The Impact of Energy , Agriculture , Macroeconomic and Human-Induced Indicators on Environmental Pollution from 1971 to 2011

In this study, the impact of energy, agriculture, macroeconomic and humaninduced indicators on environmental pollution from 1971 to 2011 is investigated using the statistically inspired modification of partial least squares (SIMPLS) regression model. There was evidence of a linear relationship between energy, agriculture, macroeconomic and human-induced indicators and carbon dioxide emissions. Evidence from the SIMPLS regression shows that a 1% increase in crop production index will reduce carbon dioxide emissions by 0.71%. Economic growth increased by 1% will reduce carbon dioxide emissions by 0.46%, thus supports the environmental Kuznets curve hypothesis that an increase in a country’s economic growth leads to a reduction in environmental pollution. An increase in electricity production from hydroelectric sources by 1% will reduce carbon dioxide emissions by 0.30%, thus increasing renewable energy sources in Ghana’s energy portfolio will help mitigate carbon dioxide emissions. Increasing Enteric Emissions by 1% will increase carbon dioxide emissions by 4.22% and a 1% increase in the Nitrogen content of Manure Management will increase carbon dioxide emissions by 6.69%. The SIMPLS regression forecasting exhibited a 5% MAPE from the prediction of carbon dioxide


INTRODUCTION
Climate change as gained global prominence as a result of its long-term effect on the globe.
As a result, climate change mitigation through a sustainable global action deems essential to limit the rising levels of greenhouse gas emissions (Asumadu-Sarkodie and Owusu 2016b; Owusu and Asumadu-Sarkodie 2016).This global effort has propelled a lot of research interest in environmental, energy and agricultural sustainability.This paradigm shift in scientific research has increased the interest in using historical data to predict and/or explain the causal-effect between response and predictor variables.
Several studies have employed modern econometric techniques such as; Engle-Granger's method of cointegration (Engle and Granger 1987), Johansen's method of cointegration (Johansen 1995) Two strands of studies are analysed in the study.The first strand of existing literature estimates the causal nexus between carbon dioxide emissions and gross domestic product (GDP), or carbon dioxide emissions and energy-intensity/consumption/production, or a combination of carbon dioxide emissions, energy consumption, GDP and/or population.Huang et al. (2008) examined the causal relationship between energy consumption and GDP in 82 countries using the GMM model which found no evidence of causal relationship between energy consumption and GDP.Soytas and Sari (2009) examined the relationship between CO2 emissions, energy consumption and GDP using the long-run Granger causality test which found evidence of causal direction running from CO2 emissions to energy consumption but not valid in the reverse.Lozano and Gutié rrez (2008) examined the causal relationship between GDP, CO2 emissions, energy consumption and population in USA using a non-parametric frontier method which found evidence of steady efficiency increase in the average of the modelled variables for the estimated period.Jammazi and Aloui (2015) examined the relationship between CO2 emissions, energy consumption and economic growth for 6 countries spanning from 1980-2013 by employing a wavelet-window-crosscorrelation approach.Their study found evidence of a bidirectional causality between energy consumption and GDP while there was evidence of a unidirectional causality running from energy consumption to CO2 emissions.Zhang and Cheng (2009) examined the Grangercausality between CO2 emissions, energy consumption, GDP, and population.Their study found no evidence of causality between CO2 emissions and energy consumption and GDP.Gul et al. (2015) examined the relationship between energy consumption and carbon dioxide emissions in Malaysia with a data spanning from 1975-2013 by using the maximum entropy bootstrap method (MEBM).Their study found evidence of a unidirectional causality running from energy consumption to CO2 emissions.Qureshi et al. (2016) examined the relationship between energy crisis, GHG emissions and GDP for the Caribbean, Europe, Asia, Africa and Latin America with a panel data spanning from 1975-2012 by employing Johansen's method of cointegration and variance decomposition.Their study found evidence of negative significant relationship between electricity access and power shortage in certain regions of Europe and Asia.Remuzgo and Sarabia (2015) employed the factorial decomposition to examine the global distribution inequality of carbon dioxide emissions in IEA considered regions with data spanning from 1990-2010.Their study found evidence of causality between GDP per capita and global distribution inequality of CO2 emissions.Ouyang and Lin (2015) examined the relationship between energy consumption and GDP in China with a data spanning from 1991-2010 by employing the Logarithmic Mean Divisa Index (LMDI) method.Their study found evidence of a long-run equilibrium relationship between CO2 emissions, energy consumption, labour, industrial sector value added and fossil-fuel consumption.Ozturk and Acaravci (2010) examined the relationship between energy consumption and economic growth in Romania, Albania, Hungary and Bulgaria with a data spanning from 1980-2006 by employing the ARDL bounds testing method which showed evidence of a long-run equilibrium relationship.Their study found evidence of bidirectional causality between energy consumption and economic growth in Hungary.Ozturk and Acaravci (2011) examined the relationship between electricity consumption and economic growth in 11 Middle East and North Africa countries with a data spanning from 1971-2006 by employing the ARDL bounds testing method which showed no evidence of long-run equilibrium relationship for Iran, Morocco and Syria.Their study found evidence of longrun equilibrium relationship in Egypt, Oman, Israel and Saudi Arabia.The overall evidence of the 7 Middle East and North Africa countries showed no evidence of long-run equilibrium relationship.Acaravci and Ozturk (2010) examined the relationship between carbon dioxide emissions, electricity consumption and economic growth in 19 European countries with a data spanning from 1965-2005 by employing the ARDL bounds testing method which showed no evidence of long-run equilibrium relationship for Switzerland, Portugal, Iceland, Greece, Italy, Denmark and Germany.Their study supported the validity of the environmental Kuznets curve hypothesis.Apergis and Payne (2011) examined the relationship between renewable energy consumption, labour force, gross fixed capital formation and economic growth in 6 Central American countries with a data spanning from 1980-2006 by employing the FMOLS method which showed evidence of a long-run equilibrium relationship for all the 6 Central American countries.Their study found evidence of a bidirectional causality between renewable energy consumption and economic growth.
Asumadu-Sarkodie and Owusu (2016d) examined the multivariate causality analysis of the Kaya factors in Ghana by taking into consideration the total primary energy consumption, GDP, carbon dioxide emissions and population with a data spanning from 1980-2012 using the vector error correction model.Evidence from their study showed a long run causality from population, GDP and energy consumption to carbon dioxide and a bidirectional causality from CO2 emissions to energy consumption.
The second strand of existing literature estimates the causal nexus between carbon dioxide emissions, energy consumption, GDP and/or population by adding more variables like; foreign investments, industrialization, trade, urbanization, financial development, etc.. Shahbaz et al. (2015) examined the relationship between energy consumption from road transportation, transport sector value added, fuel prices and CO2 emissions in Tunisia with a data spanning from 1980-2012 by employing the ARDL bounds testing method in the presence of structural breaks which showed an evidence of a long-run equilibrium relationship.Their study found evidence of a bidirectional causality between energy consumption and CO2 emissions while fuel prices exhibit a unidirectional causality on energy consumption, road infrastructure, CO2 emissions and transport sector value added.Ben Abdallah et al. (2013) examined the relationship between energy consumption from road transportation, transport sector value added, fuel prices and CO2 emissions in Tunisia with a data spanning from 1980-2010 by employing the Johansen's method of co-integration and Granger-causality test which showed an evidence of a long-run equilibrium relationship.
Their study refuted the validity of the environmental Kuznets curve hypothesis and found an evidence of unidirectional causality running from fuel price to energy consumption from road transportation.Cerdeira Bento and Moutinho (2016) examined the relationship between CO2 emissions per capita, non-renewable electricity production per capita, renewable electricity production per capita, GDP per capita and international trade with a data spanning from 1960-2011 in Italy by employing the ARDL bounds testing method in the presence of structural breaks which showed an evidence of a long-run equilibrium relationship.Their study confirmed the validity of the environmental Kuznets curve hypothesis and found an evidence of unidirectional causality running from GDP per capita to renewable electricity production per capita.Seker et al. (2015) examined the relationship between foreign direct investment and environmental quality with a data spanning from 1974-2010 in Turkey by employing the ARDL bounds testing method in the presence of structural breaks which showed an evidence of a long-run equilibrium relationship.Their study confirmed the validity of the environmental Kuznets curve hypothesis and found an evidence of unidirectional causality running from foreign direct investment, GDP per capita and energy consumption to CO2 emissions.Rafindadi and Ozturk (2015) examined the relationship between natural gas consumption, capital, exports and labour with a data spanning from 1971-2012 in Malaysia by employing the ARDL bounds testing method in the presence of dual structural breaks which showed an evidence of a long-run equilibrium relationship.Their study found no evidence of causality running GDP per capita and natural gas consumption.Salahuddin et al. (2015) examined the relationship between CO2 emissions, electricity consumption, economic growth and foreign direct investment with a data spanning from 1980-2012 in the Gulf Cooperation Council by employing DOLS, FMOLS and DFE methods which showed an evidence of a long-run equilibrium relationship.Their study found evidence of unidirectional causality running from electricity consumption to CO2 emissions and a bidirectional causality between economic growth and CO2 emissions.Tiwari et al. (2013) examined the relationship between coal consumption, capital, trade openness, economic growth and CO2 emissions with a data spanning from 1966-2011 in India by employing the ARDL bounds testing method in the presence of structural breaks which showed an evidence of a long-run equilibrium relationship.Their study confirmed the validity of the environmental Kuznets curve hypothesis and found an evidence of bidirectional causality between economic growth and CO2 emissions, and between coal consumption and to CO2 emissions.Shahbaz et al. (2012) examined the relationship between CO2 emissions, energy consumption, trade openness and economic growth in Pakistan with a data spanning from 1972-2009 by employing the ARDL bounds testing method which showed an evidence of a long-run equilibrium relationship.Their study confirmed the validity of the environmental Kuznets curve hypothesis and found an evidence of unidirectional causality running from economic growth to CO2 emissions.
Spurious regression occurs as a result of multicollinearity problems among study variables in a model.As a result, informative variables that are essential in explaining a specific response are dropped.Multicollinearity problems have been reported to exist when analysing variables that include environmental pollution, energy consumption and economic growth (Asumadu-Sarkodie and Owusu 2016a; Asumadu-Sarkodie and Owusu 2016e).As a result of multicollinearity problems, the inability for some econometric models (VAR, VECM and ARDL) to estimate coefficients per equation due to insufficient observations and the inability to estimate large number of candidate models due to the large number of regressors or the maximum number of lags used in a model, many studies employ few predictor variables which may not be informative in explaining a response variable(s).
Contrary to the aforementioned econometric models, the partial least squares regression is able to estimate the relationship between variables that exhibit strong multicollinearity, have the maximum number of lags and have more or equal number of variable observations than the predictor variables.The partial least squares regression analysis has been employed in previous studies outside the scope of the study (Ceglar et al. 2016;Mehmood et al. 2012;Xu et al. 2012), and has been labelled as a flexible method for multivariate analysis.Against the backdrop, the study explores the worth of estimating the impact of energy, agriculture, macroeconomic and human-induced indicators on environmental pollution in Ghana with 21 variables using the statistically inspired modification of partial least squares (SIMPLS).The study contributes to the global debate on climate change through the use of a versatile methodology and more informative variables that are essential in explaining environmental pollution in Ghana.
The remainder of the study comprises of: "Methodology", "Results and Discussion", and "Conclusion and Policy recommendations".

METHODOLOGY Data
The study examines the impact of energy, agriculture, macroeconomic and human-induced indicators on environmental pollution using the SIMPLS regression model.A time series data spanning from 1971 to 2011 were employed from World Bank (2014) and FAO (2015) as shown in Table 1.Table 2 presents the descriptive statistical analysis of the study variables.
Evidence from Table 2 shows that with the exception of EPH, all the remaining variables are

Partial Least Squares Regression
The partial least squares regression involves the use of principal component analysis to decide on a regression model with the latent variables and the response variables.The SIMPLS regression model directly computes the partial least squares factors as linear combination of the original study variables (De Jong 1993).The SIMPLS regression model was developed with the aim of explaining a specific optimality problem (i.e.maximizing covariance criterion between the predictor variables and the response variable, on condition that the predictor variable scores are orthogonal).As a result, it has been suggested that the SIMPLS regression model is to some extent superior over the non-linear iterative partial least squares (NIPALS) (Asumadu-Sarkodie and Owusu 2016c; Wise 2004).

Model Estimation
Prior to the SIMPLS regression model, a linear regression analysis is estimated to examine the relationship between energy, agriculture, macroeconomic and human-induced indicators and environmental pollution, expressed as: Where,  0 represents the intercept,  1 ,  2 ,  3 ,…,  20 represent the "projected change in the mean response of the predictor variables for each unit change in their value" and   represents the white noise.
For brevity, the SIMPLS regression model is expressed as:

Assessment of SIMPLS Regression
A leave-one-out cross validation and a randomization test known as Van der Voet T 2 is used in the SIMPLS method to ascertain whether a model with a specific number of latent variables are different from the selected optimal model by the root mean predicted residual error sum of squares (PRESS).Evidence from Table 4 shows that the minimum root mean PRESS is 0.2407 and the minimizing number of factors in the SIMPLS method is 5.After the selection of the number of optimal latent variables, the study examines the percentage variation explained in SIMPLS model.Evidence from Figure 2 shows that almost 98% of variation is explained for the predictor variables while 97% of variation is explained for the response variable.The Variable Importance Analysis The importance of the predictor variables in explaining carbon dioxide emissions is estimated using the variable importance of projection (VIP), which examines the contribution of each predictor variable based on the variance explained by each SIMPLS latent variable (Mehmood et al. 2012;Wold et al. 2001).The first analysis in the SIMPLS model led to the deletion of some variables that had VIP value less than 0.83, since literature considers them as irrelevant (Gosselin et al. 2010).Eriksson et al. (2001) categorizes the explanatory variables based on their VIP thus, "highly influential" ( > 1), "moderately influential" (0.8 <  < 1) and "less influential" (  ℎ 0.8).In the examination of the variable importance analysis in Table 5, the explanatory variables can be categorized based on their VIP value.IND, RES, TRD, GDPP, FOSI, EN2OMA, MNMA and MNMM can be classified as moderately influential variables while POP, LIVE, EPH, CROPI, EMCH4BCR, EMN2OBCR, BBCRDM, EMN2OCR, RCR, Stock, ECH4En and can be classified as highly influential variables.Figure 3 presents a plot of VIP versus coefficients for centred and scaled data.The black dotted lines represent the "threshold line" while the red-solid-circles represent the level of carbon dioxide emissions.Values of the coefficient below zero represent a negative contribution towards carbon dioxide emissions while values greater than zero represent a positive contribution towards carbon dioxide emissions.It appears from Figure 3 that, the higher the VIP, the higher the coefficient contribution and the lower the VIP, the lower the coefficient contribution.

Impact Analysis
The impact analysis of energy, agriculture, macroeconomic and human-induced indicators on environmental pollution is evident in Table 6.Evidence from In contrast, a 1% increase in industrialization will increase carbon dioxide emissions by 0.44%.A 1% increase in Biomass burned crop residues will increase carbon dioxide emissions by 0.69%.Methane Emissions from Burning crop residues increased by 1% will increase carbon dioxide emissions by 0.74%.population increased 1% will increase carbon dioxide emissions by 1.03%.A 1% increase in Crop Residue will increase carbon dioxide emissions by 1.56%.A stock of animals increased 1% will increase carbon dioxide emissions by 3.29%.Increasing Enteric Emissions by 1% will increase carbon dioxide emissions by 4.22% and a 1% increase in the Nitrogen content of Manure Management will increase carbon dioxide emissions by 6.69%.However, the value of the intercept indicates that when there is no increase in the energy, agriculture, macroeconomic and human-induced indicators, there will be no carbon dioxide emissions.

Diagnostics of the SIMPLS Method
The SIMPLS regression method is subjected to diagnostic checks to estimate the independence of the residuals in the model.Evidence from the diagnostic plot s shows that the actual response and the predicted variable nearly fit on the regression line along with evidence of normal distribution by the residual normal quantile plot. 22

SIMPLS Forecasting and Performance
Figure 5 shows the Actual versus SIMPLS Predicted CO2.The study explores the usefulness of the SIMPLS regression method in predicting the response variable.Evidence from Figure 5 shows that the SIMPLS predicted carbon dioxide emissions nearly fits the original data with a mean absolute percentage error (MAPE) of 5.16% which is acceptable.By the addition of a trend line, the following equation is generated based on sixth-polynomial: y = 3E-05x 6 -0.003x 5 + 0.1045x 4 -1.0881x 3 -2.7697x 2 + 127.4x + 2200.3 with an R² = 0.9742.97% of the SIMPLS predicted carbon dioxide emissions were explained by the sixth-polynomial equation.

CONCLUSION AND POLICY RECOMMENDATION
The study examines the impact of energy, agriculture, macroeconomic and human-induced indicators on environmental pollution from 1971 to 2011 using the SIMPLS regression model.There was evidence of a linear relationship between energy, agriculture, macroeconomic and human-induced indicators and carbon dioxide emissions.
Evidence from the SIMPLS regression model shows that, Crop production index increased by 1% will reduce carbon dioxide emissions by 0.71%.GDP per capita increased by 1% will reduce carbon dioxide emissions by 0.46%, thus supports the environmental Kuznets curve hypothesis that an increase in a country's economic growth leads to a reduction in environmental pollution.Electricity production from hydroelectric sources increased by 1% will reduce carbon dioxide emissions by 0.30%, thus increasing renewable energy sources in Ghana's energy portfolio will help mitigate carbon dioxide emissions.
In addition, a 1% increase in industrialization will increase carbon dioxide emissions by 0.44%.A 1% increase in Biomass burned crop residues will increase carbon dioxide emissions by 0.69%.Methane Emissions from Burning crop residues increased by 1% will increase carbon dioxide emissions by 0.74%.population increased 1% will increase carbon dioxide emissions by 1.03%.A 1% increase in Crop Residue will increase carbon dioxide emissions by 1.56%.A stock of animals increased 1% will increase carbon dioxide emissions by 3.29%.Increasing Enteric Emissions by 1% will increase carbon dioxide emissions by 4.22% and a 1% increase in the Nitrogen content of Manure Management will increase carbon dioxide emissions by 6.69%.However, the value of the intercept indicates that when there is no increase in the energy, agriculture, macroeconomic and human-induced indicators, there will be no carbon dioxide emissions.
Evidence from the SIMPLS regression forecasting shows that the predicted carbon dioxide emissions fit the original data with a performance of 5% MAPE and a 97% R squared value from the performance of a trend line data analysis based on sixth-polynomial equation.The following policy recommendations are proposed in the study: Integrating climate change measures into national energy policies, sustainable agricultural policies, strategies and planning will increase institutional capacities towards adopting climate change measures, adaptation, early warnings and impact reduction.
Industrial policies that aims at promoting local technological development, scientific research, innovation and creativity, access to affordable internet and the provision of conducive environmental policies will propel Ghana's effort towards achieving sustainable industrialization.
Finally, there is the need for an enhanced local financial institutional capacity to boost access to financial services, insurance and banking in order to attract foreign investments, create jobs and propel the country's financial development.
positively skewed.In addition, EN2OMA, ECH4MM, EPH, FOSI, GDPP, IND, MNMA, MNMM, RES and STOCK exhibit a leptokurtic distribution while the remaining variables exhibit a platykurtic distribution.From the Jarque-Bera test statistic, EN2OMA, ECH4MM, EPH, FOSI, GDPP, IND, MNMA, MNMM, RES and STOCK do not fit the normal distribution while the remaining variables are normally distributed based on 5% significance level.Figure shows the trend of the predictor variables versus the response variable.With the exception of EPH, it appears from Figure 1 that all the variables have a positive monotonic relationship with CO2.

16Figure 2 .
Figure 2. (a) Percent Variation Explained for X Effects and (b) Percent Variation Explained for Y Responses

Figure 4 .
Figure 4. Diagnostics Plots (a) Actual by Predicted Plot (b) Residual Normal Quantile Plot

Table 1 .
Data and Variable Definition

Table 3 .
Linear Regression Analysis

Table 4 .
Cross Validation with SIMPLS Note: The minimum root mean PRESS is 0.2407 and the minimizing number of factors is 5.

Table 5 .
Variable Importance Analysis

Table 6 .
Model Coefficients for Centred and Scaled Data