Undoubtedly, one of the main determinants of the growth and development of countries is the per capita GDP (gross domestic product). The fact that OECD countries, which the present study was carried out on, consist of high- or upper-medium-income countries strengthens the correlation between income and growth. Growth is a factor that increases the CO₂ emission. Thus, it was aimed to observe the effect of GDP on pollution by adding this variable into the model (Shang et al., 2023). Many studies have been carried out to determine the relationship between growth and environmental pollution by making use of different methods. The most important effect of the increase in the pollution rate on the environment is the increase in the increasing greenhouse gas emissions. Carbon dioxide (CO₂) is the main source of greenhouse gas. The causal relationship between greenhouse gases produced by human activities and global warming was proven in the fifth assessment report of the United Nations Intergovernmental Panel on Climate Change (IPCC) in 2014 (Shao et al., 2019; Huang et al., 2022). Gross domestic product is the most widely used measure of the size of an economy (Ponce and Khan, 2021; Maroufi and Hajilary, 2022). The origin of research on GDP and environmental pollution is defined under the Environmental Kuznets Curve (EKC) hypothesis, which is based on the Kuznets curve developed by Kuznets (Kuznets, 1955). However, a lot of empirical research has been done to investigate the negative effects of growth on the environment.

In recent studies, the relationship between the energy consumption variable and pollution is one of the frequently researched subjects. Any use of energy is necessary to sustain many economic activities and produce goods and services. Therefore, while the increase in production increases the need for energy, it also leads to an increase in CO₂ emissions and environmental pollution (Çamkaya et al., 2022). Countries need energy to continue many economic activities, produce goods and services, and maintain social and economic developments. Moreover, it is considered as a fundamental factor in reducing the poverty and increasing the welfare level and living standards (Alkan and Albayrak, 2020). ENC variable was added to the model since the fact that OECD countries include both developing and developed countries would cause these countries to have different energy consumption and pollution levels. In many studies, it was tried to measure the amount of pollution created by using different types of energy, such as fossil fuel use such as oil and coal and CO₂ emissions (Acar and Lindmark, 2017), electricity consumption and CO₂ emissions (Asumadu-Sarkodie and Owusu, 2016; Asumadu-Sarkodie and Owusu, 2017), and the relationship between renewable energy sources and CO₂ emissions (Hasnisah et al., 2019; Kirikkaleli and Adebayo, 2021).

From a general perspective, urbanization describes the migration from rural areas to cities. Urbanization has numerous and versatile effects on CO₂ emissions (Niu and Lekse, 2017). Considering the structures of OECD countries, developed and developing countries differ in terms of urbanization rate. URB variable was added to the model in order to observe the differences between the countries in terms of CO₂ emissions. The energy-growth-emission relationship is related to the urbanization and environment. Studies showed that urbanization can increase both income and energy use, and such an increase is evident at all stages of urbanization progression (Khan et al., 2019; Zhou, 2019; Guo et al., 2023). Many factors such as household size, changing industrial structure, increase in the number of new residences and new public facilities, and city size distribution increase the level of pollution (Abbasi et al., 2020). Many studies have been carried out on a local and national basis and they showed that rapid industrialization and urbanization caused an increase in CO₂ emissions. CO₂ emissions, whether overall or per capita, are affected by regional characteristics (Khoshnevis and Golestani, 2019).

The relationship between trade and the environment is one of the important research topics. Since the 1990s, the positive and negative effects of trade openness on the environment have been the subject of the Environmental Kuznets Curve (EKC), Pollution Paradise, and Race to the Bottom hypotheses (Fan et al., 2019). Some of the studies revealed that the increase in foreign trade increased CO₂ emissions (Ansari et al., 2019; Liu et al., 2019; Nurgazina et al., 2021). Some other studies revealed that foreign trade has no effect on CO₂ emissions or it changes depending on the export rate and time (Mutascu, 2018; Salahuddin et al., 2018; Lv and Xu, 2019; Iqbal et al., 2021). Adding DAE into the model, it was aimed to test the effects of trade relationships between countries (neighbors or not) on CO₂ emissions according to both approaches by considering the geographical locations of OECD countries.

Spatial analysis methods have been frequently used in recent years. Thus, the classical methods used to explain real-life relationships have been replaced by spatial models that take location labels into account while explaining the relationships between variables (Başar, 2009). Using the spatial analysis method, the trends of the factors affecting CO₂ emissions in the form of increase, decrease, and remaining constant has been observed in many studies (Panayotou, 2016; Chen et al., 2019; Yang et al., 2019; Jezierska-Thöle et al., 2022). This aspect of the study is expected to contribute to the literature.

In this study, the variables that analyze the effect of economic growth on environmental degradation and are selected in parallel with the studies in the literature were first analyzed with the EKC method. Then, for the spatial analysis, the GDP² variable was removed from the model developed and the model was established as follows. ln C O 2 = β 0 + β 1 l n E N C + β 2 l n G D P + β 3 l n D A E + β 4 l n U R B + ε (1)

In Eq. 1, CO₂ refers to carbon dioxide emissions (metric ton/person), ENC to energy use (oil equivalent kg/person), GDP per capita to real gross domestic product (US$ in 2010 constant prices). DAE to the trade openness index (share of exports and imports in the gross domestic product), and URB to urbanization rate (share of the urban population in the total population).

The data used in the study cover the years 1990–2015, and in model no 1, the coefficient β₀ is the constant term and it represents other variables that are not included in the model. The coefficient β₁ (ENC) shows the change in carbon dioxide emissions versus a 1% change in energy consumption. The coefficient β₂ (GDP), however, shows the change in carbon dioxide emissions versus a 1% change in gross domestic product and the variable β₃ represents the openness index (DAE) and represents the 1% impact on carbon dioxide emissions. Finally, the variable β₄ (URB) shows the effect of a 1% change in the urbanization rate on carbon dioxide emissions. The data used in the analysis were obtained from the World Bank (World Development Indicators-WDI) database.

In this research, firstly, a spatial model of the variables was created with the explanatory spatial data analysis method. Regional decompositions were determined by map distributions and Moran scatter diagrams. The parameters of the model were estimated by the least squares method. Spatial dependency tests were conducted to determine whether the model contains spatial dependence. After determining the type of spatial dependence, the spatial model was created by including the dependency in the model. The study was completed by using the maximum likelihood method for the parameters of the spatial regression model. In the study, the trends of increase, decrease, and remaining constant in all variables for the beginning and ending years are also presented visually with thematic maps. In addition, by applying regression analysis to the model, it was determined that the error variances were equal and normal distribution so that the model could be used.

Moran scatter diagrams were prepared to observe the spatial interaction of the map distributions and the variables in which regional differences are shown visually. The graph is scattered across four regions by the Moran I values of the countries. The maps by the regions of these countries in the graph were obtained by making use of LISA analysis. Spatial regression analysis was performed with the data set method determined in the last stage. In Spatial Regression, the decision-making process was followed step by step. The results were interpreted according to the significance levels of ***0.01, **0.05, *0.10.

In the regression model, some tests are conducted to detect the presence of spatial dependence. These.

• Lagrange multiplier test,

Before the spatial effects are predicted, it should be tested if the spatial dependence will be added into the model as a spatial delay or as a spatial error. Therefore, Anselin proposed the Lagrange Multiplier (LM) test, which was developed by Anselin in 1988 and determines the structure of the dependency for the spatial error self-relation in the presence of a lagged dependent variable (Anselin, 1988; Yılgör, 2019). L M λ = e ′ W e / e ′ e n − 1 2 t r W ′ W + W 2 (2) where, tr (trace) represents the sum of the diagonal elements of the matrix. σ 2 = e ′ e n − 1 indicates error variance. Except for tr (W'W + W²), the L M λ statistic equals to the square of the Moran I statistic (Fıscher Manfred and Jinfeng, 2011). This test statistic has an asymptotic X² distribution with 1 degree of freedom.

Contrary to Moran I test approach, it is based on the maximum likelihood method. The hypotheses for the L M λ test are as follows. H 0 : λ = 0 H 1 : λ = 0

The LM _lag test, another spatial dependency test related to the LM method, is formulated as follows. L M ρ = e ′ W y / e ′ e n − 1 2 W X β ^ ′ M W X β ^ / e ′ e n − 1 + t r W ′ W + W 2 (3) M = I − X X ′ X − 1 X is the projection matrix in the test. β ^ shows the LCC estimates of the β ^ included in the linear regression model. This test statistic has an asymptotic X² distribution with a degree of freedom of 1.

The hypotheses for the L M ρ test are as follows. H 0 : ρ = 0 H 1 : ρ ≠ 0

While the H₀ hypothesis states that the spatial autoregressive parameter is not significant, the H₁ hypothesis states that this parameter is significant (Anselin and Rey, 1991).

The most common method used to test spatial autocorrelation is Moran’s I statistic applied to regression residuals. The statistics is expressed as follows. I = n W 0 * e ′ W e e ′ e

Here, n represents the number of observations, W represents the spatial weight matrix, e = y − X β represents the nx1 dimensional vector of the errors estimated by the EKK in the regression model, and e’ indicates the sum of the squares of the wastes (Cliff and Ord, 1972). W 0 = ∑ i = 1 n ∑ j ≠ i n W i j (4)

It is the standardization factor that shows the sum of the non-zero weights of the spatial weight matrix. In the weight matrix standardized according to the rows, this total number of units gives n. Since n / W 0 = 1 , the Moran statistics is as follows. I = e ′ W e e ′ e (5)

The null hypothesis of this test clearly states that there is no spatial dependence, whereas the alternative hypothesis does not express a sharp type of dependence. Since the spatial correlation structure under the alternative hypothesis is not clear, this test only investigates spatial sequential dependence and does not provide information about the type of addiction (Anselin and Rey, 1991; Anselin, 1998; Aral, 2016).

They created the Robust determination test, which was developed by Bera and Yoon in 1993 and using the asymptotic distributions of the LM tests. This test contains differences in the covariance and mean of the LM statistic. Based on this study, Anselin et al. developed Robust statistics in 1996 (Anselin et al., 1996; Aral, 2016).

Robuts transformation tests are used to determine which model originates from the spatial dependence. In the presence of a spatial lag-dependent variable, it tests the spatial error dependence, and in the presence of a spatial error dependence, it tests the spatial lag dependency.

Under the assumption of λ ≠ 0, 0, the hypothesis H₀: ρ = 0, Under the assumption of ρ ≠ 0, it tests the H₀: λ = 0 hypothesis (Zeren, 2011). R L M λ = e ′ W 2 e / σ ^ 2 − T 21 n J ^ p . β − 1 e ′ W 1 y / σ ^ 2 2 T 22 − T 21 2 n J ^ p . β − 1 (6)

Here n J ^ p . β − 1 = σ ^ 2 W 1 x β ′ M W 1 x β + T 11 σ ^ 2 − 1 , T i j = t r W i W j + W i W j ands M = I − X X ′ X − 1 X . Moreover e = y − x β ^ and σ 2 = e ′ e n − 1 .

In case of W 1 = W 2 = W , T 11 = T 21 = T 22 = T = t r W ′ W + W 2 , the test is shown as follows (Anselin et al., 1996). R L M λ = e ′ W 2 e / σ ^ 2 − T n J ^ p . β − 1 e ′ W 1 y / σ ^ 2 2 T 1 − T n J ^ p . β − 1 (7)

Here n J ^ p . β − 1 = σ ^ 2 W 1 x β ′ M W 1 x β + T 11 σ ^ 2 − 1 T i j = t r W i W j + W i W j and M = I − X X ′ X − 1 X . Moreover, e = y − x β ^ and σ 2 = e ′ e n − 1

If W 1 = W 2 = W , it demonstrates that T 11 = T 21 = T 22 = T = t r W ′ W + W 2 and the thest is shown as below: R L M ρ = e ′ W 1 y / σ ^ 2 − T 12 T 22 − 1 e ′ W 2 e / σ ^ 2 2 n J ^ p . β − T 21 2 T 22 − 1 (8) W 1 = W 2 = W , the test statistic is as follows (Anselin et al., 1996). R L M ρ = e ′ W y / σ ^ 2 − T 12 T 22 − 1 e ′ W 2 e / σ ^ 2 2 n J ^ p . β − T (9)

The joint LM test tests the co-significance of the spatial delay and spatial error models. The joint LM test is expressed as follows. R L M ρ = e ′ W y / σ ^ 2 − e W e / σ ^ 2 2 n J ^ p . β − T + e ′ W e / σ ^ 2 2 T (10)

The joint LM test statistic yields the sum of the LM statistics obtained from the L M ρ and R L M λ or L M λ and R L M ρ tests. L M ρ λ = L M ρ + R L M λ = L M λ + R L M ρ (11)

This test has an asymptotic χ^2 distribution with a degree of freedom of 2 (Anselin et al., 1996).

In the estimation of spatial regression models, the most widely used method is the maximum likelihood (ML) method. This method was first introduced by Keith Ord in the year 1975. In the ML approach, the probability calculation of the common distribution of the observations is maximized according to a set of parameters. The ML estimation has the desired asymptotic properties (efficiency, consistency, and asymptotic normality, etc.).

The logarithmic likelihood function for the spatial delay model is as follows (Anselin, 1988). L = − n / 2 ln 2 π − n / 2 ln 1 / n e 0 − ρ e L ′ e 0 − ρ e L + ln I n − ρ W − 1 / 2 σ 2 I − ρ W Y − X β I − ρ W Y − X β (12)

To estimate the parameters with the maximum likelihood method, the model should be maximized with respect to σ² and β. However, the parameters of the logarithmic likelihood function are not linear. Anselin proposed substituting some of the parameters in the original likelihood function in order to reduce the number of unknowns in the problem. For this purpose, the likelihood function is reduced to a condensed likelihood function through the estimates obtained for σ² and β. The parameters of the spatial autoregressive model and the estimation steps of β by Anselin in 1988 are as follows (Zeren, 2011).

• In the classical regression model, y = X β 0 + ε 0 equals to β 0 ^ = X ′ X − 1 X ′ y .

By applying these steps, the coefficients for the lag model are obtained (Anselin, 1988; LeSage and Pace, 1999).

The steps in estimating the parameters of the spatial error model developed by (Ord, 1975) are as the following.

• Residues are obtained from the y = X β 0 + ε 0 classical regression model.

The l n L c function is maximized with respect to λ .

• The variables y ^ = I − λ W y and x ^ = I − λ W x are calculated.

In the model developed by Anselin in 2005, the steps and operation are followed as schematically illustrated in Figure 1. The first step is to conduct a classical regression analysis. Classical regression model results are examined and LM statistics are examined considering if there is a dependency or not. If LM_λ and LM_ρ are statistically insignificant, then it is concluded that there is no spatial dependence. In this case, the spatial error model and the spatial delay model will not be valid. Thus, it is decided to use the results obtained from the classical regression model.

If the LM_λ test, which examines only the spatial error dependence of the LM statistics, is statistically significant, the study is continued with the spatial error model. If the LM_ρ test, which examines the dependence of the spatial delay, is statistically significant, then the study continues with the spatial delay model.

If both LM_λ and LM_ρ tests are statistically significant, then RLM_λ and RLM_ρ tests are conducted in order to determine where the spatial dependence originates from. If the RLM_λ test in which the error dependence is tested is significant, then the spatial error model is used. If the RLM_ρ test in which the delay dependence is examined is significant, then the spatial delay model is used.

Map plays an important role in descriptive spatial data analysis. Maps allow for visualizing the pattern of the spatial distribution. In applications, quartile maps that divide the dataset into four equal parts are used in general (Fıscher Manfred and Jinfeng, 2011). At this point, for the variables used in the study, maps were created using CO₂ emissions, ENC, GDP, URB, and DAE data for the years 1990 and 2015 to observe the increase, decrease, or stability in the variables.

It can be seen in 1990–2015 CO₂ change maps (Figure 2) that, in the CO₂ emission map, the countries with the highest CO₂ emissions are shown in the darkest colors. As the colors become lighter, the CO₂ emission rates decrease. Comparing the data of 1990 and 2015, CO₂ emissions increased in Japan, Korea, New Zealand, and Norway. On the other hand, CO₂ emissions decreased in Denmark, France, Germany, and Hungary. No change in CO₂ emissions was observed in Australia, Austria, Belgium, Canada, Chile, Colombia, Czech Republic, Estonia, Finland, Greece, Iceland, Ireland, Israel, Italy, Latvia, Lithuania, Luxembourg, Mexico City, Netherlands, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden, Switzerland, Turkey, the United Kingdom, and the United States.

As can be seen in 1990–2015 ENC rates maps (Figures 3, 4), considering the energy usage rates and comparing the years 1990 and 2015, ENC rates increased in Austria, Israel, Japan, Korea, Slovenia, and Spain. It decreased in Estonia, Hungary, Latvia, Lithuania, Slovakia, Switzerland, United Kingdom. There was no change in Australia, Belgium, Canada, Chile, Colombia, Czech Republic, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Italy, Luxembourg, Mexico, Netherlands, New Zealand, Norway, Poland, Portugal, Sweden, Turkey, and the United States.

It can be seen in 1990–2015 GDP rates maps (Figures 3, 4) that, examining the GDP indicators of OECD countries, GDP decreased in Denmark, Hungary, and Mexico in 1990 and 2015. It increased in Chile, Korea, and Poland. No change was observed in Austria, Australia, Belgium, Canada, Colombia, Czech Republic, Estonia, Finland, France, Germany, Greece, Iceland, Ireland, Israel, Italy, Japan, Latvia, Lithuania, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Slovakia, Slovenia, Spain, Sweden, Switzerland, Turkey, the United Kingdom, and the United States.

As can be understood from 1990–2015 URB rates maps (Figures 3, 4), given the data on urbanization rates of OECD countries, urbanization rates decreased in Australia, Czech Republic, Estonia, Latvia, Lithuania, New Zealand, Spain, and Switzerland in 1990 and 2015. It increased in Colombia, Hungary, Italy, Japan, the Netherlands, Norway, and Turkey. No change was observed in Austria, Belgium, Canada, Chile, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Israel, Korea, Luxembourg, Mexico, Poland, Portugal, Slovakia, Slovenia, Sweden, the United Kingdom, and the United States.

It can be seen in 1990–2015 openness index rates (DAE) maps (Figures 3, 4) that, given the openness index DAE indicators, DAE decreased in Chile, Germany, Israel, Latvia, Lithuania, New Zealand, Norway, Switzerland, and the United Kingdom in 1990 and 2015. It increased in the Czech Republic, Greece, Hungary, Mexico, Poland, Spain, and Sweden, whereas no change was observed in Austria, Australia, Belgium, Canada, Colombia, Denmark, Estonia, Finland, France, Iceland, Ireland, Italy, Japan, Korea, Luxembourg, Netherlands, Portugal, Slovakia, Slovenia, Turkey, and the United States.

The spatial interaction of the variables, regional differences of which are shown visually with the map distributions, are examined by making use of Moran scatter diagrams. Moran I value in the scatter diagram refers to the slope of the regression curve between the values of the relevant variable and the values obtained by multiplying by the spatial weight matrix of the variable (Anselin, 1996). In spatial autocorrelation, global effects originate from regional variations, whereas local effects originate from the spatial dependence between neighboring regions (Bivand and Wong, 2018).

LISA analysis, which is a local indicator of spatial relationship, is used to examine whether the observation values show regionally significant spatial clustering or segregation. The Moran scatter diagram consists of four main regions, and the countries are scattered over the four regions of the graph by their Moran I values. These countries can be mapped to the regions in the chart by using LISA analysis. In the LISA analysis, countries at the 1% significance level can be grouped into four clusters, which are HH, LH, LL, and HL. Moran scatter diagrams and LISA clustering maps for the years 1990 and 2015 of green economy indicators of OECD countries are given below.

Figure 5 shows that, given the CO₂ emission data, Moran I value was determined to be 0.121 in 1990 and 0.187 in 2015. These values show that there is a positive autocorrelation between the CO₂ emission values of the countries. It was found that autocorrelation increased in 2015.

LISA maps show that, considering the 1990s CO₂ emissions, Canada was clustered in HH, Mexico in LH, and Greece in LL. Other countries were not clustered into any of these four significance levels.

Considering 2015s CO₂ emissions, Canada is clustered in HH, Estonia in HL, and Mexico in LH. While Estonia could not be included in any cluster at the significance level in 1990, it was clustered in HL at the significance level of 1% in 2015. Mexico was clustered in LH in 1990 and LH in 2015. While Canada was clustered in HH in 1990, it was clustered in HH in 2015.

Figure 6 shows that Moran I value of the ENC ratio variable was determined to be 0.313 in 1990 and 0.486 in 2015. These values indicate that there is a positive autocorrelation between the values of the countries. It was found that autocorrelation increased in 2015.

LISA clustering maps were created using ENK data of 1990 and 2015 for OECD countries. In 1990, Canada was clustered in HH, Mexico in LH, and Greece in LL at the significance level of 1%. In 2015, Canada, Finland, Norway, and Sweden were clustered in HH and Mexico in LH. While Finland, Norway, and Sweden were not included in any cluster at the significance level of 1% in 1990, they were clustered in the HH cluster in 2015. While Greece was clustered in LL in 1990, it was not included in any cluster in 2015.

Figure 7 shows that Moran I value of the GDP variable was calculated to be 0.012 in 1990 and 0.035 in 2015. Considering the GDP variable, it was determined that there was positive autocorrelation between countries. LISA maps show that Canada was clustered in HH, Denmark and Mexico in LH, and Estonia and Latvia in LL in 1990–2015 at a significance level of 1%.

Figure 8 shows that, while the Moran I value of the urbanization rate variable was 0.115 in 1990, this rate increased to 0.474 in 2015. It was determined that there was positive autocorrelation between countries regarding the URB variable.

LISA maps show that France, Germany, and Luxembourg were clustered in HH, Hungary in LL, in the Netherlands in LH in 1990. In 2015, Austria, Czech Republic, Hungary, Slovakia, and Slovenia were clustered in LL, whereas Belgium, France, Luxembourg, and the Netherlands were clustered in HH. For the KO variable, while the Netherlands was clustered in LH in 1990, it was clustered as HH in 2015. While Belgium was not included in any cluster in 1990, it was clustered in HH in 2015. Austria, Czech Republic, Slovakia, and Slovenia were not included in any cluster in 1990, but clustered in LL in 2015.

Figure 9 shows that Moran I value of DAE variable was calculated to be 0.119 in 1990 and 0.096 in 2015. It was determined that there was positive autocorrelation between countries in terms of the DAE variable.

LISA maps shows that Canada, Greece, and Mexico were clustered in LL in 1990 by the DAE variable. In 2015, Canada, Greece, and Mexico were clustered in LL, and the United Kingdom in LH. In 1990, the United Kingdom was not clustered in any cluster. The United Kingdom was clustered in LH in 2015.

Least Squares (LS) method in regression analysis is a statistical method that aims to minimize the sum of squares of error. In this study, while the CO₂ emission rate was accepted as the dependent variable for the years 1990 and 2015, the ENK, GDP, URB, and DAE variables were considered as independent variables. The estimation results regarding the parameters of the model are as follows (Tables 1, 2).

It was determined that, on CO₂ emissions, the GDP and URB variables were significant at the confidence level of 90%, while the other variables were not statistically significant. In the model established for the data of the year 1990, the error variances must have an equal and normal distribution in order for the LCC estimations to be valid and for the model to be used. For the normality test of error variances, the relevant assumptions in this model were met considering the Jarque-Bera test results. Given the Breusch-Pagan and Koenker-Bassett test results for the varying variance, the model met the relevant assumptions. Error variances of the model have an equal and normal distribution.

It was determined that, considering the CO₂ emission, the GDP was statistically significant in the 99% confidence interval, and other variables were not statistically significant. In the model established for 2015 data, the error variances must have equal and normal distribution in order for the LCC estimations to be valid and for the model to be used. For the normality test of error variances, the relevant assumptions in this model were met according to the Jarque-Bera test results. Considering the Breusch-Pagan and Koenker-Bassett test results for the varying variance, the model met the relevant assumptions. Error variances of the model have equal and normal distribution.

Considering the fact that the data used in the application belong to OECD countries and their geographical locations are taken into account, it is thought that it might contain spatial dependence since it might be affected by neighborhood relations. Thus, it was investigated whether there is a spatial dependence of the countries in terms of green economic indicators (CO₂, ENC, GDP, URB and DAE), taking into account the neighborhood relationships of the countries. For this purpose, spatial dependency tests were conducted. Spatial dependency test results are shown in the Tables 3, 4 below.

For the Moran I test statistic, the probability value was calculated to be 0.66038. Since this value is higher than 0.01, the H₀ hypothesis, which states that there is no spatial dependence, is accepted. Although it was determined that there is no spatial dependence in OECD countries for 1990 data, the Moran I test does not provide information about the structure of spatial dependence. Examining the LM test values in the spatial dependence tests for the 1990 data, it was determined that they were not statistically significant. Thus, the interpretation should be made using the results of the regression analysis.

In Moran I test statistic, the probability value was calculated to be 0.73665. Since this value is higher than 0.1, H₀, which indicates that there is no spatial dependence, is accepted. For the data of the year 2015, it was determined that there is no spatial dependence in OECD countries. For the same year, it was also determined that they were not statistically significant when the spatial dependence tests and LM test values were examined. Thus, the results of the regression analysis should be taken into account.

Given the results of the 1990 and 2015 regression analysis, there is a significant relationship between the dependent variable CO₂ and the variables ENC, GDP, and DAE. The URB variable does not have a significant relationship. Considering the 1990 data, ENC variable was found to be 0.01, GDP and DAE variables are statistically significant at the significance level of 0.05. Considering the data for the year 2015, ENC, DAE, and GDP variables are significant at the significance levels of 0.1, 0.05, and 0.01, respectively. ENC, GDP, and DAE variables are positively related to the CO₂ variable.

The present study was carried out by conducting spatial analysis and geographical regression analysis. All the OECD countries were examined individually. The results achieved from the methods used are explained comparatively. Since multiple countries and multiple variables were used in this study, the data were gathered from the World Bank for the period until 2015 and it poses a limitation for this study. However, it would be an honor for us if the current study team will be included in future studies. The results achieved here are explained below.

1) Comparing the data of the years 1990 and 2015, CO₂ emissions increased in Japan, Korea, New Zealand, and Norway. On the other hand, the CO₂ emissions decreased in Denmark, France, Germany, and Hungary. Considering the energy usage rates; Comparing the years 1990 and 2015 are compared, ENC rates increased in Austria, Israel, Japan, Korea, Slovenia and Spain. It decreased in Estonia, Hungary, Latvia, Lithuania, Slovakia, Switzerland, United Kingdom. In 1990 and 2015, GDP decreased in Denmark, Hungary, and Mexico but increased in Chile, Korea, and Poland. No changes were observed in other countries. The reason for this is estimated to be that the countries examined here consist of high- and upper-middle-income countries. Given the data on urbanization rates of OECD countries, URB decreased in Australia, Czech Republic, Estonia, Latvia, Lithuania, New Zealand, Spain, and Switzerland in 1990 and 2015. However, it increased in Colombia, Hungary, Italy, Japan, the Netherlands and Norway, and Turkey. Considering the DAE indicators, DAE decreased in 1990 and 2015 in Chile, Germany, Israel, Latvia, Lithuania, New Zealand, Norway, Switzerland, and the United Kingdom. It increased in the Czech Republic, Greece, Hungary, Mexico, Poland, Spain, and Sweden. The distributions of the maps of the variables in the form of increase, decrease and remaining constant are given in Table 5.

The world have been facing a major environmental disaster in the last century. Temperatures are rising rapidly, precipitation cycles are changing, and glaciers and snow are melting, raising seawater levels on a global scale. Both natural processes and human activities create greenhouse gases. The most important natural greenhouse gas in the atmosphere is the water vapor. However, human activities cause the release of large amounts of greenhouse gases, increasing the atmospheric concentrations of these gases, which increases the greenhouse effect and also increases the temperature. It is essential to reduce these emissions in order to mitigate the climate change. Emergency measures were discussed at the United Nations Climate Change Conference held in Glasgow, Scotland, between 31 October and 12 November 2021. The world has warmed by 1°C when compared to the pre-industrial period and only 1°C more increase in the temperature cannot be tolerated. Therefore, environmental problems are core problems not only in developed, developing, or underdeveloped countries, but all the countries. However, China, the USA, and the European Union countries, which increase pollution the most, rank the first three globally. In addition, it causes a dramatic increase in pollution in underdeveloped countries.

Since the present study covers OECD countries and all of these countries are developed and upper-middle-income developing countries, no significant changes were observed in the findings of the countries. In addition, the data covering a period of 25 years poses a time constraint. However, it is hoped that it will be a stepping stone for comparison in future studies or combinations of different countries. Countries whose carbon dioxide emissions are clustered in HH, HL should give importance to environmental measures on the basis of policies. There is an increase in CO₂ emissions throughout the study. Countries with the highest increase in CO₂ include Japan, Korea, New Zealand and Norway.

The relationship between income growth and pollution, which the present study aims to prove as the reason for CO₂ emission, is a common problem not only for OECD countries, but also for the whole world. In addition, considering the critical warnings mentioned above, the issue is too serious to be left to the initiatives of states. For this reason, it is essential for states to immediately implement practices that are supported by sanctions and include obligations, rather than basing the process on a voluntary basis, in order to protect the environment. From this point of view, the following suggestions can be taken into account on what should be done to reduce CO₂ emissions all over the world.

Although developments advance technology and make our lives easier, green growth should be preferred in order to reduce its negative impact on CO₂ emissions. Growth and sustainability should be made the fundamental principles in every field of life and every sector. In doing so, priority should be given to advancements in areas that significantly affect CO₂ emissions. For example, transportation preferences should be quickly harmonized with nature. In particular, air transportation is one of the factors that cause the most CO₂ emissions in the world. All transportation systems, especially airlines, should be taken under control with policies such as mode of transportation and pricing according to distance, in a way that will reduce CO₂ emissions. This should be implemented as a sustainability policy, not a restriction on freedom. The share of fossil fuels in energy consumption should be reduced and the use of clean energy sources such as solar, wind, and biomass energy should be promoted. People should use their nutritional preferences towards organic and local agricultural products instead of products such as meat, refined oil, alcohol, and sugar. Disposable products should not be used outside, except for the case of preventing risk factors. Consumption of need rather than consumption of desire should be a way of life. Sorting garbage, eliminating wild landfills, and supporting recycling should be involved in the immediate action plans. In addition, laws should be made use of when necessary in order to reintroduce organic wastes to the soil and to prevent the natural resources from pollution. The usage period (except for single-use) of products should be applied not only for electronic goods but also for all products with multiple uses.

It is clear that it is not sustainable for people in the world to continue living upon the same habits. Even if a series of decisions have been taken in the studies on the environment, most of them could not yield the desired results due to the non-participation of the big states or the states that have implemented the practice later for economic or political reasons. For instance, Australia, Canada, Japan, and Russia, which had taken on commitments for the First Commitment Period of the Kyoto Protocol, did not taken on any commitment for the Second Commitment Period. The Second Commitment Period (Doha Amendment) of the Kyoto Protocol, which was signed by 144 participating countries, came into effect on 31 December 2020. Due to the initiation of the ‘Paris Agreement’ that governs the climate regime post-2020, the Second Commitment Period has only been symbolically acknowledged. As a result, the Kyoto Protocol has fulfilled its role as the initial implementation tool of the Framework Convention on Climate Change (unfccc.int/kyoto_protocol). Binding policies under the auspices of the United Nations can ensure collective action by all countries. This could lead to a general improvement. Otherwise, the chances of success for isolated policies remain slim.