5 Results: Time-Series Cross-Sectional Analysis & Geographic Information
5.1.2 Results
This paper examines whether the presence of a certain authoritarian regime type within a country can have an effect on the level of greenhouse gas emissions in a country, specifically carbon dioxide (CO2) emissions. Further, I hypothesize that regime type will have an effect on carbon dioxide emissions and that out of all authoritarian regime types, party regimes will have the greatest effect at reducing carbon dioxide emissions. Table 3 gives the results for the three models used in the time-series cross-sectional analysis, showing the effect on CO2 emissions in general (column one), and fixed effects (column 3). Column 2 gives the full model without time-fixed effects.
Table 3 outlines all three models used in this thesis. All models used in this thesis include random effects. In addition to this, monarchies are used as the baseline model, and are therefore omitted because of collinearity. The first column outlines the first model used. Model 1 is a generalized least squares regression with random effects where carbon dioxide emissions are regressed with all authoritarian regime types in order to see initial results. In this regression, all 51 regimes are included in this regression, and all regimes (party, military, and personal) are shown to be statistically significant at the 95th percentile (p-value<0.05). The results show that if all authoritarian regimes are present, then carbon dioxide emissions will decrease. However, contrary to my hypothesis, it is military regimes that have the strongest relationship and are less likely to have higher levels of CO2 emissions.
On the other hand, the results from Model 2 show to have a more profound effect. The second model is another generalized least squares regression which includes all explanatory and control variables. This model, which includes only 46 out of 51 regimes, shows that once the control variables are included, regime type does not have as profound of an effect. While all regimes are still statistically significant in the 95th percentile, once controlling for the economy, demographics, and industrialization, all regime types are not as effective at decreasing CO2 emissions. Again, in this regression, the results are contradictory to my hypothesis that party regimes will outperform all other authoritarian regime types, and show that military regimes are performing the best.
The control variables included in this regression that are controlling for the economy are trade as percentage of GDP and GDP per capita measured in US dollars. In this regression, controlling for demographics are the variables population density based on total percentage of population and
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
urban population showing the rate of urbanization in different regimes. Lastly, controlling for industrialization is the value added by industry as a percentage of GDP.
When looking at Model 2, one can see that not only do the control variables decrease the effect of authoritarian regime types, but they are also all statistically significant. GDP per capita has the greatest effect on carbon dioxide emissions out of any of the control variables, so that as GDP per capita increases, so does CO2 emissions. This could be a result of the Environmental Kuznets curve.
This is also true for urban population and industrialization, whereas urban population and industrialization increases, so do CO2 emissions. This could easily be due to the fact that as industrialization increases, more people will move to urban areas to look for job opportunities, and with more industrial output and human congestion, carbon dioxide emissions will increase.
However, for trade, the higher the level of CO2 emissions, the lower the amount of trade a country does. This effect is contradictory to my original understanding because normally more trade would mean more production, which would result in increased CO2 emissions. Population density, while statistically significant, also shows the relationship that the less dense a population is, the more CO2 emissions a country will have. This is contradictory as well because normally the more densely populated a regime is, the more likely it is to have higher CO2 emissions.
‧
Table 3: Authoritarian Regime Type and Carbon Dioxide Emissions: Models with Random Effects
VARIABLES (1) Model 1 (2) Model 2 (3) Model 3
Standard errors in parentheses; Monarchy regimes used as the baseline
*** p<0.01, ** p<0.05, * p<0.1
‧
carbon dioxide emissions could possibly be affected by the performance in previous years’, time-fixed effects need to be included. By doing so, it allows me to determine whether my hypothesis is correct and whether CO2 emissions are affected by authoritarian regime types rather than by time.Furthermore, Model 3 (column 3) shows that military regimes outperform all other authoritarian regime types, then party regimes, and then personal regimes, while monarchies are still omitted because of collinearity. When looking at authoritarian regimes, when large amounts of carbon dioxide emissions are present, it is less likely that the authoritarian regime is a military regime. In addition to this, the results also show that all authoritarian regime types besides personal regimes are statistically significant.
In Model 3, when looking at the control variables, GDP per capita is broken down into two variables: GDP per capita (log) and GDP per capita (log2). By logging GDP per capita and by then also generating a square term, I am able to control for the Environmental Kuznets Curve (EKC).
The EKC is based off the idea that the average income increase would first worsen then lead to an enhancement of the environmental conditions in a country (Grossman, Krueger 1995). Therefore, since GDP is the main control for the economy, it needs to be logged in order to test the effect of the EKC. In this model, GDP per capita is statistically significant and shows that the higher the CO2 emissions, the lower the levels of GDP. Therefore, these results do not support the Environmental Kuznets Curve hypothesis.
Model 3 also shows that all other control variables, except for industry are statistically significant.
Therefore, the results show that industry does not have an effect on carbon dioxide emissions in authoritarian regimes. In terms of urban population, the results are similar to Model 2, where as urban populations grow in an authoritarian regime, so will the levels of CO2 emissions. Lastly, the results for trade and population density are also the same as in Model 2. As a result, authoritarian regimes with a more dispersed population and lower levels of trade will be more likely to emit more CO2 emissions.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
5.2 Geographic Information System Analysis
The second form of analysis will consist of a spatial analysis that utilizes Geographic Information Systems (GIS). Therefore, in order to better visualize the data, the software “GeoDa” allows me to incorporate better methods to determine how the data is structured.
5.2.1 Summary Statistics: GIS Analysis
In figure 5.3, a boxplot is used to determine whether there are any significant outliers in the data that could skew the results. This figure shows a boxplot with the hinge set at 1.5. Another boxplot that can be used is one where the hinge is equal to 3.0. However, this type of boxplot is more restrictive. For example, with the hinge set at 1.5, there are six out of fifty-one regimes that are outliers. These outliers are Kazakhstan, Kuwait, Oman, Russia, Saudi Arabia, and United Arab Emirates (UAE). However, when the hinge is set to 3.0, there are only two outliers. These outliers are Kuwait and UAE. Therefore, by using the hinge equal to 1.5, one will have a better understanding as to which countries or regimes skew the data and make it higher than necessary.
Most of the countries that will skew the data are monarchies, with only one of them, Russia, being a personal authoritarian regime.
Figure 5.3: Box Map (Hinge = 1.5) to Determine Outliers in CO2 Emissions (Metric Tons per Capita)
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Another way to visualize and better interpret the data is through a parallel coordinate map, as seen in figure 5.4. A parallel coordinate plot allows one to plot multiple variables to see how they interact. Therefore, the variables I chose to use in this map are carbon dioxide (metric tons per capita) emissions for the years 1997 and 2010. As one can see from this map, most of the regimes in this dataset have lower levels of CO2 emissions. Also, most of these emissions do not increase drastically over the course of thirteen years. Even Kuwait, the largest emitter, does not drastically increase but slightly decreases with going from 32.11 to 29.96. However, with the larger CO2
emitters, they all increase their emissions except for Singapore which decreased its CO2 emissions from 15.39 to 10.96 metric tons per capita. Also, one can notice that most of the countries that are at the top have increased quite quickly regardless of whether or not they signed onto the Kyoto Protocol in 1997, such as Saudi Arabia (11.14 to 18.91) and United Arab Emirates (15.59 to 18.81).
Figure 5.4: Parallel Coordinate Map Representing the Change of CO2 Emissions (Metric Tons per Capita) From 1997 to 2010
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
one to see the relationship between the variables and how the different authoritarian regimes perform. Figure 5.5 is a map that shows the main relationship between authoritarian regime type and carbon dioxide (per capita) emissions. As one can see in this map, over time not only are monarchy regimes being outperformed by all other regime types, but also, the majority of these countries have increased their CO2 emissions.
In addition to this, the other maps show varying results. In Figure 5.6, for the relationship between GDP per capita and authoritarian regime type, most of the regimes do not change much over time, and only a few are increasing or decreasing. Figure 5.7 shows the trend of trade levels during this time period. One can see that most countries, regardless of regime type, have an increase in their levels of trade. Figure 5.8 is a map that shows the changes in population density levels. Most countries maintain a stable population density, except for a few countries which have increased over the years, such as United Arab Emirates (37.61 to 120.39). Next, Figure 5.9 shows the trend of urbanization by seeing how the urban population has changed in each country. Most countries have maintained a stable urban population. It is mainly party regimes with the most urban population growth: China, Vietnam, Laos, Eritrea, and Tanzania, with the exception of Yemen (personal). Lastly, Figure 5.10 is a map of industrialization; it shows the value added from industries as a percentage of GDP. Most regimes have seen growth in this area with a few having larger increases, such as the Democratic Republic of the Congo, Namibia, Laos, etc. However, some data skews the results due to the fact that there are no entries for Libya in 1997 and 2010 and for Kuwait in 1997.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Figure 5.5: Map for Carbon Dioxide Emissions in Authoritarian Regimes (1997-2010)
Figure 5.6: Map of GDP per Capita Levels in Authoritarian Regimes (1997-2010)
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Figure 5.7: Map of Trade Levels in Authoritarian Regimes (1997-2010)
Figure 5.8: Map of Population Density Levels in Authoritarian Regimes (1997-2010)
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Figure 5.9: Map of Urban Population Levels in Authoritarian Regimes (1997-2010)
Figure 5.10: Map of Industry Levels in Authoritarian Regimes (1997-2010)
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y 5.2.2 Results: GIS Analysis
This section presents the results of the estimations for carbon dioxide emissions measured in metric tons per capita and its relationship to authoritarian regime type. The results were produced using the GIS software GeoDa, and are presented using multiple forms of analyses. As stated prior, I hypothesize the effect of authoritarian regime type on carbon dioxide emissions will be more profound in party regimes than in all other types of regimes, where party regimes will be less detrimental to the environment and have fewer CO2 emissions. The following models allow for a visualization of these results.
Figure 5.11 and Figure 5.12, gives some preliminary results on the effect of adding weighted variables in the data to test the relationship of regimes and their CO2 emissions by using a Local G* clustering map. In order to do this, you must include a weighted variable that lets you test the spatial relationship of regimes with their own neighboring regimes. For this research, we are using distance weight rather than contiguity weight because the nature of the data is based on point data rather than polygon data. Therefore, using distance weight, I set the nearest neighbors function to the five closest neighbors, so that the spatial relationship for CO2 levels will be based on the five nearest neighbors to each regime. Because of this, one can see that there are three areas in red that denote high levels of spatial clustering. These regimes are Ethiopia, Oman, and Tanzania. This shows that these regimes not only have high levels of CO2 emissions, but also their five neighboring regimes also have high levels of CO2 emissions. Furthermore, there are five areas in blue that denote low levels of spatial clustering. These regimes are Cameroon, Cuba, Gabon, Syria, and Togo. This shows that these five regimes have low levels of CO2 emissions and so do their five neighboring regimes.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Figure 5.14: Local G* Cluster Map with the Distance Weight Set to the 5 Nearest Neighbors
Additionally, in Figure 5.12, I set the nearest neighbor function to the ten closest neighbors, so that the spatial relationship for CO2 levels will be based on a wider range of neighboring regimes.
Because of this, one can see that the amount of spatial clustering is reduced in both sections. There are now only two areas in red that denote high levels of spatial clustering. These regimes are Ethiopia and Oman. This shows that these regimes not only have high levels of CO2 emissions, but also their ten neighboring regimes also have high levels of CO2 emissions. Moreover, there is now only one regime instead of five in blue that denotes low levels of spatial clustering. This regime is Togo. Therefore, this shows that Togo has low levels of CO2 emissions and so does it’s ten neighboring regimes.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Figure 5.15: Local G* Cluster Map with the Distance Weight Set to the 10 Nearest Neighbors
Regression Analysis
This next section will include several regression analyses using GIS. The three models will include a classical regression analysis, and the other two will incorporate spatial analysis with a spatial lag model and a spatial error model. All three of these models will utilize a weighted file that measures the relationship between a specific observation and its surrounding observations based on distance (km). The distance is set at a standard rate generated by GeoDa: 6,639.41 km. By utilizing spatial analysis, we can determine whether carbon dioxide emissions can be influenced by a regime’s geographic location.
Model 1 in Table 4 uses the classical regression model. When first running this regression, in order to test the effect and significance of the relationship between the different authoritarian regime types and CO2 emissions, I only included the dummy variables for party, military, and personal without any control variables present. Also, just like in the main time-series cross-sectional model, monarchies are used as the baseline model and omitted due to collinearity. Therefore, as consistent with the main analysis’s results, Model 1’s results concluded that all three regimes are statistically significant with military regimes having the greatest effect.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
However, when adding in the economic controls such as GDP per capita (log) and GDP per capita (log2), the effect of the regime dummy variables completely diminished and was no longer statistically significant. In addition to this, even when adding in all control variables, GDP per capita still has the largest effect on carbon dioxide emissions in authoritarian regimes. The control variables used in this classical regression are the same as the ones used in the time-series cross-sectional analysis, yet only GDP per capita (log) and GDP per capita (log2) are statistically significant. In addition to this, the results are very similar in regards to the coefficients and statistical significance of the variables. Trade and population density are still contradictory, which could be a results of omitted variable bias (OVB).
Lastly, when utilizing a weighted file, one must also determine whether spatial auto correlation is an important factor. Therefore, spatial lag and spatial error must also be tested for. Table 5 shows the diagnostics for spatial dependence based on the classical regression. This figure determines whether or not to test for spatial lag and spatial error based on Lagrange Multiplier (lag) and Lagrange Multiplier (error). Based on probability, both of these variables show that they are significant in the 95th percentile (probability < 0.05). Therefore, the results show that spatial lag model and spatial error model must be tested. I also tested this relationship using different weights that incorporate different distances besides the standard distance set up by GeoDa: 10,000 km and 20,000 km. However, there was no evidence to support the use of spatial lag and spatial error using those weights.
‧
Table 4: Authoritarian Regime Type and Carbon Dioxide Emissions: GIS Results
(1) Model 1 (2) Model 2 (3) Model 3
‧
Table 5: Diagnostics for Spatial Dependence
TEST MI/DF Value Probability
Moran’s I (error) 0.1115 4.5781 0.00000
Lagrange Multiplier (lag) 1 10.9584 0.00093
Robust LM (lag) 1 5.3553 0.02066
Lagrange Multiplier (error) 1 5.6068 0.01789
Robust LM (error) 1 0.0037 0.95142
Lagrange Multiplier (SARMA) 2 10.9621 0.00416
Hence, Model 2 in Table 4 shows the regression results for the spatial lag model. In this model, the results for the coefficients have the same effect as the previous model in the classical regression.
In addition to GDP per capita (log) and GDP per capita (log2), trade is the only other control variable which is statistically significant. Therefore, as GDP per capita increases in a regime, CO2
emission levels are more likely to be higher and increase. Furthermore, the spatial lag regression is not only used to determine the relationship between the dependent and independent variables, but it also calculates for the spatial relationship between neighbors. Therefore, it also adds a weighted variable for CO2 (metric tons per capita) emissions: Weighted: CO2. This variable is shown to be statistically significant from its z-value (z-value > 2). However, while this improves the relationship, all of the coefficients are still very small, which still shows that these variables are not enough to determine a strong relationship with CO2 emissions.
Lastly, Model 3 in Table 4 provides the regression results for the spatial error model. Again, like the previous model, the spatial error model shows that GDP per capita (log) and (log2) have the highest significance. The variables are the same as in the last two models, whereas GDP per capita increases, carbon dioxide emissions will increase. However, population density and trade are still contradictory where as they decrease, CO2 emissions will increase. Also, the spatial error method also adds in another variable known as Lambda. Lambda is used to calculate the spatial error, which explains whether one has enough variables to determine the validity of the relationship being tested. If Lambda is high, then more variables need to be added. As a result, because Lambda
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
coefficient is greater than any other variable: 0.77, more variables possibly need added. With these results, we can conclude that what we have is not enough to satisfy proper results.
Furthermore, because these results are contradictory to our first method of analysis, time-series
Furthermore, because these results are contradictory to our first method of analysis, time-series