) 0.807 (7.41) Ln(A) 3.079 (19.95) 2.517 (13.36) 3.078 (19.83) 2.542 (13.57) Ln2(A) -0.137 (-13.93) -0.101 (-7.84) -0.137 (-13.85) -0.107 (-8.04) Ln(T) -1.166 (-27.39) -1.032 (-19.81) -1.167 (-27.32) -1.022 (-19.41) Intercept -18.529 (-31.15) R2 0.942 0.741 0.942 0.503 LogL -302.38 399.51 -302.16 403.79 LM Spatial lag 44.51 14.07 30.93 11.32 LM Spatial error 4.75 1.78 4.89 1.39
Robust LM spatial lag 39.79 19.52 26.41 18.77
18 specification in our analysis. These results represent the steps of the specific-to-general approach. The result of (classical) LM tests and (robust) LM test are represented by table 2 and the result of Wald tests, LR tests, and Hausman test are shown by Table 3.
Table 3: Estimation results of IPAT equation: spatial Durbin model specification with spatial specific effects
Hausman test-statistic, degrees of freedom and probability = 114.3853, 9, 0.000
The result of (classical) LM tests (spatial lag = 14.07 and spatial error 1.78) shows that we should reject the null hypothesis that there are no endogenous interaction effects but we fail to reject the null hypothesis that there
Determinants (1) (2) (3)
Spatial fixed effects Spatial fixed effects Bias correction Random spatial effect, Fixed effect W*Ln(I) 0.082 (1.61) 0.082 (1.61) 0.384 (8.97) Ln(P) 0.683 (5.91) 0.683 (5.69) 0.879 (16.58) Ln(A) 2.235 (10.25) 2.235 (9.88) 1.721 (7.79) Ln2(A) -0.089 (-6.11) -0.089 (-5.89) -0.055 (-3.87) Ln(T) -0.992 (-17.83) -0.992 (-17.19) -0.923 (-17.75) W*Ln(P) 0.011 (0.08) 0.011 (0.08) -0.651 (-10.26) W*Ln(A) 0.661 (1.86) 0.661 (1.80) -1.444 (-5.03) W*Ln2(A) -0.028 (-1.24) -0.029 (-1.19) 0.087 (4.75) W*Ln(T) -0.281 (-2.58) -0.281 (-2.51) 0.298 (3.26) Phi 0.048 (5.92) σ2 0.011 0.012 0.0132 R2 0.997 0.997 0.9962 Corrected R2 0.752 0.752 0.7889 LogL 411.065 410.41 -140672.13
Wald test spatial lag 7.848 (0.097) 7.447 (0.114) 240.506 (0.000) LR test spatial lag 6.772 (0.148) 6.7724 (0.148)
19 are no interaction effects among the error terms. Since this result already points to one of the spatial models, we no longer need to consider the (robust) LM tests. As illustrated by the flow chart of the testing procedure in Figure 4, now we should conduct spatial Durbin model estimation and test the hypothesis whether the spatial Durbin model can be simplified to the spatial lag or spatial error model.
Based on the results of Wald and LR tests, we fail to reject the null hypothesis (Ho: θ=0) that the spatial Durbin model can be simplified to the spatial lag model at 5% significance (Wald test: 7.848 with p-value=0.097; LR test: 6.772 with p-value 0.148). This implies that the spatial lag model is acceptable. On the other hand, the results of LR and Wald tests regarding the hypothesis test whether the spatial Durbin model can be simplified to the spatial error model (Ho: θ+δβ=0) suggest us to reject the null hypothesis, which means that spatial Durbin model cannot be simplified to spatial error model (Wald test: 21.049 with p-value=0.000; LR test: 20.758 with p-value=0.000).
Even though we failed to reject the null hypothesis that spatial Durbin model can be simplified to spatial lag model (Ho: θ=0), we reject the null hypothesis that spatial Durbin model can be simplified to spatial error model (Ho: θ+δβ=0). Therefore the best choice of spatial econometric model is the spatial Durbin model. This decision is based on two reasons. First, since there is an evidence of the existence of exogenous interaction effects, omitting these effects in the model will lead to bias and inconsistence estimation (Greene, 2008). Second, as pointed out by Ertur and Koch (2007) and Videras (2012), all of our explanatory variables in this paper, population (P), economics activities (A), and Technology (T), are influenced by other countries on their neighboring economies. Therefore we cannot ignore spatial interaction effects in the explanatory variables.
20 Consequently, we should use spatial econometric techniques to analyze the IPAT equation.
Instead of fixed effects we can also treat µ and as random effects. Hausman’s specification test can be used to test the random effects model against the fixed effects model. However, whether the random effects model is an appropriate specification if the population may be said to be sampled exhaustively, such as all counties of a state or all regions in a country, remains controversial (Elhorst, 2011). Based on the result of Hausman test (114.3853, 9, p<0.000), we find that the random effects model must be rejected. Therefore we should use fixed effect panel data estimation.
5.3 Analysis of Effects Estimation
There are two important effects in spatial econometric models, i.e. direct effect and indirect effects. The results of these effects estimation are represented in Table 4.
Table 4: Effects estimates of IPAT equation
Note: t-values in parentheses (in column 2)
5.3.1 Direct Effects estimation
The direct effects estimates of the four explanatory variables reported in column (1) of Table 4 are significantly different from zero and have the expected signs. An increase in number of population (P) and economic activities in terms of GDP per Capita (A) lead to an increase of CO2 emission
Determinant (1) (2)
21 (I), while higher technology level in terms of GDP per unit of energy use (T) inclines CO2 emission (I).
Since we use logarithm form for all of the variables, we interpret all of these effects estimates as elasticity of our dependent variable (I) with respect to our independent variables (P, A, and T). Therefore, ceteris paribus, a one percent increase in the number of population (P) increases CO2 emission (I) by about 0.683 percent. Holding all other variables fixed, a one percent increase in GDP per capita (A) increases CO2 emission (I) by about 2.249 – 0.18 Ln(A) percent. A one percent increase in GDP per unit of energy use (T) decreases CO2 emission (I) by about 0.999 percent, ceteris paribus. Note that these elasticities are different from their coefficient estimate, which is due to feedback effects that arise as a result of impacts passing through neighboring countries and back to the countries themselves.
Based on the result of this effect, the relationship between CO2 emission (I) and GDP per capita (A) can be expressed with the equation below:
Ln(I) = 2.249 Ln(A) - 0.090 Ln2(A) [9]
The marginal effect of GDP per capita (A) to the CO2 emission (I) is 2.249 – 0.18 Ln(A). Equation [9] and the marginal effect of GDP per capita (A) to the CO2 emission (I) ensure the existence of an inverted U-shaped EKC since the polynomial form of the GDP per capita (A) has a negative sign and is statistically significant. Therefore, at the beginning, economic activities will lead to an increase of CO2 emission. However, as the economic activities continues to rise, the effect declines.
22 Figure 5: Graph of EKC analysis based on equation [9]
Figure 6: Marginal effect of GDP per capita on CO2 emission based on direct effects estimation
23 Figure 7: Graph of EKC analysis based on equation [9] using continuous value of GDP per capita.
Figure 7 gives strong evident for the existence of an inverted U-shaped EKC. However, as pointed out by Figure 5 and 6, there is no country in our data set that exceeds turning point. All of them are in the left side of the turning point (in the first and second stage of economic development). The nearest country to the turning point is Japan. Its highest value of GDP per capita is equal to US $ 40,707. The second and third nearest country to the turning point are United Arab Emirates and Singapore with their highest GDP per capita equal to US $ 35,316.34 and US $ 31,227.64 respectively. China, which is now become one of the country that achieve the highest Economic development level in Asia, is only in the 18th position. This is because our data set only until 2008 and in per capita term. Graphical illustrations of EKC per country based on this direct effects estimation are illustrated in Appendix C.
Therefore, to tackle environmental problem related to reduce CO2 emission, Asian countries need to increase their economic activities so that be able to reach the third stage of economic development. Another way that they can do is increasing their level of technology. By doing these policies the bad impact of human activities to the environmental condition will reduce.
Ln (A)
Ln
(
I)
24 5.3.2 Indirect Effects estimation
The spatial spillover effects (indirect effects estimates) of all four variables have the same sign as the direct effect. However, these effects are not significant for population and polynomial form of GDP per capita. Economic activities of a country will lead to an increase of CO2 emission not only in the own country, but to a limited extent also in its neighboring countries (elasticity 0.899). Conversely, an increase in technology will decline CO2 emission not only in the own country but also in neighboring countries (elasticity -0.386). Therefore economic activities and technology become the most important factors to the level of CO2 emission in a country since it produce direct and indirect effects.
Just as the direct effects estimation, we analyze the EKC using the relationship between CO2 emission (I) and GDP per capita (A). Based on indirect effects estimation, this relationship can be expressed as follow:
Ln(I) = 0.899 Ln(A) - 0.038 Ln2(A) [10] The Marginal effect of GDP per capita to the CO2 emission is = 0.899 – 0.076 Ln(A). Turning point of the EKC based on this effect estimation is when Ln(A) is
equal to 11.83. This point is equal to GDP per capita level equal to US $ 137,166.03 and CO2 emission level equal to 203.79 metric ton per capita.
Graphical illustration of equation [10] and the marginal effect of GDP per capita to the CO2 emission based on the data set that we have are represented by Figure 8 and Figure 9 respectively. To give additional evidence for the existence of an inverted U-shaped EKC, we use graphical illustration of equation [10] using continuous value of GDP per capita from 0 until its maximum value. This graph is represented by Figure 10.
25 Consequently, to reduce CO2 emission, the most important policy for all Asian countries is to increase their technology level since it produces direct and indirect effect. Another policy that they should do is increasing their economic activities so that they will exceed the turning point.
Figure 8: Graph of EKC analysis based on equation [10]
26 Figure 10: Graph of EKC analysis based on equation [10] using continuous value of GDP per capita.
6. Conclusion
Based on the analysis, there are three main conclusions from this paper. First, the decision on using spatial Durbin model with spatial specific effects as the model to be estimated on analyzing human impact to the environment condition is a strong evidence to the existence of spatial interaction effects in the data set. This argument is based on the analysis results that we must reject the null hypothesis, in which there are no endogenous interaction effects, no exogenous interaction effects and no spatial specific effects. Therefore, ignoring these effects will lead to biased estimation results. This conclusion is in line with the base theory that when the data set is spatial in nature so we should include spatial interaction effects in our analysis as mentioned by Videras (2012).
27 per capita. The slope of the marginal effects of this relationship is negative. Therefore, in the early stage of economic development, economic activities will lead to environmental quality deterioration. Then, as the economic activities continue increasing, this negative impact is sagging. However, based on the data set, we cannot illustrate a perfect an inverted U-shaped EKC graphically. The curve ends near EKC turning point. This means that all of the Asian countries are either in the first stage of economic development (pre-industrial economies) or in the second stage of economic development (industrial economies).
The last main conclusion from this paper is that population, economic activities, and technology have strong influence to the growth of CO2 emission. Yet, the strong influences are given by economic activities and technology since they produce direct and indirect effects to the level of CO2 emission in the atmosphere. On the other hand, population only gives strong influence through its direct effect. Thus, based on this conclusion, policy makers should put higher attention on economic activities by increasing it until exceed the turning point and increase their technology level on their policy consideration related to CO2 emission reduction. Consequently, by taking into account this conclusion on their policy consideration, they will obtain an effective and efficient policy in reducing CO2 emission as one of the answers of global warming problems.
28 References
Andreoni, J. & Levinson, A., (2001), The Simple Analytics of The Environmental Kuznets Curve, NBER Working Paper, Vol.80, pp. 269-286.
Anselin, L. (2006), Spatial Econometrics, In: Mills, T. and K. Patterson (eds) Palgrave
Handbook of Econometrics, vol. 1, pp. 901-969. Palgrave McMillan, Basingstoke.
Candia, L. (2003), “Does the Environmental Kuznets Curve Explain How a Growing Economy Can Achieve Better Environmental Qualityi?” Dundee Yearbook of
International Natural Resources and Energy Law and Policy, Vol. 7, No. 8, (University
of Dundee, UK).
Dhanda, K. K., Adrangi, B. & Chatrath, A. (2005), Linkage Between GDP and Emissions: A Global Perspective On Environmental Kuznets Curve, Journal of
Business & Economics Research (May), Vol 3, No.5.
Dietz, T. & Rosa, E.,A. (1997), Effects of Population and Affluence on CO2 Emissions,
Proceedings of the National Academy of Sciences, USA, Vol.94, pp.175-179.
Elhorst, J.P. (2010a), Spatial panel data models, In Handbook of applied spatial
analysis, eds. M.M. Fischer and A. Getis, pp. 377-407, Berlin: Springer.
Elhorst, J.P. (2010b), Applied Spatial Econometrics: Raising the Bar, In Spatial
Economic Analysis, Vol. 5, No.1, pp.9-28.
Ertur, C. & Koch, W. (2007), Growth, technological interdependence and spatial externalities: Theory and evidence, Journal of Applied Econometric, Vol. 22(6), pp. 1033-1062.
Greene, W.H. (2005) Econometric Analysis, 6th edition, Upper Saddle River, NJ, Pearson Prentice Hall.
Grossman, G.M. & Krueger, A.B. (1994), Economic Growth and the Environment,
NBER Working Paper (February): No. 4634.
LeSage, J.P. & Pace, R.K. (2009), Introduction to spatial econometrics, Boca Raton, US: CRC Press Taylor & Francis Group.
Malthus T.R. (1798), An essay on the principle of population, Oxford World's Classics reprint.
Manski, C.F. (1993) Identification of endogenous social effects: the reflection problem,
Review of Economic Studies, Vol. 60, pp. 531-542.
Onozaki, K. (2009), Population is a Critical Factor for Global Carbon Dioxide Increase,
Journal of Health Science (November), Vol. 55(1), pp. 125-127.
Panayotou, T. (1993), Empirical Tests and Policy Analysis of Environmental Degradation at Different Stages of Economic Development, Working Paper WP238
Technology and Employment Programme, Geneva: International Labor Office.
29 Seldadyo, H., Elhorst, J.P. & De Haan, J. (2010c), Geography and governance: Does space matter?, Papers in Regional Science, vol. 89, pp. 625-640.
Shi, A. (2001), Population Growth and Global Carbon Dioxide Emissions, International
Union for the Scientific Study of Population (IUSSP) Conference (June).
Videras, J. (2012), Spatial Patterns of Carbon Emissions in the U.S.: A Geographically Weighted Regression Approach, Hamilton Collage, United State.
30 Appendix A: List of Countries
ID Country Name 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 Bahrain Bangladesh Brunai Darussalam Cambodia China India Indonesia Iran Iraq Israel Japan Jordan Kazakhstan Korea, Rep. Kuwait Kyrgyz Rep Lebanon Malaysia Mongolia Nepal Oman Pakistan Philippines Russia Saudi Arabia Singapura Sri langka
Syrian Arab Republic Tajikistan
Thailand Turkmenistan United arab mirates Uzbekistan
31 Appendix B: General overviews of all of the data
Figure B1: Overview of Cumulative CO2 emission in Asia continent
Figure B2: Overview of Cumulative Total Population in Asia continent 0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00 18.00 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 x M ill io n s (k ilo to n ) Cumulative CO2 3.10 3.20 3.30 3.40 3.50 3.60 3.70 3.80 3.90 4.00 4.10 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 B ill io n s
32 Figure B3: Overview of Cumulative GDP per Capita in Asia continent
Figure B4: Overview of Cumulative GDP per Unit of Energy Use in Asia continent 190 200 210 220 230 240 250 260 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Th o u san d s (c o n stan t 2000 US $)
Cumulative GDP per Capita
0 1 2 3 4 5 6 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 M ill io n s (PPP $ p e r kg o f o il e q u iv al e n t)
33 Appendix C: Scatter Plot of EKC per Country Based on Direct Effects
37 Turning Point 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00 0 5 10 15 Ln ( I) Ln (A)
Saudi Arabia
Turning Point 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00 0 5 10 15 Ln ( I) Ln (A)Singapura
Turning Point 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00 0 5 10 15 Ln ( I) Ln (A)Sri langka
Turning Point 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00 0 5 10 15 Ln ( I) Ln (A)Syrian Arab Republic
38 Turning Point 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00 0 5 10 15 Ln ( I) Ln (A)
Turkmenistan
Turning Point 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00 0 5 10 15 Ln ( I) Ln (A)United arab mirates
39 Appendix D: Scatter Plot of EKC per Country Based on Indirect Effects
43 Turning Point 1.00 2.00 3.00 4.00 5.00 6.00 0 5 10 15 Ln ( I) Ln (A)
Saudi Arabia
Turning Point 1.00 2.00 3.00 4.00 5.00 6.00 0 5 10 15 Ln ( I) Ln (A)Singapura
Turning Point 1.00 2.00 3.00 4.00 5.00 6.00 0 5 10 15 Ln ( I) Ln (A)Sri langka
Turning Point 1.00 2.00 3.00 4.00 5.00 6.00 0 5 10 15 Ln ( I) Ln (A)Syrian Arab Republic
44 Turning Point 1.00 2.00 3.00 4.00 5.00 6.00 0 5 10 15 Ln ( I) Ln (A)
