Corruption Control, Renewable-led Energy Transition and Carbon Emissions: Empirical Evidence from Panel Data of Multi-countries

This paper focuses on the carbon emission reduction effect of anti-corruption mediated by renewable-led energy transition based on the panel data of 98 10 countries from 1996 to 2015. Since the mediation model is estimated by estimating a series of multiple regression equations, the total, direct and indirect effects can be 11 separated out to clarify the conduction path between corruption control, renewable energy and carbon emissions. Owing to confirmation of the Sobel-Goodman 12 mediation tests, renewable energy acts as a significant mediator through which corruption control contributes to emission reduction, regardless of the indicator is total 13 carbon emissions, carbon emissions per capita or carbon emission intensity. For policymakers and regulators, there needs to be more emphasis on eliminating corruption 14 in the increased penetration of renewable energy and not being seduced by traditional lobbyists. Particularly for developing countries, the effective way to reduce 15 emissions is to remove institutional barriers in priority areas including the energy and resources sectors to cleaner-oriented energy transition. investigate the spatial

of resources it causes is the root cause of many social problems (Cole, 2007;Fredriksson and Svensson, 2003; De Jong and Bogmans, 2011). Resource-intensive and 23 capital-intensive industries are often hard hit by corruption (Kolstad, et al., 2008). The energy sector is resource-intensive and capital-intensive, so it is easy to imagine 24 the endless corruption cases (Gisladottir et al., 2020). For example, in China, many of the most egregious cases of corruption have occurred in the energy sector, which 25 has not only caused serious economic losses, but also harmed environmental governance and improvement (Hao et al., 2020). 26 Ever since the Paris climate agreement in 2015, in which 195 countries pledged to limit global temperatures to no more than 2℃ above pre-industrial levels and do 27 their best to limit rises to around 1.5℃, energy transition has become synonymous with the future of energy. It is the general trend of the world energy transformation 28 to implement clean substitution on the energy supply side and electric energy substitution on the energy consumption side, so as to form an energy pattern dominated 29 by clean energy and centered on electricity. Major countries in the world have simultaneously accelerated the development of low and decarbonization energy systems. 30 The development of renewable energy in countries around the world is mainly focused on solar, wind and biomass energy to accelerate the energy transition process, 31 improve energy security and reduce dependence on fossil fuels. The positive transformation to green, low-carbon and other clean energy and renewable energy is 32 mainly reflected in the shift of energy investment focus to green and clean energy, and the further optimization of industrial structure and energy consumption structure. 33 Developed countries in Europe and the United States have put forward clear energy transition plans and measures, which is an important pilot signal of the profound 34 changes in the international energy system under the background of new scientific and technological revolution, climate change and green and low-carbon environment. 35 Corruption, however, has become a major obstacle to the energy transition. Bribery and corruption cost the world economy at least $1.5 trillion annually, about 2 36 percent of global GDP, which could have been used to eradicate poverty, create jobs and protect the environment(IMF,2016). A widening gap in environmental regulation 37 would amplify the impact of trade convergence, forcing polluting firms to move to polluting havens. After decades of globalization, some countries, especially 38 developing and emerging economies, have moderate levels of regulatory corruption that attract polluters to move in, and the existence of traditional lobbies makes 39 pollution havens and corruption paradise suitable for some European companies (Candau and Dienesch, 2017). Polluting corporate lobbies still have ample bribe capital 40 to influence the quality of environmental policy implementation. When there is a certain level of corruption, the quality of environmental policy implementation will 41 be the balanced result of the game between various interest groups, which cannot make up for the negative externalities caused by environmental pollution. 42 In this paper, we estimate the total and intensity effects of corruption control on carbon emissions by using multi-country panel data and instrumental variable method. 43 Different from the previous literature, considering that increasing the share of renewable energy and reducing the dependence on fossil energy are the basic ways of 44 environmental pollution, we study the corresponding mediating effect, that is, how corruption affects renewable energy and then carbon emissions. We adopt the fixed 45 effect model of instrumental variables to eliminate the errors caused by endogeneity. High income countries, middle-income countries and low-income countries show 46 obvious heterogeneity in corruption control. In order to ensure the robustness of the parameters, we make group estimation and draw more enlightening conclusions. corruption campaign can not only directly reduce pollution levels, but also indirectly reduce pollution through improving the intensity of environmental regulations 84 and the level of economic development. The anti-corruption movement is more of an external shock, which is more objective than the shock form of studying the 85 change of official positions in previous literature. At the same time, the study found that the pollution reduction effect of the anti-corruption campaign is more significant 86 in China's resource-based cities, northern cities and western regions. 87 The indirect effects of corruption on pollution are studied by scholars from the perspectives of economic development level, hidden economy and resource allocation, discussed this indirect effect for the first time, but the author did not control the potential endogeneity between income and corruption and did not consider the 90 heterogeneity of different countries. Cole (2007) overcome these two problems. Based on the data of 94 countries from 1987 to 2000, Cole constructed two joint 91 equations in which corruption and pollution are mutually causal variables. Through analysis, it was found that the direct impact of corruption on SO2 and CO2 emissions 92 per capita is positive, but the indirect impact based on income is negative, and the absolute value of indirect impact is greater than that of direct impact. Except for 93 some high-income countries, the total effect of corruption on pollution is negative, that is, corruption will not lead to an increase in pollution level. In contrast, when 94 . However, the existence of corruption will cause the turning point of the environmental Kuznets curve to move backward, that is, the 102 turning point will appear at a higher per capita income level and pollution level (Lopez and Mitra 2000). Moreover, the higher the degree of corruption, the greater the 103 deviation between the actual path of income and pollution and the social optimal path without corruption factors (Leitão 2010). 104 Corruption can also affect the quality of the environment by affecting the hidden economy. Blackman (2000) discovered that hidden economic activities can seriously used spatial measurement analysis method to investigate the spatial agglomeration effect of environmental pollution in China and the environmental effects of FDI and 114 regional corruption. The study found that the inflow of FDI reduces China's environmental pollution, while regional corruption reduces the environmental benefits 115 brought by FDI. Dong 2020) and other indicators. The best way to measure corruption is to directly observe the bribery of officials, but the bribery behavior of officials is generally 120 more obscure, so it is not easy to be observed. Most of the existing research on corruption focuses on its impact on economic development. There is still a lack of discussion on the relationship between corruption 129 and pollution, and there is no consensus on the relationship between corruption and pollution, and the mechanism of corruption on pollution is not clear. Moreover 2012; Sekrafi and Sghaier, 2016), the sequential conduction route of corruption, renewable-led energy transition and carbon emissions has not been clarified (Nicolli 137 and Vona, 2015). Theoretically, as presented in Fig.1, there are two accessory pathways from corruption to carbon emissions, one is directly acting on carbon emissions, 138 the other is indirectly acting through the renewable-led energy transition, in which the renewable-led energy transition plays a mediating role. The increasing material 139 wealth of mankind has driven economic growth at the cost of energy consumption. The polluting nature of fossil fuels requires that they be replaced by clean and 140 renewable energy sources. But in this process, the short-sighted behavior of established interest groups, corrupt practices such as bribery and collusion, impede the 141 clean-oriented energy transition. Specifically, in this paper, in the middle of the relationship between corruption and carbon emissions, the share of renewable energy 142 in total energy consumption serves as a mediator. These mediating effects are assessed by estimating the following equations: 143 (1) 144 (2) 145 where refers to the corban emissions of country at year , which can be numerically expressed as total carbon emissions( ), carbon emission per capita( ) 147 and carbon emission intensity( ),respectively. denotes is the annual corruption control index of the evaluated country. is share of renewable energy 148 in total energy consumption.
is the set of control variables, including GDP per capita( ), the proportion of industrial added value in total output value( ), 149 the share of fossil energy in total energy consumption( ), and the share of net energy import in energy use( ). , and are the error terms of the 150 corresponding equations. The estimated coefficient presented in Eq.
(1) and Fig.1 should be referred to as the total effect of corruption control on carbon emissions, 151 which is distinguished from the estimated coefficient ′ in Eq.(3) that represented as the direct effect of corruption control on carbon emissions after controlling for 152 the share of renewable energy. The mediational effect, in which corruption control leads to carbon emissions through the share of renewable energy, is called the 153 indirect effect, which represents the portion of the relationship between corruption control and carbon emissions that is mediated by renewable-led energy transition. 154 At first, a four step approach proposed by Baron and Kenny (1986) which tends to miss some true mediational effects(MacKinnon et al., 2007). However, turning to 155 the equivalent approach calculates the indirect effect by subtraction between the total effect and the direct effect of the two regression coefficients in Eq.(1) and Eq.(3) 156 as following, 157 The goal of conducting mediation analysis is to indirectly assess the effect of a proposed cause on dependent variable through a proposed mediator. The utility of 159 mediation analysis stems from its ability to go beyond just describing the relationships between variables to understanding the relationships between variables. A 160 necessary component of mediation is that indirect effects are statistically and practically significant(Preacher and Hayes, 2004). Thus, in this paper, we apply the Sobel-161 Goodman tests to inspecting whether renewable-led energy transition performance as the mediator that carries the influence of corruption control on carbon emissions. 162 Once the coefficient of the indirect effect has been estimated, its significance needs to be tested. In particular, the mediating effect in this paper can be said to occur 163 when (i) the independent variable ( ) significantly affects the mediator( ) and the dependent variable( ) , (ii) the significantly affects in the absence 164 of ,(iii) has a significant unique effect on , and (iv) the effect of on shrinks upon the addition of the mediator to the model.

Data descriptions 169
The panel data used in this paper generally comes from the world development indicators (WDI) of the World Bank, which is compiled from officially recognized 170 international sources. As for carbon emissions, they are statistically derived from the burning of fossil fuels and the manufacture of cement, including solid, liquid and 171 gaseous fuels as well as natural gas. In addition to total carbon emissions, per capita carbon emissions and carbon intensity are also used as dependent variables. The 172 corresponding variable definitions, units of measurement and statistical descriptions are presented in Table 1 For mediation variables, in this paper, it specifically refers to the share of renewable energy consumption in the total final energy consumption labeled by Ren. Due 184 to the limitation of resource endowment, technical conditions and absorption mechanism, there are obvious differences in renewable energy among countries. 185 Statistically speaking, the value of the index is between 0 and 98%, which means that some countries have negligible renewables, while others derive almost all their 186 energy consumption from renewables. In terms of countries, it can be seen from Fig.3 that the higher penetration of renewable energy is mainly distributed in regions 187 such as Africa and Northern Europe. In addition, we also provide a set of control variables, the main purpose of which is to control income level, industrial structure, 188 proportion of fossil energy and energy import dependence, etc. The statistical description and global distribution of these variables are shown in Table 1 and Fig.3  189 respectively. From the conduction mechanism of various variables and multicollinearity of variables, the correlation coefficient between variables should be within a 190 reasonable range( see Table 2), not too high to lead to multicollinearity, nor too low to lack of basic statistical correlation.  As the core variable of this paper, corruption control indicator comes from the Worldwide Governance Indicators (WGI) project, covering indicators of six governance 195 indicators of more than 200 countries in the world since 1996, which has become an important basis for global policy makers and non-governmental organizations to 196 measure the level of government administration. The WGI is a research data set that summarizes the views of a large number of respondents from enterprises, citizens 197 and experts in industrial and developing countries on the quality of governance. The data come from a number of research institutions, think tanks, non-governmental 198 organizations, international organizations and private enterprises. Among them, corruption control observes the extent to which public power is exercised for self-199 interest, including large and small forms of corruption, and the occupation of national resources by the elite and private interests. In terms of the index itself, the higher 200 the value indicates that the better the country controls corruption, and the lower the corruption level. Overall, corruption control continues to improve in most countries, 201 despite the huge gap between high-income and low-income countries. Corruption leads to the distortion of renewable and non-renewable energy resources allocation, 202 which affects the carbon emission situation. However, whether it meets the significance needs to be tested by the above methods. 203

Basic results 205
Before the Sobel-Goodman mediation test, we first estimate the total effect of corruption control on carbon emissions under path , the results of which are presented 206 in Table 3. Since there are three measurement methods of carbon emission in this paper, namely total carbon emissions, per capita carbon emissions and carbon emission 207 intensity, their respective estimation parameters can be obtained. For the total effect, The estimated coefficient of is significantly negative, and one unit increase 208 in the corruption control indicator will reduce total carbon emissions by about 39.1%, indicating that from the perspective of total effect, corruption control can reduce 209 total carbon emissions, and reducing corruption is an important way to reduce pollution emissions. The reflects the overall situation of a country's carbon emissions. 210 Considering the total population of each country, we will get another picture. A typical example is China, which is the largest emitter in the world, but it is in the middle 211 level when considering per capita. The total effect of is also significantly negative. Specifically, increasing one unit of the corruption control indicator will reduce 212 carbon emissions per capita by 15.2%. Turning to carbon emission intensity, the total effect is basically consistent with the former two( and ), and even the 213 estimated coefficients of and are very close, although their global distribution based on quantity varies greatly. In general, corruption control has an 214 effective negative effect on carbon emissions from the perspective of total volume, per capita and intensity, which shows that path is positive and feasible. 215 Restraining corruption can reduce pollution in both absolute and relative quantity, which fully shows that anti-corruption is an important way to reduce emissions, and 216 in turn explains as it is proposed by Candau and Dienesch (2017) that why some low-income countries are both pollution havens and corruption paradise. 217 218 Table 3 Estimation results of the total effect(path ) 219 220 Regarding the path which reflects the impact of anti-corruption on renewable energy consumption, although major countries in the world have set medium-and 225 long-term renewable energy consumption share targets, and even some countries have put forward 100% renewable energy planning for carbon neutralization, and 226 technological development has led to a significant reduction in the cost of renewable energy production, institutional and governance deficiencies that plague renewable 227 energy remain, notably widespread corruption. The estimated coefficients of the corruption control indicator in Table 4 illustrate this situation, which is significantly 228 positive, indicating that anti-corruption is beneficial to increase the share of renewable energy consumption. The effective path is a necessary condition for the 229 existence of mediating effect. More specifically, at the 1% significance level, one unit increase in the corruption control indicator will lead to a 25% increase in the 230 share of renewable energy consumption. 231 In considering the direct effect of corruption control on carbon emissions, both the independent variable and the mediator should be introduced into 232 the econometric model. The estimated coefficients in Table 5 can show whether paths ′ and is effective. Consistent with the previous estimation parameters, the 233 coefficient of meets the significance requirement of 1%, and the value is as negative as expected. However, for different indicators of carbon emissions, the 234 estimated parameters are slightly different. When the carbon emission indicator is the total carbon emissions, the estimated coefficient is -0.3321, that is to say, one 235 unit increase of corruption control indicator will reduce the total carbon emissions by about 33.2%, and the role of anti-corruption in reducing emissions has been 236 confirmed again. In comparison, if the total carbon emissions are replaced by the carbon emissions per capita, the sign and significance level of the estimation 237 coefficient do not change, but only the value of the parameters to be estimated. The estimated coefficient of is reduced to 0.1052, which indicates that the 238 increase of corruption control index of a unit reduces the per capita carbon emission by about 10.5%. The same situation occurs when the carbon emission index is 239 replaced by , which further confirms the significance of the direct effect. With regard to path , regardless of whether the dependent variable is , or 240 , the estimated coefficients of are all significantly negative, indicating that increasing the share of renewable energy consumption is conducive to the 241 realization of emission reduction targets. In addition, it is necessary to mention that this paper introduces a number of control variables in the model, most of their 242 estimated coefficients are in line with expectations, but the sign of GDP per capita in the intensity model is contrary to expectations, which may be related to the 243 environmental Kuznets curve. 244 245 Table 6 Results of the Sobel-Goodman mediation tests 246 247 Since the path , and ′ are significantly effective, the Sobel-Goodman mediation test presented in Table 6 can be further used to characterize the mediating 248 effect, because there are requirements not only for sign of the direct effect, but also for significance of the direct effect. As far as is concerned, the statistics of 249 Sobel, Goodman-1 and Goodman-2 all meet the significance requirement at 1%, which confirms the feasibility of renewable energy consumption share as the 250 intermediary of corruption control and total carbon emissions. The specific value of the indirect effect is -0.0587, that is to say, the increase of corruption control 251 indicator by one unit not only reduces the total carbon emissions by 33.2% directly, but also reduces the total carbon emissions by 5.87% indirectly through the 252 intermediary role of the share of renewable energy consumption in the total final energy consumption. When the dependent variable is replaced by , the significant 253 indirect effect also indicates as a mediator. Furthermore, the indirect effect coefficient of carbon emissions per capita shows that the mediator contributes about 254 4.7% to the decrease of carbon emissions per capita. Once the dependent variable is replaced by , not only the mediating effect is supported by all statistics, but 255 also its contribution to carbon emission intensity is roughly equal to the result in the model. Briefly, the mediating effect of renewable energy on corruption 256 control has been confirmed in all aspects, from which it is not difficult to find that anti-corruption can remove the obstacles of clean-oriented energy mix and energy 257 transition and play a positive role in carbon emission reduction. 258

Robustness checks 259
In order to get robustness, this paper selects total carbon emissions, per capita carbon emissions and carbon emission intensity as the objects to be investigated, which 260 can ensure the reliability to a certain extent. However, a situation that cannot be ignored is that in almost all empirical analysis cases, endogeneity will appear in various 261 forms, including simultaneous causality, omitted variables and measurement errors (Bound et al., 2001;Wooldridge, 2002 ;Wooldridge, 2006), so it is necessary to 262 overcome endogeneity to obtain robustness. The usual practice is to select the appropriate instrumental variables(IV) to eliminate the endogeneity problem. Table 7  263 reports the regression results of fixed-effects(FE) and fixed-effects IV(FE-IV) method. Although the estimated coefficients are different from the previous ones 264 mentioned above, there are no significant differences in sign and significance. The two main variables, and , are negatively correlated with carbon emission 265 indicators, which further shows that anti-corruption and renewable-led energy transition can promote emission reduction. We selected the lag term of corruption control 266 indicator as its instrumental variable to examine the robustness of path ′ and . The results of the last three columns in Table 6 show that there is no obvious mutation 267 in sign and significance for and , and the direct effect is statistically significant. Furthermore, considering the outstanding heterogeneity between high-268 income countries and low-and middle-income countries, we divide the study sample into high-income groups and low-and middle-income groups for group regression, 269 and the results are shown in Table 8. An obvious difference is that the estimation coefficient of the high-income countries does not meet the significance  Another feature is the universal role played by renewable-led energy transition for carbon emission reduction. Different from the direct effect of corruption control 279 on the atmospheric environment measured by carbon emissions, which shows the deviation between high-income countries and low-and middle-income countries, the 280 emission reduction effect of increasing the share of renewable energy consumption has universal significance. For high-income countries, the estimation coefficient 281 shows that, on average, a 1% increase in the share of renewable energy consumption can reduce the total carbon emissions, carbon emissions per capita and carbon 282 emission intensity of these countries by about 0.12%, 0.11% and 0.10% respectively. Comparably, taking into consideration of low-and middle-income countries, including controlling total energy consumption, increasing the supply of non-fossil energy, and promoting energy conservation and emission reduction as the main 290 measures, making outstanding contributions to the global energy transition. 291

Conclusions and policy implications 292
This paper investigates the environmental effects of anti-corruption and the role of renewable energy as an mediator between corruption control and carbon emissions 293 in this process of energy transition and carbon neutralization. The dependent variables of environmental quality are measured by three carbon emission indicators 294 originating from the database of World Bank, which are total carbon emissions( ), carbon emissions per capita( )and carbon emission intensity( ). The 295 independent variable is the corruption control indicator, while the mediator is the share of renewable energy consumption in total final energy consumption. The sample 296 data is made up of 98 countries from 1996 to 2015, covering major developed and developing countries, and is therefore adequately representative. In view of the past 297 experience in the relationship between anti-corruption, energy transition and environmental pollution in various countries, the energy and environment sectors are the 298 hotbeds for corruption (Kolstad et al.,2008), especially for low-and middle-income countries, corruption paradise and pollution haven are often accompanied by each 299 other. Corruption has both direct effects and indirect effects mediated by renewable energy on environmental pollution, and the significance of these effects is precisely 300 the purpose of this study using the mediation effect model. 301 Some meaningful findings can be obtained from the empirical analysis of this paper as follows: First of all, anti-corruption can restrain carbon emissions to a certain 302 extent, especially for low -and middle-income countries. Due to the long-term implementation of deindustrialization, a large number of high polluting enterprises are 303 transferred and dispersed to developing countries and emerging markets with low intensity of environmental regulation, and the rent-seeking space of high-income 304 countries is greatly compressed, thus the corruption problem is relatively light. The relationship between corruption and carbon emissions in these countries is less 305 prominent than that in developing countries. Whether it is corruption or environmental pollution, developing countries are more serious, and even some problems have 306 become a persistent disease hindering sustainable development, because their pollution haven is often shaped by corruption with embeddedness of high pollution 307 enterprises (Damania, 2002;Barassi and Zhou, 2012;Egger and Winner, 2005;Sarmidi, et al., 2015). In the developing and emerging economies, only a few countries, 308 such as China, have taken measures to break the link between corruption and pollution. The mode of co-existence of corruption and pollution is still difficult to be 309 completely changed in the short term. The experience of these countries fully shows that anti-corruption is equivalent to emission reduction. Secondly, there is sufficient 310 empirical evidence for renewable energy as a mediator between corruption control and carbon emissions. The similar indirect effects obtained from the three carbon 311 emission indicators indicate that the mediating effect of renewable-led energy transition is certain. Corruption is seen as one of the biggest threats to effective climate 312 action and a major obstacle to the global transition from fossil fuels to renewable energy (Oniango, 2020). The threat of corruption to renewable energy is not only 313 rampant in developing countries and emerging markets, but even in advanced economies such as Italy (Gennaioli and Tavoni, 2016) and Iceland (Gisladottir et al., 314 2020), the risk of corruption in the renewable energy sector is difficult to avoid. Additionally, only when the total, direct and indirect effects are significant and the 315 corresponding numerical conditions are met, can the conduction paths of corruption control, energy transition and carbon emissions be considered to be unimpeded. In 316 addition to the Sobel-Goodman mediation tests of three carbon emission indicators, the estimation method that introduce instrumental variables to eliminate 317 endogeneity may be beneficial to supporting the mediating effect of renewable energy between corruption and carbon emissions. 318 These findings imply that leaving corruption monitors absent increases the risk of corruption in both the non-renewable and renewable resource sectors, thereby 319 hampering sustainable management of renewable resources and the environment. Anti-corruption policies need to be adequately enforced during the energy transition. 320 For countries with abundant renewable natural resources, measures should also be taken to address the underlying situation of current mismanagement. Relevant policy 321 optimizations should include collaboration between government agencies, businesses and local resource users and other actors to avoid collusive behavior by emitters. 322 For developing and emerging markets, it is necessary to improve market access for resource-intensive businesses without interference from interest group lobbying and 323 bribery, and to insist on higher standards of environmental regulation. Give up lax environmental policies and fill the gap in environmental regulation, regardless of 324 short-term monetary losses. In general, the affected countries will have to transform the functions of government, improve the institutional framework of integrity, 325 strengthen the capacity of the supervision institutions, and develop reasonable and monitoring mechanisms to curb the breeding and spread of corruption, so as to 326 improve the quality of environmental policies. 327

Declarations 328
Ethics approval and consent to participate: The present study work was not conducted on human or experimental animals where national or international guidelines 329 are used for the protection of human subjects and animal welfare. Hence, ethics approval and consent to participate are not applicable.    Note:***, **, and * refer to significance at 1%, 5%, and 10% confidence level, respectively 468 469