Identifying Economic Factors of Renewable Energy Consumption—A Global Perspective

Abstract

This study aims to identify the factors most likely to affect renewable energy consumption (REC) across mostly homogenous country groups worldwide. Classifying countries into a relatively homogenous group is taken from their economic and social development level measured with the Human Development Index. We delimited highly, medium-, and low-developed countries and checked whether the sets of determinants for using renewable energy sources are the same. We constructed a panel dataset as a basis for the panel Bayesian model averaging (panel BMA) as a factor selection method. The most likely factors were found and compared between the groups of countries. Then, the panel fixed-effects models for each country group were estimated. The results allowed us to confirm that CO₂ per capita emissions, terms of trade, GDP, foreign direct investment, crude oil price, and energy consumption from alternative sources are the most critical drivers of REC in group I. The most important factors in group II are CO₂ per capita, labor force, forest area, and gas and coal consumption. In the third group, REC consumption differs from that of the more advanced groups and strongly depends on foreign direct investment inflow. The results allow the formulation of policy recommendations on a global scale.

1. Introduction

Environmental damage, biodiversity loss, global warming, rapid population growth, increased fuel costs, military and geopolitical conflicts, and their eventual impact on all other economic sectors are the leading causes of extracting energy from renewable sources. Renewable energy consumption, or REC for short, has increased from 2011 to 2021 in the following scale: 8.3% in Europe, 29.4% in the Commonwealth of Independent States (CIS), 37.3% in Middle Eastern countries, 20.5% in Africa, 8.8% in North America, 20.5% in Asia–Pacific, and 10.8% in South and Central America. Conversely, the increase in the world’s energy consumption from non-renewable sources was only 1.25 percent. It showed negative growth in the European Union (EU), Europe, and Euro-Asian countries (−1.7%, −0.9%, and −0.6%, respectively), as well as relatively minor increases in locations like Africa (2.9%) and the Middle East (3.6%). Moreover, the increase in primary energy in 2021 was driven by emerging economies, with China being the leader in growth [1]. These facts motivate the studies that seek relationships between renewable energy consumption and its stimulants and de-stimulants across countries. Finding the factors that increase or decrease the ratio of REC in total energy consumption could help governments stimulate investments in renewable energy sources and create an appropriate climate policy. This knowledge is also essential for investors who can identify the investment potential and needs of countries and/or regions.

The current study examines economic-, energy-, and environmental-based determinants of renewable energy consumption from a global perspective. The main objective of the current study is to identify the most influential REC determinants, which are crucial for technologists, economists, and policymakers. Accurately identifying REC determinants is essential in developing a suitable policy mechanism to address the significant energy demand. Furthermore, it explores ways to reduce our long-term reliance on fossil fuels to achieve the Sustainable Development Goals (SDGs). Moreover, the appropriate policy gives momentum to control carbon emissions and achieve climate change targets under the SDGs. The creation of low greenhouse gas emissions is closely tied to the objectives of the Paris Agreement, and this can be accomplished by altering the structure of energy production and consumption.

The motivation behind considering subgroups of countries classified according to the level of social and economic development is the technological frontier, which affects technology absorption and efficient use across countries. The drivers for catching up are human and physical capital acquisition, socio-institutional factors, knowledge development, infrastructural bottlenecks, product diversification from low- to high-value-added activities, and participation in global and regional value chains [2]. Countries across the globe are also diversified in their exploitation of renewable energy sources. For example, the share of REC in the energy mix is diversified across countries and regions, being, in 2021, highest in Asia–Pacific (43.1%), Europe (25.4%), and North America (21.2%), and lowest in the CIS (0.3%), Middle East (0.4%), and Africa (1.2%) [1].

Therefore, the research problem discussed in this study refers to the differences between the key factors affecting energy consumption from renewable sources identified for different groups of countries. The analysis includes the energy consumption structure as determined by its sources. Initially, all countries worldwide were considered for this study, but data limitations corrected this scope. Eventually, 109 countries observed from 1995 to 2019 were considered. The countries were divided into subgroups according to the Human Development Index values reported in the last year of the sample. The Human Development Index is one of the crucial indicators demonstrating how countries could combine economic performance and social development. The HDI helped to identify 46 countries with “very high” HDIs (group I), 16 countries with “high” HDIs (group II), and 47 countries with “medium and low” HDIs (group III) [3]. The current study merged the medium and low HDI countries into one group. The HDI level groups do not correspond directly to the geographical regions but are related to these regions, with some exceptions.

A comprehensive methodology that integrates economic, environmental, and multiple REC factors forms the foundation of the current research. The panel BMA method serves as the first-tier methodological framework. When there are many possible variables, BMA’s primary benefit is that it ranks each component based on probability. Moreover, it guarantees comparable findings and recommends the model parameters most likely out of various competing ones [4]. The second tier is related to the panel econometric model construction. The panel fixed-effect models were estimated to extract the essential energy consumption factors from renewable sources. The relationships demonstrate differences and similarities between the countries’ subgroups. Such an approach is uncommon in the literature, mainly because of the panel BMA application. Combining the Bayesian approach relying on posterior probability with a classic one (panel FE models) has an advantage that justifies the results from two perspectives, recognized in contemporary statistics and econometrics.

The novelty of the current research lies in the applied methodology and in directly comparing the renewable energy consumption factors by focusing on HDI groups in REC across the globe and identifying the most likely factors affecting renewable energy consumption across mostly homogenous country groups worldwide. However, the subject’s popularity in the literature inevitably confirmed that the Paris Pact increased global awareness of climate change and its consequences. It aligns with the results obtained by [5]. They suggested that environmental concern is essential in explaining renewables participation in different countries.

The outcomes of this study indicate that CO₂ emissions, measured per inhabitant, are crucial in implementing and increasing energy use from renewable sources. This factor was confirmed in countries included in groups I and II. This is in line with studies that confirmed the awareness of the impact of greenhouse gases on climate change, such as [5]. The results also revealed a similarity of factors determining REC in country groups I and II. Group III, which is less developed, remains different and requires special attention.

The remainder of this paper is organized as follows: Section 2 reports the relevant literature review. Section 3 provides data characteristics and methodology. Section 4 presents and discusses the empirical results, and Section 5 concludes.

2. Literature Review

In the literature, researchers explored the influencing factors of REC, such as income, energy prices, environmental degradation, energy security, energy consumption, political environment, renewable energy potential, international flows, financial development, and regulatory systems. The research reported in [6,7] provided a robust framework for subsequent studies that examined the factors influencing REC. Most studies initially just regarded income and prices as primary factors of REC [8,9,10,11,12,13,14,15,16].

Several approaches tackle energy issues, and there is a plethora of literature on the causality between aggregate and disaggregated energy consumption from various sources (including renewable energy) and economic growth [6,17,18,19]. On the other hand, several studies in the economic literature have focused on key economic factors in increasing REC [20,21,22,23,24,25,26,27,28]. However, only a few studies have objectively investigated the factors determining renewable energy generation [5,8,29,30,31,32,33]. Recently, researchers have introduced environmental variables to explore the relationship between REC and economic development [10,11,13,14,16,17].

Ref. [9] included potential RE sources and the regulatory framework for fossil fuels in the list of possible determinants of REC in the existing literature using panel data from 46 industrialized and developing nations between 1980 and 2011. The study analyzed the primary short- and long-term drivers of the percentage of non-hydro renewable energy sources using panel cointegration estimation approaches. The study’s findings demonstrate that many driving factors have varying effects on the amount and proportion of non-hydro renewable energy, most of which have long-term effects. Increases in oil prices and financial development significantly impact the quantity and percentage of electricity generated by RE sources other than hydropower. Trade openness has an ambiguous effect. On the other hand, energy generation from non-hydro renewable sources is not significantly impacted by resource rent, gross fixed capital formation, ratification of the Kyoto Protocol, or foreign direct investments.

Considering the countries and regions, the literature focuses on both developing and developed ones. The multi-country approach is also very popular.

Referring to developing countries, Ref. [10] investigated the relationship between GDP, CO₂ emissions, and REC using panel data for 24 Asian nations between 1990 and 2012. This study discovers a long-run equilibrium between the variables. RE use is positively impacted by CO₂ emissions in China, the Philippines, Pakistan, Yemen, Iraq, and Saudi Arabia. The GDP of India, Sri Lanka, the Philippines, Thailand, Turkey, Malaysia, Jordan, the United Arab Emirates, Saudi Arabia, and Mongolia significantly impacts renewable energy. Two bidirectional causal linkages exist between CO₂ emissions and REC and between GDP and RCE.

Ref. [20] analyzed the factors that influence the use of REC in the five biggest African nations—Nigeria, Egypt, Ethiopia, the Democratic Republic of Congo, and South Africa—using the data from 1996 to 2016. The researchers included macroeconomic, socioeconomic, and institutional variables in their investigation. The study applied the BMA (Bayesian model averaging) technique. The empirical findings depict that an increase in any of these determinants (population growth, urban population, energy demand/use, electricity power demand/consumption) causes an increase in REC. Ref. [25] also consider African countries in the context of REC.

Ref. [34] examined the factors influencing REC in a panel of six significant emerging economies: Brazil, China, India, Indonesia, the Philippines, and Turkey. This paper determines the long-term relationships between REC and pollutant emissions and income. These relationships are significant in Brazil, China, India, and Indonesia, while income plays a significant role in the Philippines and Turkey. Short-term bidirectional causal links have been identified between renewable energy and income and between renewable energy and pollution emissions. These findings imply that the initiatives made by developing nations to lower their carbon footprint by boosting energy efficiency and significantly raising the proportion of renewable energy sources in their total energy mix are justified.

Ref. [35] found that innovation contributed to a rise in REC, while CO₂ emissions, real GDP per capita, and economic freedom caused a decrease in REC in the ASEAN+3 economies between 1998 and 2018. Ref. [36] investigated the factors influencing the adoption of renewable energy in Africa using data from 1990 to 2019. The study confirmed the feedback hypothesis, which examined the tripartite economic, environmental, and sociopolitical effects of adopting renewable energy in Africa. The study found a positive correlation between the adoption of renewable energy and economic factors. The findings also demonstrated how adopting RE in Sub-Saharan African economies adversely affected environmental issues, including carbon emissions and ecological footprints. Some researchers have recently focused on reducing CO₂ emissions while growing the renewable energy industry without slowing economic growth [37,38,39].

It is widely acknowledged that renewable energy sources may replace crude oil in energy production and consumption. It is also anticipated that there will be a negative correlation between the demand for renewable energy and oil prices, as higher oil prices would incentivize people and companies to switch to renewable energy sources, lower their consumption, and buy more energy-efficient products [40]. Refs. [6,34,40] investigated the impact of oil prices on REC and found some evidence of significant effects on oil prices. The relationship between economic growth and REC is currently receiving much attention in the literature, e.g., [6,18,41,42,43].

Ref. [44] examined the impact of both renewable and non-renewable energy resources on greenhouse gas emissions. The panel data from 25 emerging Asian nations from 2000 to 2016 were taken. The study found a short- and long-term positive correlation between economic growth and renewable energy and suggested a feedback hypothesis. Conversely, the study observed that a 1% increase in renewable energy causes a 0.193% decline in carbon emissions. The results show that non-renewable energy resources significantly contribute to greenhouse gas emissions, while renewable resources have a favorable effect on reducing greenhouse gas emissions. This report also emphasizes how emerging Asian economies may protect the environment by utilizing renewable energy sources and stronger regional environmental regulations.

Ref. [45] analyzed the relationship between energy use and economic growth for the BRICS nations between 1990 and 2012 using a multivariate panel approach. The findings suggest that economic growth is the critical factor driving rising energy consumption in the BRICS nations.

To address environmental sustainability, Ref. [46] investigated the impacts of trade and natural resources on REC for Sub-Saharan African countries spanning 1990–2015. According to the findings, trade openness substantially negatively influences the amount of REC in the region. On the other hand, increased FDI and income levels were found to have a substantial positive impact on REC.

Trade openness’s impact on REC has received less attention than the relationship between trade openness and energy demand [47,48,49,50,51]. Trade expansion fosters economic growth, enabling the development and application of more energy-efficient technologies, leading to the expectation that trade openness will enhance renewable energy. A high level of trade openness can increase national income, giving affected countries more significant financial resources to create renewable energy that meets the demand for improved environmental quality. Trade openness is critical to economic development because it expands the domestic market for export-focused industries, improves bilateral trade relations, and provides access to energy-efficient technologies. Furthermore, economic activity has developed, and energy consumption in various channels (e.g., meeting the international requirements for surplus production and importing more machinery and equipment) has increased energy demand in South America [51]. Ref. [52] investigated the relationship between energy consumption and trade openness in four developing and oil-importing nations: Bangladesh, India, China, and Pakistan. The study observed a unidirectional causal relationship between trade openness and energy consumption. Summing up, REC in developing countries mainly depends on greenhouse gas emissions, income, oil price changes, population growth, and FDI inflow.

Different perspectives were considered when referring to developed countries. Ref. [6] investigates the determinants of REC for G7 countries and finds that, in the long run, real GDP per capita and CO₂ per capita emissions are a major driving force that fluctuates per capita REC. The usage of renewable energy is negatively impacted by increases in oil prices, but less so. Ref. [53] examine how social and macroeconomic factors affected REC in the G7 between 1990 and 2014. The study empirically supports the positive and significant effects of energy imports, the Human Development Index, and research investment (% of GDP) on REC.

Ref. [54] illustrated empirical work, using growth and environmental functions, on how renewable energy can effectively spur economic growth and reduce carbon emissions in the context of 15 major developed countries that consume renewable energy. The findings demonstrate how effective renewable energy is at boosting economic growth and lowering carbon emissions. Ref. [55] used the panel fixed-effects model to analyze 28 EU countries from 1990 to 2015. Their findings demonstrated the impact of financial development on REC. They confirmed that energy prices, the banking sector, bonds, and capital markets increased the share of REC, whereas FDI was statistically insignificant.

Ref. [56] explored the case of OECD economies for data on the factors influencing REC from 1990 to 2017. The findings indicate that several factors, including income, human capital, energy productivity, energy pricing, and eco-innovation, can explain REC. Ref. [57] also examined the relationships between capital, labor, international trade, and non-renewable and renewable energy. The study concludes that trade between the EU-15 countries increases renewable energy use.

Ref. [58] use panel estimation techniques robust to cross-sectional dependence to examine the effects of real income, trade openness, and renewable and non-renewable energy on CO₂ emissions in the Environmental Kuznets Curve (EKC) model for the European Union from 1980 to 2012. The study demonstrated that trade and renewable energy reduce carbon emissions while non-renewable energy increases CO₂ emissions, supporting the EKC hypothesis using the dynamic ordinary least squares estimator.

Ref. [59] examined the most likely factors influencing REC in European countries. The study discusses various institutional, social, and economic factors that impacted REC and renewable energy policies in selected European nations before and after the Paris Agreement. To determine the most likely factors affecting REC in 2015 and 2018, the Bayesian model averaging (BMA) is utilized. The comparison results show that the GDP level and the nuclear and hydro energy consumption were the key drivers in both studied years. It also became evident that, in contrast to 2018, when trade openness and FDI contributed to an increase in REC, REC in 2015 heavily depended on the structure of energy consumption. Compared to developing countries, the factors of REC in developed ones are often similar, e.g., oil prices, CO₂ emissions, and real GDP. However, financial innovations like green finance instruments (bonds and other capital market instruments) are also highly important.

Another stream of the literature refers to the multi-country perspective. They are typically divided according to high, medium, and low income. For example, the study of [60] considered a global panel of countries divided into high, middle, and low income, but it only refers to 64 countries. Also, Ref. [61] examined the long-term, bidirectional relationship between GDP growth and REC in high-income, upper-middle-income, and lower-middle-income nations. The findings are inconsistent, but higher-income countries are those where the bidirectional long-term association between the variables is more significant and stable. Ref. [62] investigated the relationship between energy demand and foreign direct investment (FDI). The study found that FDI increased renewable energy sources while decreasing non-renewable energy consumption in 74 countries between 1985 and 2012. The study breaks down FDI inflows into mining, manufacturing, financial services, and total services. The findings generally suggest that energy consumption is reduced regarding non-renewable energy sources and increased regarding renewable energy. Ref. [63] examines the relationship between output, FDI inflows, and energy consumption in 75 countries between 1990 and 2010. The current findings show that, in the three categories of countries (developed, all, and developing), there is evidence of a bidirectional relationship between FDI and output per capita, REC and GDP per capita, and non-renewable energy and GDP per capita. Ref. [64] analyzed low-income countries, middle-income, and upper-middle-income panels and revealed the long-run asymmetric correlations between financial development, trade openness, capital flows, and REC. The asymmetric relationship is also confirmed in the near run, except for lower-income nations. The results showed that long-term causation was present, especially when REC was included in the equation as a dependent variable. Ref. [65] examined the relationships between RCE, CO₂ emissions, FDI, and economic growth in 190 countries from 1980 to 2018. The study used static and dynamic models, showing that carbon emissions, REC, FDI, and economic growth significantly impact one another. On the other hand, REC has been found to improve environmental quality and reduce FDI inflow.

The set of variables used typically to analyze renewable energy consumption is similar across different studies and constitutes a basis for further development and comparison.

3. Materials and Methods

This section is divided into two data characteristics and methodology subsections, primarily based on the panel Bayesian model averaging and the fixed-effects model. The selection of variables is based on both the principles of environmental economics [12] and a review of the empirical literature [5,40]. The chosen variables, GDP, oil price, and oil consumption, were utilized by [35,36], while foreign direct investment and net inflows (% of GDP) were adopted from [51]. BMA is a well-known approach to identify energy consumption [21,27], which aims to rank variables according to their relationship with the dependent variable.

3.1. Data Characteristics

The current study utilizes panel data on REC and its determinants across 109 selected countries from 1995 to 2019. This overall energy consumption includes electricity, heating and cooling, and transport.

Table A1. The list of selected countries ranked 1–66 according to their HDI.

Country	Code	Country	Code	Country	Code
United Arab Emirates	ARE	France	FRA	Malaysia	MYS
Argentina	ARG	United Kingdom	GBR	Netherlands	NLD
Australia	AUS	Greece	GRC	Norway	NOR
Austria	AUT	Hong Kong SAR, China	HKG	New Zealand	NZL
Belgium	BEL	Croatia	HRV	Poland	POL
Belarus	BLR	Hungary	HUN	Portugal	PRT
Canada	CAN	Ireland	IRL	Qatar	QAT
Switzerland	CHE	Iceland	ISL	Romania	ROM
Chile	CHL	Israel	ISR	Russian Federation	RUS
Czech Republic	CZE	Italy	ITA	Singapore	SGP
Germany	DEU	Japan	JPN	Slovak Republic	SVK
Denmark	DNK	Lithuania	LTU	Slovenia	SVN
Spain	ESP	Luxembourg	LUX	Sweden	SWE
Estonia	EST	Latvia	LVA	Thailand	THA
Finland	FIN	Mauritius	MUS	Turkiye	TUR
-	-	-	-	United States of America	USA

,

Table A2. The list of selected countries ranked 67–115 according to their HDI.

Country	Code	Country	Code	Country	Code
Bulgaria	BGR	Peru	PER	Ecuador	ECU
Sri Lanka	LKA	Mexico	MEX	Egypt	EGY
Ukraine	UKR	Brazil	BRA	South Africa	ZAF
North Macedonia	MKD	Colombia	COL	Jamaica	JAM
China	CHN	Algeria	DZA	Indonesia	IDN
-	-	-	-	Vietnam	VNM

and

Table A3. The list of selected countries ranked 116–191 according to their HDI.

Country	Code	Country	Code	Country	Code
Philippines	PHL	Comoros	COM	Eritrea	ERI
Venezuela	VEN	Papua New Guinea	PNG	Guinea	GIN
Iraq	IRQ	El Salvador	SLV	Liberia	LBR
Morocco	MAR	Pakistan	PAK	Congo, Dem. Rep.	COD
Bangladesh	BGD	Haiti	HTI	Afghanistan	AFG
India	IND	Nigeria	NGA	Sierra Leone	SLE
Nepal	NPL	Rwanda	RWA	Guinea-Bissau	GNB
Equatorial Guinea	GNQ	Benin	BEN	Yemen	YEM
Cambodia	KHM	Lesotho	LSO	Burkina Faso	BFA
Angola	AGO	Malawi	MWI	Mozambique	MOZ
Myanmar	MMR	Senegal	SEN	Mali	MLI
Syrian Arab Republic	SYR	Djibouti	DJI	Burundi	BDI
Cameroon	CMR	Sudan	SDN	Central African Republic	CAF
Kenya	KEN	Madagascar	MDG	Niger	NER
Congo	COG	Gambia	GMB	Chad	TCD
Solomon Islands	SLB	Ethiopia	ETH	-	-

in Appendix A provide a list of all the countries included in the analysis. We use the Human Development Index (HDI) to classify countries with “very high” HDI, “high” HDI, and “medium and low” HDI.

It is well known that a growth and development measure is a crucial tool for economists, energy and environmental experts, ecologists, and policymakers. Therefore, to evaluate national socioeconomic development, the United Nations Development Programme (UNDP) created the Human Development Index (HDI) in 1990 and established a new methodology in 2010 [66]. It prioritizes development over gross domestic product (GDP) when assessing living standards. Therefore, the HDI has completed GDP as a different metric to gauge a nation’s development [67]. Increasing human development is vital for developing a nation [68]. However, human development and economic growth have an inevitable bidirectional relationship. On the one hand, economic growth provides a means of promoting progress in human development. On the contrary, it enhances human development (well-being, health, education, etc.). Compared to economic statistics like GDP per capita, the HDI offers a more thorough assessment of human development. It considers health and education critical to human well-being, wealth, and economic progress. Human development levels across nations can be compared using the HDI.

For this reason, we use this index to illustrate the impact of factors on renewable energy consumption by country at different stages of development. The ranking of countries based on the HDI was used to set the countries’ groups. Specifically, countries ranked from 1 to 66 in 2021 are designated as having a “very high” HDI (group I), those ranked from 67 to 115 are defined as having a “high” HDI (group II), and those ranked from 116 to 191 are denoted as having a “low and medium” HDI (group III). Following this simple classification, we have 46 countries ranked with a “very high” HDI, 16 countries ranked 67 to 115 (with a “high” HDI), and 47 countries ranked 116 to 191 (with medium and low HDI). Due to the lack of data, some countries from groups II and III are excluded from the study.

As the HDI levels fully define the economic and social characteristics of the countries and regions, the energetic characteristics are very interesting and diversified. On the one hand, the increase in the consumption of energy from renewable sources is observed. The facts and figures show that the growth rate of renewable energy consumption was 8.8% in North American countries from 2011 to 2021. Similarly, there was a 10.8% enhancement in REC in South and Central America. In Europe, the growth in REC at the same time was 8.3%. The insight of African countries shows a 20.5% growth rate in REC. The Asian–Pacific countries attained a 20.5% growth rate in REC [1]. On the other hand, the energy mix structure for the total energy supply brings quite a new perspective. It should be emphasized that countries qualified to be in the first group, according to the HDI (1–66), are located mainly in Europe, North America, Eurasia, the Middle East, and Asia–Pacific. According to the data provided by the International Energy Agency (www.iea.org), in 2021, the energy mix for developed countries consisted of coal 13.1%, oil 30.5%, natural gas 26.7%, nuclear 12.0%, and biofuels and waste 10.2%, with the remaining share related to hydro energy and wind and solar energy—for Europe; coal 10.4%, oil 35.9%, natural gas 35.0%, nuclear 9.2%, and the remaining share related to biofuels and waste, hydro energy, and wind and solar energy—for North America; coal 16.5%, oil 19.6%, natural gas 54.5%, and nuclear 5.8%, with the remaining share related to biofuels and waste, hydro energy, and wind and solar energy—for Eurasia; coal 0.5%, oil 40.2%, natural gas 58.3%, and nuclear 0.4%, with only 1% in share related to biofuels and waste, hydro energy, and wind and solar energy in the Middle East; and finally, coal 48.3%, oil 23.4%, natural gas 11.6%, hydro energy 2.4%, and the remaining share is for biofuels and waste, nuclear, and wind and solar energy—for Asia–Pacific. The second group of countries (67–115) consists partially of less-developed European countries (Bulgaria, Ukraine, North Macedonia), some Asian (China) and South American countries, and three countries located in Africa. The energy mix for Central and South America consists of coal 4.9%, oil 39.6%, natural gas 21.0%, biofuels and waste 21.9%, hydro energy 8.9%, and the remaining share relates to nuclear energy and wind and solar energy. The third group—the middle and low HDI countries (116–191)—are primarily associated with Africa, apart from Algeria, Egypt, and South Africa, which are included in group II. The energy mix there is as follows: coal 12.3%, oil 22.9%, natural gas 16.5%, biofuels and waste 45.5%, and the remaining share is related to hydro energy and wind and solar energy. The analysis of the energy mix across the groups of countries implies that traditionally industrialized countries have a lower share of renewables in the energy mix than developing countries. Additionally, countries grouped according to the HDI level have similar energy characteristics within the groups.

Figure A1, Figure A2, Figure A3, Figure A4, Figure A5 and Figure A6 (in Appendix A) present a detailed view of how renewable energy consumption (REC) and CO₂ emissions per capita (CO₂pc) vary across different groups of countries with ”very high”, ”high”, and ”medium and low” HDI. We notice substantial variation in renewable energy consumption across the countries in all three groups. Some countries have a wide range of REC values, while others have a much narrower range, presented in the left-hand panel. The right-hand panel plots show a steady increase in the mean renewable energy consumption for countries with a ”very high” HDI, while in the other groups (“high” and “medium and low”), we can observe a decrease. The situation is reversed in the case of CO₂ emissions. Among the group of countries with the highest HDI, we observe a systematic decrease from 2015, while in the other groups, the volume of emissions has even increased.

Additionally, the sample GINI coefficient was calculated for REC and CO₂pc variables to illustrate the scale of inequality concerning the variables of interest in each country group. Comparing the values for groups I and II, we can notice more diversity in REC consumption (0.48 and 0.43, respectively) than in CO₂pc emission (0.34 and 0.35, respectively). That means the emission levels are similar, but countries still need to implement more energy from renewable sources. On the contrary, in group III, the countries are much more diversified by CO₂pc emission (0.67) and indirectly in their development than in REC (0.25), which remains very low.

The study draws on data from various sources, including the World Development Indicators (WDI–2019) and Statistical Review of World Energy (BP–2019). Below is a list of all the variables considered in our study, with corresponding symbols provided in parentheses:

• Economic variables: gross domestic product (GDP), gross domestic product per capita (GDPpc), FDI net inflow (FDI/GDP), inflation (INF), trade openness (TO), adjusted savings: energy depletion (ADSED), the labor force (LF);
• Energy-based variables: oil (OC), coal (CC), gas (GC), nuclear (NC), hydro (HC), renewable energy consumption (% of total final energy consumption, REC), crude oil price (COP);
• Environmental: CO₂ emission per capita (CO₂pc);
• Others: forest area (FA).

Detailed descriptions of all variables and their units are provided in

Table A4. Variable Descriptions.

No.	Variable Abbreviation	Variable Name	Proxy/Scale of Measurement	Data Source
			Energy-based variables
1	REC	Renewable Consumption	Renewable energy consumption (% of total final energy consumption)	BP-2019
2	OC	Oil consumption	Million tonnes of oil equivalent to exajoules (Mtoe)	BP-2019
3	GC	Gas consumption	Million tonnes of oil equivalent to exajoules (Mtoe)	BP-2019
4	CC	Coal consumption	Million tonnes of oil equivalent to exajoules (Mtoe)	BP-2019
5	HC	Hydro consumption	Million tonnes of oil equivalent to exajoules (Mtoe)	BP-2019
6	NC	Nuclear consumption	Million tonnes of oil equivalent to exajoules (Mtoe)	BP-2019
7	COP	Crude oil price	Crude oil price (US dollars per barrel)	BP-2019
			Economic variables
8	GDP	Gross domestic product	Constant 2010 US dollar	WDI-2019
9	GDPpc	Per capita gross domestic product	Constant 2010 US dollar	WDI-2019
10	TO	Trade openness	Trade openness = Exports of goods and services (% of GDP) + Imports of goods and services (% of GDP).	WDI-2019
11	ADSED	Adjusted savings: energy depletion	(current US dollar)	WDI-2019
12	LF	Labor force	Total labor force	WDI-2019
13	FDI/GDP	Foreign direct investment, net inflows	(% of GDP)	WDI-2019
14	INF	Inflation, GDP deflator	(annual %)
			Environmental Variables
15	CO2pc	CO2 emissions	(metric tons per capita)	WDI-2019
			Other variables
16	FA	Forest area	Square. Km	WDI-2019

in Appendix A.

Table A5. Descriptive statistics of variables under study in groups of countries.

Summary Statistics for Ranked 1–66
Variable	REC	TO	ADSED	CO2pc	CC	COP	FA	FDI/GDP	GC	GDP	GDPpc	HC	INF	LF	NC	OC
Mean	16	105	5,250,000,000	9	24	55	416,874	5	37	921,000,000,000	31,060	8	6	14,938,128	11	52
Med	11	84	197,000,000	8	4	54	34,282	3	8	252,000,000,000	31,185	1	2	4,794,475	0	14
Max	81	443	164,000,000,000	48	546	112	8,153,116	86	728	19,900,000,000,000	112,418	86	662	167,000,000	194	964
Min	0	16	0	0	0	13	0	−58	0	5,120,000,000	1886	0	−25	154,357	0	0
S.D.	15	73	16,500,000,000	7	72	32	1,334,422	10	96	2,400,000,000,000	21,794	16	25	26,266,159	31	132
Skew	2	2	6	3	6	0	5	3	5	6	1	3	18	4	4	6
Kurt	6	9	44	13	38	2	25	22	25	36	4	14	432	18	23	35
JB	876	2381	88,097	6133	63,470	85	27,737	18,971	26,451	59,601	191	8140	8,882,952	13,331	21,881	54,380
Prob.	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
Obs.	1150	1150	1150	1150	1150	1150	1150	1150	1150	1150	1150	1150	1150	1150	1150	1150
Summary Statistics for Ranked 67–115
Mean	22.21	65.72	11,700,000,000	3.24	99.16	55.17	678,820.9	3.12	19.79	694,000,000,000	4542.80	15.85	11.98	73,925,151	3.18	48.82
Med	16.64	56.69	3,260,000,000	2.52	4.79	54.19	131,335.4	2.65	6.24	151,000,000,000	4238.37	2.66	6.47	19,578,367	0.00	13.21
Max	69.94	164.70	237,000,000,000	8.44	1969.07	111.67	5,699,933.0	31.23	264.26	14,300,000,000,000	10,155.49	270.33	913.21	781,000,000	74.24	666.52
Min	0.06	15.64	0	0.32	0.00	12.72	449.80	−2.76	0.00	5,750,000,000.0	906.13	0.00	−26.30	815,445	0.00	0.00
S.D.	17.53	30.31	25,000,000,000	2.06	356.71	31.71	1,294,186.0	2.90	32.08	1,850,000,000,000.0	2175.66	40.89	50.90	179,000,000	8.02	100.39
Skew	0.80	0.92	4.97	0.76	4.29	0.41	2.82	3.92	3.74	4.95	0.63	3.94	15.24	3.45	4.66	3.93
Kurt	2.66	3.31	34.44	2.49	20.50	1.94	10.25	31.49	22.62	29.55	2.72	20.12	257.03	13.34	31.53	19.59
JB	44.54	57.83	18,116.49	43.15	6329.39	29.69	1408.15	14,550.23	7350.19	13,385.32	27.86	5918.48	1,090,975	2575.21	15,014.37	5614.66
Prob.	0	0	0	0	0	0	0	0	0	0	0.000001	0	0	0	0	0
Obs.	400	400	400	400	400	400	400	400	400	400	400	400	400	400	400	400
Summary Statistics for Ranked 116–191
Mean	61.62	66.00	1,550,000,000	0.72	5.96	55.17	142,645	3.57	2.36	63,900,000,000	1308.50	1.05	19.84	18,849,678	0.13	5.49
Med	72.71	52.73	1,967,612	0.24	0.00	54.19	55,022.90	1.51	0.00	10,900,000,000	856.25	0.00	6.33	4,570,977	0.00	0.00
Max	98.34	348.00	36,400,000,000	10.35	444.73	111.67	1,472,640	161.82	51.84	2,690,000,000,000	14,222.55	34.46	4800.53	501,000,000	9.62	244.53
Min	0.31	0.00	0.00	0.02	0.00	12.72	56.00	−11.20	0.00	0.00	0.00	0.00	−31.57	109,672	0.00	0.00
S.D.	28.76	52.94	4,460,000,000	1.31	40.97	31.68	249,621	9.23	7.87	228,000,000,000	1475.87	4.25	171.79	62,887,753	0.83	23.40
Skew	−0.71	2.92	4.42	3.78	8.02	0.41	3.18	8.71	3.72	7.48	4.51	4.85	22.46	6.26	7.97	6.74
Kurt	2.30	15.35	24.83	19.68	70.83	1.94	14.56	110.37	16.74	67.91	31.59	27.45	567.52	42.55	71.16	53.84
JB	122.81	9144.08	27,158.45	16,420.69	237,879.70	87.21	8524	579,290.50	11,947.12	217,268.80	44,016.60	33,876.73	15,700,836	84,240.46	239,899.10	135,424
Prob.	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00
Obs.	1175	1175	1175	1175	1175	1175	1175	1175	1175	1175	1175	1175	1175	1175	1175	1175

Note: Med= median; Min = minimum; Max = maximum; S.D. = standard deviation; Kurt = Kurtosis; Skew = skewness; JB = Jerqa–Berra; Prob = probability; Obs = observations.

presents descriptive statistics for the 109 selected countries from 1995 to 2019, confirming a general shift in energy consumption structure from various sources.

3.2. Methodology

An econometric panel model is a valuable tool for solving problems on an international scale; however, finding a suitable model for panel data requires testing and model comparisons. It is possible to develop and select a statistically and economically reasonable model, starting with the most straightforward construction and using the more advanced ones. The current study is based on fixed-effects (FE) models and employs the Bayesian model averaging approach for variable selection.

3.2.1. Panel Fixed-Effects Model

The panel dataset contains observations from different entities (countries) over time. This data type effectively identifies impacts not visible in cross-sectional or time-series data alone. The FE model controls the cross-sectional variations in the data and allows a separate intercept for each cross-section. The model takes the following form:

The FE model allows for an independent intercept for each cross-section and manages the cross-sectional variances in the data, according to [69]. The model has the following form:

$${REC}_{it}={\alpha }_i+\beta \mathrm{’}{x}_{it}+{\epsilon }_{it}, i=1,\dots ,N,t=1,\dots ,T,$$

(1)

where each unit (in this case, a country) has a vector of individual impacts denoted by ${\alpha }_i$, $x_{it}$ is a matrix of observed values of exogenous variables (in our study, the factors affecting REC), $\beta \mathrm{’}$ is a matrix of parameter estimates, ${\epsilon }_{it}$ is the error term, $1,\dots ,N,$ represents the units in the panel (here, countries), and $t=1,\dots ,T$ denotes the time units (in our case, years).

In our research, we also test other panel model alternatives: the random-effects model (RE), which assumes that individual effects are random, and the pooled model, which ignores all individual-specific effects [69]. All models, i.e., pooled, fixed, and random effects, can be compared and tested by using the standard $F$ test for the common intercept in all units (PM versus FE hypotheses) and Hausman’s test to decide between the fixed- and random-effects model (FE vs. RE hypotheses). In the F test, we assume that the cross-sectional units all have a common intercept in the null hypothesis. In turn, if the Hausman and Taylor test [70] rejects the null hypotheses, it implies that the fixed- and random-effects estimators are different. Therefore, fixed effects should be preferred over random effects.

3.2.2. Bayesian Model Averaging

One challenge when constructing an econometric model is identifying significant variables from a large set of potential determinants. We considered 14 potential variables in the reported study, resulting in 16,384 linear combinations. We have employed the BMA—Bayesian model averaging—method to address this issue, which can determine the significance of competing specifications across the model space (see [71]). To obtain a current summary of the BMA methodology and its applications, refer to [72]. The following are the basics of BMA that we use in our paper.

For every competitive model $M_1,\dots ,M_{2^K}$, we set the prior probability $P\left(M_1\right),\dots ,P\left(M_{2^K}\right)$, where $\sum _{i=1}^{2^K}P\left(M_i\right)=1$. Based on Bayes’ formula, the posterior probability of the model can be calculated as

$$\mathrm{Pr}\left(M_i\mid y\right)\approx {\displaystyle \frac{\mathrm{Pr}\left(M_i\right)P\left({y\mid M}_i\right)}{\sum _{i=1}^{2^K} \mathrm{Pr}\left(M_i\right)P\left({y\mid M}_i\right)}}$$

(2)

Formula (2) enables the calculation of the posterior probability for each model $M_i$ given that the density of the marginal distribution $P\left({y\mid M}_i\right)$ is known.

Another essential characteristic in model averaging is the posterior inclusion probability (PIP). The PIP represents the posterior probability that the independent variable $x_i$ is significant in explaining the dependent variable $y$ [71,72,73]. Within our framework, the PIP is computed by summing the posterior model probabilities for all models that incorporate a specific variable:

$$\mathrm{Pr}\left({\beta }_i\ne 0\mid y\right)=\sum _{i=1}^{2^K} \mathrm{Pr}\left(M_i\mid {\beta }_i\ne 0,y\right)$$

(3)

If the selected variable’s PIP value is greater than 0.7, we can consider this variable statistically robust.

4. Results

Empirical Results

To identify the determinants of renewable energy consumption (REC), we use the BMA selection procedure, which enables searching all possible combinations of potential variables and selecting the most probable candidates. We used the BMA 3.1 package [74], which is available in the Gretl program, an open-source software for econometric analysis. A simple description of how to use the BMA approach in modeling fixed-effects panel models is presented in [75]. In this study, we have 14 potential determinants of REC, so the total number of combinations equals $2^{14}=\mathrm{16,384}$. In the Monte Carlo experiment, we set 1,000,000 iterations (including 10% burn-in draws). The correlation coefficient between the analytical and numerical probabilities of the top models was above 0.99, confirming the convergence of the simulation. We set a uniform prior for all possible combinations, assuming they are equally likely.

Table 1. BMA estimates of renewable consumption determinants in 1995–2019.

GROUP I: Ranked 1–66						GROUP II: Ranked 67–115						GROUP III: Ranked 116–191
Dep. Variable: REC
Variant with GDPpc			Variant with GDP			Variant with GDPpc			Variant with GDP			Variant with GDPpc			Variant with GDP
Estimation of Country Fixed Effects						Estimation of Country Fixed Effects						Estimation of Country Fixed Effects
Name	PIP	Post. Mean	Name	PIP	Post. Mean	Name	PIP	Post. Mean	Name	PIP	Post. Mean	Name	PIP	Post. Mean	Name	PIP	Post. Mean
ADSED	0.30	7.15 × 10−12	ADSED	0.14	2.72 × 10−12	ADSED	0.07	1.07 × 10−12	ADSED	0.06	5.01 × 10−13	ADSED	1.00	2.56 × 10−10	ADSED	0.09	8.50 × 10−12
CO2pc	1.00	−0.812	CO2pc	1.00	−0.78	CO2pc	1.00	−5.54	CO2pc	1.00	−5.87	CO2pc	0.03	2.79 × 10−3	CO2pc	1.00	−6.26
CC	1.00	−0.061	CC	1.00	−0.07	CC	1.00	0.02	CC	1.00	0.02	CC	0.04	−3.84 × 10−4	CC	0.04	−3.30 × 10−4
COP	1.00	0.016	COP	1.00	0.03	COP	0.05	−6.24 × 10−5	COP	0.05	−4.51 × 10−5	COP	1.00	−0.05	COP	1.00	−0.06
FA	0.37	1.46 × 10−5	FA	0.63	3.05 × 10−5	FA	1.00	−1.78 × 10−5	FA	1.00	−1.74 × 10−5	FA	1.00	7.13 × 10−5	FA	1.00	7.63 × 10−5
FDI/GDP	1.00	−0.055	FDI/GDP	1.00	−0.05	FDI/GDP	0.05	2.56 × 10−4	FDI/GDP	0.05	5.61 × 10−4	FDI/GDP	0.88	0.06	FDI/GDP	1.00	0.15
GC	1.00	−0.088	GC	1.00	−0.10	gascons	0.99	0.08	GC	1.00	0.12	GC	0.98	−0.43	GC	0.82	−0.29
GDP pc	1.00	1.90 × 10−4	GDP	0.11	8.76 × 10−14	GDP pc	0.06	−8.14 × 10−6	GDP	0.67	−3.61 × 10−12	GDP pc	1.00	−0.01	GDP	0.61	−1.32 × 10−11
HC	0.05	2.76 × 10−3	HC	0.05	3.80 × 10−3	HC	0.40	−0.03	HC	0.17	−0.01	HC	0.05	0.01	HC	0.7	0.70
INF	0.06	−3.20 × 10−4	INF	0.05	−2.53 × 10−4	INF	0.08	−2.97 × 10−4	INF	0.08	−3.26 × 10−4	INF	0.04	3.09 × 10−5	INF	0.05	5.69 × 10−5
LF	0.06	7.19 × 10−9	LF	0.22	3.91 × 10−8	LF	1.00	−3.52 × 10−7	LF	1.00	−3.77 × 10−7	LF	0.22	−1.83 × 10−8	LF	0.57	−9.51 × 10−8
NC	0.99	−0.08	NC	1.00	−0.09	NC	0.11	−0.01	NC	0.07	−4.56 × 10−3	NC	0.04	0.01	NC	0.04	0.03
OC	0.05	−5.64 × 10−4	OC	0.04	3.79 × 10−4	OC	0.09	2.71 × 10−3	OC	0.54	0.07	OC	0.04	4.25 × 10−4	OC	0.53	0.15
TO	1.00	0.037	TO	1.00	0.05	TO	0.17	4.05 × 10−3	TO	0.16	3.45 × 10−3	TO	0.14	−2.74 × 10−3	TO	0.26	−0.01

Note: Bold font indicates PIP values greater than 0.7.

shows the posterior results for two sets of possible determinants, where Variant I includes GDPpc and Variant II uses GDP instead. The other variables in both variants remain the same. Therefore, Table 1 shows the posterior inclusion probabilities, the average value of the coefficient (parameter estimate overall considered models), and the corresponding average standard error. The posterior inclusion probability (PIP), equal to at least 0.7, shows a high probability of being included in the model. Although there is no formal requirement for high posterior probability, it is reasonable to assume that it is at least higher than 0.5 and treats the results higher than 0.7 as reliable.

In the next step, we estimate a fixed-effects model once we have established which variables can be considered significant. For this purpose, we also use the solutions available in the Gretl package [76]. The results in Table 1 demonstrated that including GDP pc or GDP as a measure of economic performance does not change the results in countries assigned to groups I and II. The only difference is in very highly developed countries (group I; 1–66). Taking GDP pc, it is also suggested that energy consumption from nuclear sources be included. The first group of countries is mostly technologically advanced and the richest. Therefore, among the factors determining REC remained CO₂pc. TO, GDP pc/GDP, FDI/GDP, COP, and energy consumption from alternative sources, mainly gas, coal, and nuclear. It is worth noting that the posterior mean was negative for CO₂pc, suggesting a negative correlation between the emissions and consuming energy from renewable sources. Ultimately, negative values of the posterior means for gas, coal, and nuclear energy consumption demonstrate a competition between using energy from alternative sources. This is obvious due to the current energy production and necessary investments. The GDP pc/GDP and terms of trade (TO) support the more intense use of energy from renewable sources. The role of FDI as a percentage of GDP is not homogenous because it depends on the structure of the investments. If investments are located in the renewable energy sector, they can increase REC; otherwise, we can expect any direction of influence. The last significant factor is the price, represented by crude oil world prices (COP). The expected sign should be positive; however, when GDP is considered, it is slightly negative. The COP increase can indirectly negatively influence REC due to its negative impact on the inflation rate and, consequently, entire economic activity. In such a case, the investment in the energy sector is slowed down, negatively impacting REC.

The second group of countries (group II; 67–115) represents a high level of development according to the HDI level. The results show the fewest factors with a posterior inclusion probability (PIP) higher than 0.7. These are CO₂pc with negative signs and labor force (LF), forest area (FA), and gas and coal consumption with positive signs. Here, there is no visible competition effect, which means that the countries are simultaneously developing energy consumption from gas, coal, and renewables. The labor force and forest area also positively impact REC, which is logical from the renewable energy sector development viewpoint.

The last group contains countries with medium and low levels of development (group III; 116–191). Here, the results are relatively diversified, depending on whether GDP pc or GDP is included. When GDP pc is included, its negative impact on REC is demonstrated, confirming that the effort to develop the poorer countries does not care much about clean energy use but uses all available energy sources. Their situation is much harder than developed ones [77].

The presence of FDI/GDP with positive signs indicates that foreign investors invest in the renewable energy sector, which highly supports the development and SDG achievement. The presence of adjusted savings: energy depletion (ADSED), with a positive sign, indicates that changes in the ratio of the value of the stock of energy resources to the remaining reserve lifetime stimulate countries to focus on renewable sources. Surprisingly, crude oil prices (COP) are negative, showing a similar impact as in group I.

In the empirical part, we propose using an FE estimator to extract the primary determinants of renewable energy consumption in the country subgroups. The results of the panel FE model estimation are presented in

Table 2. Fixed-effect estimates of renewable consumption determinants in 1995–2019.

GROUP I: Ranked 1–66					GROUP II: Ranked 67–115				GROUP III: Ranked 116–191
Dep. Variable: REC
Model 1: Fixed Effects, Using 1150 obs.					Model 2: Fixed Effects, Using 400 obs.				Model 3: Fixed Effects, Using 1175 obs.
Name	Coefficient	Std. Error	t-Ratio	p-Value	Coefficient	Std. Error	t-Ratio	p-Value	Coefficient	Std. Error	t-Ratio	p-Value
const	22.030	1.290	17.130	<0.01	73.770	3.920	18.830	<0.01	62.020	2.050	30.230	<0.01
CO2pc	−1.060	0.070	−14.54	<0.01	−5.470	0.410	−13.360	<0.01
GDPpc	1.91 × 10−4	2.35 × 10−5	8.130	<0.01					−0.0057	2.71 × 10−4	−21.02	<0.01
TO	0.030	0.010	6.000	<0.01
GC	−0.102	0.010	−8.450	<0.01	0.060	0.020	4.240	<0.01	−0.4780	0.070	−7.312	<0.01
CC	−0.060	0.010	−6.190	<0.01	0.020	3.25 × 10−3	6.280	<0.01
FDI/GDP	−0.050	0.010	−4.580	<0.01					0.070	0.020	3.220	<0.01
COP	0.020	3.69 × 10−3	4.090	<0.01					−0.049	0.010	−9.155	<0.01
NC	0.080	0.020	−4.255	<0.01
FA					−1.75 × 10−5	3.39 × 10−6	−5.168	<0.01	7.19 × 10−5	1.36 × 10−5	5.300	<0.01
ADSED									2.59 × 10−10	6.06 × 10−11	4.280	<0.01
LF					−3.42 × 10−7	3.39 × 10−8	−10.09	<0.01
Model Diagnostics
	Mean dep var	16.150	S.D. dep var	15.110	Mean dep var	22.210	S.D. dep var	17.530	Mean dep var	61.620	S.D. dep var	28.760
	Sum sq resid	10,702.60	S.E. of reg	3.120	Sum sq resid	4681.220	S.E. of reg	3.510	Sum sq resid	30,263.640	S.E. of reg	5.190
	LSDV R-sq	0.960	Within R-sq	0.420	LSDV R-sq	0.960	Within R-sq	0.570	LSDV R-sq	0.970	Within R-sq	0.470
	LSDV F(53, 1096)	486.260	p-value(F)	<0.01	LSDV F(53, 1096)	477.580	p-value(F)	<0.01	LSDV F(53, 1096)	670.730	p-value(F)	<0.01
	Log-likelihood	−2914.44	Akaike crit	5936.890	Log-likelihood	−1059.54	Akaike crit	2161.090	Log-likelihood	−3575.85	Akaike crit	7257.70
	Schwarz crit	6209.460	Hannan-Quinn	6039.780	Schwarz crit	2244.910	Hannan-Quinn	2194.290	Schwarz crit	7526.360	Hannan-Quinn	7359.01
Joint test on named regressors
Test stat: F(8, 1096) = 98.79					Test stat: F(5, 379) = 99.440				Test stat: F(6, 1122) = 162.862
with p-value = P(F(8, 1096) > 98.7939) = 1.22432 × 10−123					with p-value = P(F(5, 379) > 99.440) = 9.102 × 10−67				with p-value = P(F(6, 1122) > 162.862) = 8.207 × 10−149
Test for differing group intercepts (null hypothesis: the groups have a common intercept)
Test stat: F(45, 1096) = 374.15					Test stat: F(15, 379) = 280.98				Test stat: F(46, 1122) = 429.77
with p-value = P(F(45, 1096) > 374.157) = 0.000					with p-value = P(F(15, 379) > 280.985) = 1.039 × 10−194				with p-value = P(F(46, 1122) > 429.778) = 0.000
Hausman test (null hypothesis: GLS estimates are consistent)
Asymptotic test stat: Chi-sq(7) = 52.95					Asymptotic test stat: Chi-sq(5) = 90.22				Asymptotic test stat: Chi-sq(5) = 27.544
with p-value = 3.78485 × 10−9					with p-value = 6.022 × 10−18				with p-value = 4.466 × 10−5

.

The empirical models support the previous findings (Table 1) for the version including GDP pc, i.e., the same directions of the parameter estimates are confirmed, and the factors with posterior inclusion probabilities greater than 0.7 are statistically significant. These findings demonstrate that the panel BMA methodology can be efficiently used as a variable selection method for fixed-effects (FE) models. Additionally, the Hausman test results presented in the last three rows of Table 2 confirmed that the FE estimator outperforms the alternative random-effects (RE) one.

Referring to the methodology, we confirmed that applying the panel BMA procedure provides a reliable result since it allows a search of the entire model space to find the most likely determinants of renewable energy consumption. The most essential advantages of model averaging were indicated in [78]. The first one includes the model uncertainty in the selection procedure, which reduces overconfidence in a single model. Furthermore, it avoids the all-or-nothing mentality associated with classical hypothesis testing, where a model is either accepted or rejected wholesale. BMA gracefully updates its estimates as the data accumulate, and the resulting model weights are continually adjusted. Finally, BMA is relatively robust in terms of model misspecification. The successful application of BMA is possible for different databases, such as cross-sectional, time-series, and panel data [79,80,81].

When we refer to highly developed and developed countries (groups I and II), we can compare the results with those obtained by [82]. The authors employed a pooled OLS Model for European Union member states, while Central and Eastern Europe and Western Europe countries were analyzed separately. They analyzed entire economies and sectors that consume energy from renewable sources, such as transport, electricity, and heating and cooling. The empirical results confirmed the different conditions for the two groups of countries, suggesting the possibility of designing specific energy policies. The positive role of R&D in both categories of member states was emphasized. The authors reported a non-significant relationship between energy consumption from renewable sources and GDP per capita in Western as well as Central and Eastern European countries. The only exception was the electricity sector in the Western countries, which reported a negative relationship between REC and economic growth. Most recently, [83] examined renewable energy consumption factors in the Eurozone and confirmed a strong inverse association between GDP and renewable energy consumption. They also found a negative relationship between foreign direct investment (FDI), energy imports, and renewable energy consumption. A positive relationship between R&D and REC was also confirmed. In our study, the impact of GDP per capita on REC was positive and significant in group I—very highly developed countries, including the EU countries, among others.

When comparing with other publications, some similarities in the results can be identified with the study by [36]. The author prepared a deep analysis of African countries. He found that the two-sided (feedback) relationship between renewable energy (RE) and economic growth in the studied economies was positive and significant. He concluded that policies supporting renewable energy adoption should be advanced. A look at other explanatory variables suggests a positive relationship exists between financial development, foreign direct investment, trade, governance, urbanization, and life expectancy that stimulates RE adoption. However, the findings presented by [36] are rather unique. In our study, the African countries are mostly assigned to group III of countries, classified as medium- and low-developed. They are accompanied by other countries, so the results are different. However, it must be emphasized that when GDPpc has a negative sign, the crowding-out effect plays a specific role. To increase growth in poor countries, it is difficult to maintain an appropriate level of investment required to increase energy consumption from renewable sources. Conversely, we confirmed the positive impact of FDI, forest area, and ADSED. The last factor is highly important because limited natural resources determine the seeking and application of new technologies.

Another study related to African countries [84] found a significant negative and non-linear impact of renewable energy consumption on the economic growth of combined African countries. Dividing African countries into regions, the authors revealed a negative effect of renewable energy consumption on economic growth and further highlighted the varying resource and renewable energy development across African countries.

The results are relatively diversified depending on the data quality and availability, the diversity of the countries taken to the investigation, and possible correlations between the countries applying similar energy policies due to more general requirements (like the Sustainable Development Goals and the European Green Deal). Despite this, we confirmed that CO₂pc. TO, GDP pc/GDP, FDI/GDP, COP, and energy consumption from alternative sources, mainly gas, coal, and nuclear (GC, CC, and NC), are the essential drivers of REC in group I. In the second group (group II), the key factors are CO₂pc, labor force (LF), forest area (FA), and gas and coal consumption (GC and CC). Significant factors influencing renewable energy consumption in groups I and II are similar, taking CO₂pc and gas and coal consumption. The results in the last group (group III) are diversified, demonstrating a huge dependence of REC on FDI. Additionally, the ADSED is essential as well. In general, group III, which is less developed, remains different and requires special attention.

5. Conclusions

In this paper, we analyzed the determinants of energy consumption from renewable sources using panel BMA and fixed-effects econometric models. The results were compared in three groups of countries delimited by the HDI level, which implies the development characteristics. The total number of countries of interest comprises 109, while the observation period was limited to 1995–2019. The results obtained from the panel BMA were confirmed when the FE models were estimated. The panel BMA approach found substantial differences between factors observed in groups of countries.

The outcomes of this study confirmed that CO₂ emissions, measured per inhabitant, are crucial in implementing and increasing energy use from renewable sources. Furthermore, governments and inhabitants are generally aware of the threats of increasing emissions. This factor was positive in countries included in groups I and II. The results also revealed a similarity of factors determining REC in country groups I and II concerning other energy sources, particularly gas and coal consumption. Group III, which is less developed, relies on FDI inflow and requires special attention.

These findings are reliable and allow for energy policy recommendations. Firstly, all countries must implement an efficient, energetic policy to decrease greenhouse gas emissions, particularly CO₂. The policy goals and the time of their realization must be determined and satisfied promptly. Secondly, the participation in costs of climate policy (emission taxes, coal taxes, etc.) must be distributed among all countries; however, due to differences in development, some discounts and deferments are allowed at both the macroscale and microscale (consumers, enterprises, particularly small and medium enterprises). Thirdly, the inflow of FDI, particularly in the middle- and low-developed countries, should be primarily directed to the renewable energy sectors.

Although including as much data as possible, we found some limitations in the study. One of them is related to the limitations of the data, particularly in poorer and non-democratic countries. It refers to both the variables and countries. We did not include the sectors of the economies; instead, we took the entire economies. Analysis across the sectors would enhance the findings and deepen the conclusions.