Does Income Inequality Influence Energy Consumption in the European Union?

Abstract

This study examines the emerging challenge of how income inequality affects household energy consumption within the European Union amidst the ongoing energy crisis. Using the Gini index and the gender pay gap as indicators of income inequality, the analysis covers the period 2000–2023 across EU member states. Dynamic panel data models reveal that the Gini index negatively impacted final energy consumption per capita as well as energy consumption specifically from gas oil and diesel, while the gender pay gap is associated with less energy consumption in the realms of natural gas and ambient heating. Causality is further explored through recent methodology developed for panel data and Bayesian networks, and the results confirm the causality between energy consumption and income inequality. To address the complex relationship between income inequality and energy consumption, policymakers should implement integrated strategies promoting energy efficiency, clean energy transitions, gender equality, and social safety nets, particularly in light of the energy crisis exacerbated by the Russia–Ukraine conflict.

1. Introduction

Income inequality and unsustainable energy consumption pose significant challenges for households worldwide [1,2]. Persistent income inequality can have detrimental economic consequences, including social instability, widening regional disparities, and inequities in social welfare [3]. Similarly, rapid increases in energy use can negatively affect livelihoods, production, and environmental quality [4]. Despite these documented negative impacts of both energy consumption and income disparities on sustainable development goals, the interrelationship between these two challenges has been relatively understudied [5]. Increased energy consumption can disproportionately burden low-income households, as energy expenditures can comprise a larger share of their income [6], potentially exacerbating income inequality. Conversely, reduced renewable energy costs may decrease the proportion of household income allocated to energy, potentially mitigating income inequality [3]. While the existing research has often focused on the role of renewable energy in reducing income disparities [7], the impact of overall energy consumption has received less attention. The main aim of this paper is to assess the causality between energy consumption and income inequality, but also the impact of income inequality on energy use.

This study concentrates on EU member states due to the prominence of energy crises and income inequality in current policy discussions. Energy inequalities are often viewed as a manifestation of underlying income disparities [8]. Recent events, including the COVID-19 pandemic, high inflation, and disruptions to Russian gas supplies, have contributed to soaring electricity prices, significantly impacting households, which have been forced to curtail their energy consumption. European governments have implemented various policy responses, ranging from mitigating the effects of higher energy costs to directly reducing prices. However, these approaches often prove inadequate due to conflicting objectives and cross-border spillover effects [9]. Policymakers must therefore consider strategies that encompass reducing energy consumption, increasing energy supply, safeguarding vulnerable consumers and internal energy markets, and addressing gender pay gaps [10]. While the European Commission has set targets to reduce electricity demand by 10% and gas demand by 15%, several EU countries have yet to implement corresponding measures. The 2023 Energy Outlook report projects a 1.3% increase in global energy consumption, with a shift toward fossil fuels, potentially hindering the transition to cleaner energy sources [11]. Regarding income inequality, current EU discussions have been amplified by the rise in the gender pay gap during the pandemic. This paper offers a novel perspective by arguing that policy proposals aimed at reducing the gender pay gap should consider the influence of energy consumption. In this way, the study contributes to a deeper understanding of the gender pay gap, specifically through the lens of the energy crisis, which may necessitate a reevaluation of existing policy recommendations.

While the Gini index is frequently used to measure income inequality, recent European policy discussions increasingly emphasize the gender pay gap, highlighting both the slow progress in its reduction and the significant disparities among EU member states. This paper offers a novel contribution by being the first to specifically investigate the impact of energy consumption on the gender pay gap. Furthermore, it considers a comprehensive set of energy indicators, including per capita final energy consumption in households, final energy consumption from natural gas, gas oil, and diesel oil, as well as energy used for specific purposes such as heating and ambient heating. The methodological approach combines panel data models (specifically, dynamic panel data models), causality analysis, and Bayesian networks. The application of Bayesian networks to the study of the income inequality–energy consumption nexus is a further innovation.

The gender pay gap is a multifaceted issue that extends beyond simple pay discrimination, encompassing a range of inequalities women experience in employment access, career advancement, and compensation. These include sectoral segregation, the unequal division of paid and unpaid work, the “glass ceiling”, and direct pay discrimination. Firstly, approximately 24% of the gender pay gap can be attributed to the overrepresentation of women in relatively lower-paying sectors like care, health, and education [12]. Jobs with a high concentration of women tend to be systematically undervalued. Secondly, while women often work more total hours per week than men, a greater proportion of their time is dedicated to unpaid work, which can impact their career trajectories. This disparity is why the EU advocates for equal sharing of parental leave, sufficient public childcare services, and flexible work arrangements within companies. Thirdly, hierarchical position significantly influences pay levels. Despite this, less than 10% of top company CEOs are women. Interestingly, management positions exhibit some of the largest gender-based hourly earnings differences in the EU, with women earning 23% less than men. Fourthly, some women receive lower pay than men for performing the same work or work of equal value, even though the principle of equal pay has been a fundamental tenet of European treaties (Article 157 TFEU) since 1957.

In 2023, Bulgaria exhibited the greatest income disparity among EU nations, with a Gini coefficient of 37.2. The Gini index, a standard measure of income distribution, ranges from 0 (perfect equality) to 100 (complete inequality). Conversely, Slovakia demonstrated the most equitable income distribution within the EU, registering a Gini coefficient of 21.6 in the same year.

Although no single, universally accepted theory directly links gender pay disparities to energy consumption, relevant frameworks can shed light on this intersection. Feminist political ecology, for example, explores the interconnectedness of gender, power, and environmental issues, suggesting that unequal power dynamics between men and women influence energy policy and resource distribution [13]. Within this framework, lower wages for women could restrict their access to energy-efficient appliances or clean energy technologies, potentially increasing their household energy consumption. Moreover, women’s disproportionate involvement in unpaid care work, such as cooking and cleaning, which are often energy-intensive, should also be considered.

The Environmental Kuznets Curve (EKC) hypothesis proposes an inverted U-shaped relationship between economic growth and environmental degradation. It suggests that, initially, pollution and resource consumption (including energy) may increase with development, but they eventually decline as societies place greater emphasis on environmental protection. While not explicitly focused on gender, the EKC framework may be relevant if women’s economic empowerment influences a shift toward cleaner and more efficient energy practices within households [12]. However, the EKC theory remains contested, and its applicability can vary across contexts.

The socioeconomic metabolic approach analyzes societal metabolism, which encompasses the flow of resources (including energy) within a society. It considers how social factors such as gender roles influence resource utilization patterns. This perspective could be applied to examine how unequal pay distribution manifests in differing energy consumption patterns within households [14]. For example, women with lower incomes may be compelled to use less efficient energy sources due to financial constraints.

The gender–energy poverty nexus explores the disproportionate impact on women of limited energy access. While not directly concerned with pay disparities, this area of research highlights women’s vulnerability to energy insecurity [15]. This connection is significant because energy poverty can lead to increased dependence on traditional fuels, such as firewood, which can have adverse health and environmental consequences.

It is crucial to recognize that these are conceptual frameworks, not empirically validated causal relationships. Further research is necessary to fully understand the specific mechanisms through which gender pay disparities may influence energy consumption patterns.

Against this backdrop within the EU context, the central objective of this paper is to assess the impact of both the Gini index and the gender pay gap, used as measures of income inequality, on energy consumption in the period 2000–2023. The analysis also incorporates several control variables, including GDP per capita, inflation rate, unemployment rate, foreign direct investment, and urban population. Given this theoretical background, the research considers two hypotheses to be tested:

H₁: 

Is there any causality between energy/electricity consumption and income inequality?

H₂: 

Does income inequality impact energy/electricity consumption?

This research offers several novel contributions that address gaps in the existing literature. It examines the effect of income inequality on household energy consumption, aiming to inform current policy debates. It employs Bayesian networks to assess the robustness of the results obtained from dynamic panel data models and panel data causality analysis. Finally, it develops policy recommendations for EU member states based on the empirical findings.

The novelty of this study is given by the value of the empirical findings that could support policy recommendations. The results partially confirm the hypotheses formulated above. The Gini index has a negative impact on final energy consumption per capita as well as on energy consumption specifically from gas oil and diesel. On the other hand, the gender pay gap appears to be linked to decreased energy consumption in the realms of natural gas and ambient heating. The causal relationship is supported in the panel data approach in almost all cases, excepting the relationship between gender pay gap and electricity consumption.

The subsequent literature review focuses on the relationship between income inequality and energy consumption, highlighting the specific gaps that this study aims to address. The following sections detail the data and methodology, present the empirical results, discuss these findings and formulate policy recommendations, and conclude the paper.

2. Literature Review

The literature on income inequality frequently explores its relationship with factors such as economic growth, human capital, investment, government policies, and institutional quality [16]. However, the connection between income disparities and energy use has received less attention, primarily focusing on the context of energy crises and the urgent need for solutions. The economic impact of income inequality remains a subject of debate, with researchers unsure whether it hinders economic performance or is simply a byproduct of development. Further investigation is needed to understand how income inequality interacts with specific crises, exploring whether it mitigates or amplifies their effects. Existing theories, such as power inequity and energy consumption elasticity, have attempted to explain the link between income inequality and energy use. However, these theories have struggled to reconcile the conflicting empirical findings observed across different country samples, lacking a comprehensive understanding of the underlying mechanisms [17]. The close relationship between these two indicators has even led some researchers to propose household electricity consumption as a proxy for income inequality [18].

The simultaneous rise in income inequality and energy consumption has spurred empirical research into their interconnectedness. While much of this research has focused on the impact of income inequality on household energy consumption, this paper examines the less-explored effect of energy use on income inequality. Studies investigating the former relationship, often using cross-national samples, have explored both the nature (linear or non-linear) and direction (positive or negative) of the link. For instance, Tan and Uprase [5] found an inverted U-shaped relationship between income disparities and energy use in ASEAN countries, suggesting a non-linear dynamic. This non-linear association was also observed by Xu and Zhong [19], who emphasized the moderating role of digitalization, in a broader sample of 108 countries between 2000 and 2019. Other studies have posited a positive relationship between income inequality and energy consumption. Du et al. [20], for example, highlighted the role of unequal power dynamics among different income groups in driving energy consumption. Qiao and Dowell [21] suggested that income disparities can expand the supply of energy products and increase demand for traditional energy sources, thereby driving up energy use. Focusing on the reverse relationship, Pillay [22] found a positive influence of residential electricity consumption on the Gini index in South Africa. This finding was corroborated by Sarkodie and Adams [23] for the same country over the period 1990–2017, using a non-linear autoregressive distributed lag model.

A contrasting perspective suggests that income inequality can actually contribute to reduced energy consumption. Odhiambo [24], for example, proposes that income inequality can be a tool for mitigating energy consumption in the context of environmental protection. Threshold effects between income inequality and environmental quality have also been observed [25,26]. Specifically for the United States, Linn et al. [27] found that income inequality led to a 9% reduction in average household electricity consumption between 1990 and 2020. Furthermore, they projected continued declines in electricity use due to necessary cost-saving measures among lower-income households, such as foregoing air conditioning or heating.

This paper aligns with a second body of research that examines the influence of energy consumption on income inequality. Sonora [16], using panel data from 144 countries spanning 1990–2018, reported a negative relationship between energy consumption and income inequality. This negative association was also observed in sub-sample analyses across four economic groups (OECD, non-OECD, OPEC, and former/current communist economies) and five regions (Asia, Eastern Europe, Western Europe, Latin America, and Asia). Other studies have investigated causal links between these variables. Galvin [28], for instance, identified a significant causal relationship from electricity consumption to income inequality in middle-income countries. Distinct from these prior works, our study employs Bayesian networks to explore causal relationships, complementing traditional panel data causality analysis.

The relationship between income inequality and renewable energy consumption has been a prominent area of research. Theoretically, increased renewable energy use can contribute to energy price stabilization due to the declining costs of renewable technologies [29]. This price stabilization, particularly with further transitions away from fossil fuels, can have implications for income inequality. In less-developed countries, where energy and natural resource expenditures constitute a larger proportion of household budgets compared with developed nations, lower energy costs resulting from renewable energy adoption can alleviate the financial burden on poorer households. This reduction in energy-related expenses can potentially lessen income inequality. However, not all research has identified a clear link between energy consumption and income inequality. For instance, Nar [8], utilizing a Pareto-optimal approach, concluded that energy consumption did not have a discernible impact on income inequality across various country groups between 1980 and 2018.

Economic theory provides a basis for understanding how various factors influence both the Gini index, a measure of income inequality, and the gender pay gap. Energy consumption is often associated with economic activity and growth. However, unequal access to affordable energy can exacerbate existing inequalities, particularly if it impedes the economic productivity of lower-income populations. A transition toward renewable energy sources can potentially mitigate these effects by creating new employment opportunities and improving energy access, which may contribute to reducing inequality [30]. Access to and affordability of energy also have implications for women’s economic participation. Limited or unreliable energy access can restrict women’s ability to engage in entrepreneurial activities, pursue education or training, or participate in paid employment, potentially widening the gender pay gap. The adoption of clean energy technologies can generate new jobs and improve energy access, potentially benefiting women and contributing to a reduction in the gap [31].

The relationship between GDP growth and income distribution is complex. Broadly shared economic growth can contribute to greater income equality through mechanisms such as job creation, wage increases, and expanded public resources for social programs. However, uneven growth patterns can exacerbate inequality, as a concentration of benefits among a select few widens the gap between the wealthy and the less affluent, potentially increasing the Gini index [32]. Similarly, robust economic expansion can potentially reduce the gender pay gap by creating more employment opportunities and empowering women to enter the workforce, negotiate better compensation, and access higher-paying industries. Conversely, if the benefits of growth are disproportionately concentrated among men, the gender pay gap may persist or even widen [33].

The relationship between urbanization and income inequality is multifaceted. While urbanization can generate opportunities and higher wages, potentially contributing to a reduction in the Gini index, it can also lead to increased living costs and unequal access to resources, potentially exacerbating inequality. Effective urban planning policies, including affordable housing initiatives, infrastructure development, and targeted job creation programs, are crucial for mitigating the negative impacts of urbanization on income distribution [34]. Urbanization can also influence the gender pay gap. Increased job opportunities and potentially higher wages in urban areas can contribute to narrowing the gap. However, challenges such as elevated living expenses and limited access to childcare services can hinder women’s workforce participation and potentially widen the pay gap [35].

Foreign direct investment (FDI) can stimulate job creation and contribute to economic expansion. The distributional impact of FDI depends on how its benefits are disseminated throughout the economy. Broadly distributed gains from FDI can contribute to greater income equality and a reduction in the Gini index. However, if FDI is concentrated in specific sectors that primarily benefit a limited segment of the population, it can worsen income inequality and potentially increase the Gini index [36]. The influence of FDI on the gender pay gap is also nuanced. While FDI can generate new employment opportunities, its effect depends on the sectoral allocation of investment. If FDI flows into traditionally male-dominated, higher-paying industries, it may exacerbate the gender pay gap. Conversely, if it promotes growth in sectors with greater gender balance, it could contribute to narrowing the gap [37].

Unemployment rates can significantly influence income distribution. Elevated unemployment can exacerbate income inequality by depriving individuals of their primary income source, pushing them further down the income distribution and consequently increasing the Gini index [38]. While unemployment affects both men and women, it can disproportionately impact women’s labor force participation. This can lead to a decrease in the measured gender pay gap, as unemployed women are often excluded from wage gap calculations. However, this does not reflect genuine progress toward gender pay equity. Conversely, low unemployment rates, coupled with robust policies that promote equal opportunities for women, can contribute to narrowing the actual gender pay gap [39].

Inflation can have differential impacts on income distribution. Elevated inflation rates can disproportionately affect lower-income individuals, as their wages may not adjust quickly enough to offset rising prices. This erosion of purchasing power can exacerbate income inequality and contribute to a higher Gini index. Conversely, low and stable inflation may not have a substantial direct impact on the Gini index [40]. Furthermore, high inflation can disproportionately disadvantage women, who are often overrepresented in lower-paying occupations. If wage adjustments lag behind inflation, women’s purchasing power may be more significantly diminished, potentially widening the gender pay gap [41].

3. Data and Methodology

This study utilizes panel data from the 27 EU member states in the period 2000–2023, given the data availability constraints for certain indicators. The methodology is presented in three stages: preliminary tests preceding model construction, dynamic panel data modeling using the Generalized Method of Moments (GMM) approach, and causality analysis employing the Juodis, Karavias, and Sarafidis [42] test (JKS).

Table 1. The variables’ presentation.

Variable	Abbreviation for Data in Natural Logarithm	Sources of Data	Unit of Measurement
GDP per capita	GDP	World Bank	Constant 2015 USD
Final energy consumption in households per capita	Energy	Eurostat	Kilogram of oil equivalent
Gini index	Gini	World Bank	%
Gender pay gap	GPG	Eurostat
Inflation rate	Inflation	World Bank
Unemployment rate as percentage of total labor force	Unemployment	World Bank, estimate provided by International Labor Office
Urban population as percentage of total population	Urban	United Nations Population Division World Bank
Foreign direct investment, net inflows as percentage of GDP	FDI	World Bank
Electricity consumption of households	Electricity	Eurostat	Thousand tons of oil equivalent
Final energy consumption of households based on natural gas	Natural gas
Final energy consumption of households based on diesel and gas oil	Diesel gas
Final energy consumption of households for heat	Heat
Final energy consumption of households for ambient heat	Ambient heat

Source: own presentation.

provides a detailed description of all variables used in these analyses. To mitigate potential multicollinearity issues, all data were transformed using natural logarithms. Correlation coefficients among the explanatory variables remained below 0.4, with the highest correlation (0.357) observed between the natural logarithms of GDP per capita and energy consumption. Consequently, multicollinearity was not deemed problematic within the models.

The gender pay gap varies considerably across EU member states. For instance, data from 2021 reveal that Luxembourg reported the smallest gap, with women earning slightly more than men on average (−0.2%). Conversely, Estonia exhibited the largest gap in 2023 (21.3%), indicating a significant disparity in earnings between genders.

3.1. Panel Data Approach

Prior to conducting estimations, it is crucial to assess the stationarity of the data through unit root tests. Furthermore, preliminary tests for cross-sectional dependence and heterogeneity are necessary to inform the selection of the appropriate panel unit root test. Cross-sectional dependence was evaluated using Pesaran’s CD test [43], which tests the null hypothesis of independence.

• Null hypothesis: ${\rho }_{ij}={\rho }_{ji}=cor\left(e_{it},e_{jt}\right)=0,i\ne j$ (cross-sectional independence)
• Alternative hypothesis: ${\rho }_{ij}={\rho }_{ji}\ne 0,for some i\ne j$

${\rho }_{ij}$—pair-wise correlation coefficient for errors:

$${\rho }_{ij}={\rho }_{ji}={\displaystyle \frac{\sum _{t=1}^T{e}_{it}·e_{jt}}{\sqrt{\sum _{t=1}^T{e}_{it}^2}·\sqrt{\sum _{t=1}^T{e}_{jt}^2}}}$$

(1)

The CD statistic is

$$CD=\sqrt{{\displaystyle \frac{2}{N(N-1)}}·}{\sum }_{i=1}^{N-1}{\sum }_{j=i+1}^N\sqrt{T_{ij}}·{\widehat{\rho }}_{ij}$$

(2)

$T_{ij}$—number of common observations between countries (i and j)

$${\widehat{\rho }}_{ij}={\widehat{\rho }}_{ji}={\displaystyle \frac{\sum _{t\in T_i\cap T_j}\left({\widehat{e}}_{it}-{\overline{e}}_i\right)({\widehat{e}}_{jt}-{\overline{e}}_j)}{\sqrt{\sum _{t\in T_i\cap T_j}{\left({\widehat{e}}_{it}-{\overline{e}}_i\right)}^2}·\sqrt{\sum _{t\in T_i\cap T_j}{({\widehat{e}}_{jt}-{\overline{e}}_j)}^2}}}$$

(3)

$${\overline{e}}_i={\displaystyle \frac{\sum _{t\in T_i\cap T_j}\left({\widehat{e}}_{it}\right)}{\#(T_i\cap T_j)}}$$

(4)

The slope heterogeneity is based on the Pesaran and Yamagata test [44]:

$$\tilde{S}={\sum }_{i=1}^N{\left({\widehat{\beta }}_i-{\tilde{\beta }}_{WFE}\right)}^{\prime }{\displaystyle \frac{X_i^{\prime }{I}_t{X}_i}{{\tilde{\sigma }}_i^2}}({\widehat{\beta }}_i-{\tilde{\beta }}_{WFE})$$

(5)

${\widehat{\beta }}_i$—OLS estimator for country i; ${\tilde{\sigma }}_i^2$—estimate of dispersion; ${\tilde{\beta }}_{WFE}$—weighted fixed effect pooled estimator; $I_t$—unit matrix.

The standardized variance ($\widehat{∆}$) and biased-adjusted variance (${\overline{∆}}_{adj}$) are:

$$\widehat{∆}=\sqrt{N}·{\displaystyle \frac{N^{-1}\tilde{S}-k}{\sqrt{2k}}}$$

(6)

$${\overline{∆}}_{adj}=\sqrt{N}·{\displaystyle \frac{N^{-1}\tilde{S}-E\left({\overline{z}}_{it}\right)}{\sqrt{var\left({\overline{z}}_{it}\right)}}}$$

(7)

$$E\left({\overline{z}}_{it}\right)=k, var\left({\overline{z}}_{it}\right)={\displaystyle \frac{2k(T-k-1)}{T+1}}$$

In the presence of cross-sectional dependence, second-generation panel unit root tests such as Pesaran’s CADF test are appropriate. Once stationarity is confirmed (or achieved through transformations like taking the natural logarithm), dynamic panel data models can be constructed.

$${ineq}_{it}={a_i+b_{1}ineq}_{it-1}+b_2{energy\_cons}_{it}+{b_{3}GDP}_{it}+b_4{inflation}_{it}+{b_{5}unemployment}_{it}+{b_{6}urban}_{it}+{b_{7}FDI}_{it}+e_{it}$$

(8)

i—index for country; t—index for year; $e_{it}$—error

ineq—income inequality (Gini index, gender pay gap)

energy_cons—energy indicator (energy, electricity, natural gas, diesel gas, heat, ambient heat)

$a_i,b_1,b_2,b_3,b_4,b_5,b_6,b_7$—coefficients.

Dynamic panel Generalized Method of Moments (GMM) estimators offer advantages in addressing endogeneity within datasets characterized by a limited number of time periods and a larger number of cross-sections. To account for potential heteroskedasticity and autocorrelation in the error terms, the Windmeijer [45] finite-sample correction is applied to the two-step system GMM estimator. Both dependent and explanatory variables are utilized as instruments. Instrument validity is assessed using the Hansen [46] overidentification test. Serial autocorrelation is examined via the Arellano and Bond [47] test. Finally, the Juodis, Karavias, and Sarafidis [42] (JKS) test is employed to investigate panel causality within the stationary data.

3.2. Bayesian Network Approach

The dataset included missing values, necessitating imputation before applying machine learning techniques. Selecting an optimal imputation strategy is a complex task, as no single method is universally superior. Based on a review of the literature [48,49]. Random Forest imputation emerged as a promising candidate due to its effectiveness and its lack of distributional assumptions [50]. Consequently, we employed the Random Forest method, which constructs an ensemble of decision trees from bootstrapped data samples to predict missing values. Descriptive statistics for the data before and after imputation are presented in Table 1 and

Table 2. Specific cross-sectional dependence and slope heterogeneity tests.

Indicator	CD stat.	${\overline{∆}}_{\mathit{a}\mathit{d}\mathit{j}}$
GDP	49.36 ***	0.121
Gini	61.23 ***	−3.549 ***
GPG	69.99 ***	−1.445
Energy	19.34 ***	−0.593
Electricity	20.18 ***	−3.808 ***
Natural gas	24.19 ***	−3.223 ***
Diesel gas	19.13 ***	−4.232 ***
Inflation	49.37 ***	−2.966 ***
Heat	11.98 ***	−2.197 ***
Ambient heat	15.18 ***	−3.184 ***
Urban	87.59 ***	−4.556 ***
Unemployment	26.48 ***	−1.556
FDI	4.86 ***	−2.665 ***

Note: *** indicates significance at 1%.

, respectively. The imputation process was performed using the R package mice, version 4.4.2 of R [51].

To explore the causal relationships between variables, we employed Bayesian networks, a machine learning approach for probabilistic graphical modeling [52,53]. These networks represent both dependent and independent relationships between random variables through a graph structure. Nodes within the graph correspond to variables, while edges signify the relationships between them. Specifically, we utilized directed acyclic graphs (DAGs) to represent these relationships, where the direction of the edges indicates potential causality. Bayesian networks are valuable tools for reasoning about uncertainty and causality by encoding probabilistic relationships within the DAG. This framework enables efficient inference, facilitating predictions and informed decision-making. Bayesian network analysis involves both structure learning (determining the DAG’s edges) and parameter learning (computing conditional probabilities). Our focus was solely on structure learning, as the resulting DAG structure provides insights into potential causal relationships between the variables.

Constructing Bayesian networks often requires learning from data, particularly in complex scenarios where expert knowledge is insufficient. The challenge of structure learning involves identifying the DAG that best represents the observed data. While exhaustive searches are feasible for small networks (up to roughly 20–30 nodes), larger networks necessitate heuristic approaches. Two primary categories of algorithms address this challenge: score-based algorithms and constraint-based algorithms. Score-based algorithms employ a scoring function to evaluate how well a network fits the data. They then search for the DAG that maximizes this score. Search strategies can include local searches (starting with an empty or complete graph and iteratively adding/removing edges) or greedy searches (assuming a topological order and adding parent nodes based on score improvement). Constraint-based algorithms utilize independence tests to identify edge constraints and subsequently search for a DAG that adheres to these constraints.

Our analysis employed one score-based algorithm, Hill Climbing [54], and a selection of five established constraint-based algorithms: PC [55], Grow-Shrink [56], Incremental Association, Fast Incremental Association [57], Interleaved Incremental Association [58], Incremental Association with False Discovery Rate [59], and Max-Min Parents and Children [60]. The resulting DAGs for each method are illustrated in Figure 1a–h. For Hill Climbing, the Bayesian Information Criterion (BIC) served as the scoring function. For the constraint-based methods, linear correlation was used as the conditional independence test. The R packages bnlearn, version 4.4.2 of R [61] and bnstruct [62] were utilized for the analysis.

4. Results

4.1. Preliminary Tests

Table 2 indicates that cross-sectional dependence is fulfilled for all the series. However, slope heterogeneity is confirmed for all variables except GDP, GPG, energy, and unemployment at 1% significance level.

Since the cross-sectional dependence is fulfilled, the Pesaran’s CADF test is employed, and it suggests that most of the data series are stationary in level, except energy consumption based on natural gas and energy consumption for heat and ambient heat, which are integrated at order 1 (see

Table 3. The results based on CADF test.

Indicator	Data
	Data in Level	Data in the First Difference
Energy	−6.112 ***	−7.088 ***
Electricity	−4.128 ***	−4.556 ***
Natural gas	0.119	−2.011 **
Diesel gas	−2.184 **	−3.944 ***
Heat	−0.754	−3.665 ***
Ambient heat	−4.543 ***	−3.878 ***
GPG	−7.445 ***	−8.144 ***
Urban	−8.766 ***	−9.599 ***
Gini	−4.117 ***	−4.544 ***
FDI	−1.667 **	−3.204 **
GDP	−3.155 ***	−4.977 ***
Inflation	−4.998 ***	−4.334 ***
Unemployment	−5.154 ***	−6.559 ***

Note: stars suggest significance at various levels: *** (significant at 1%), ** (significant at 5%), p-values in brackets.

).

4.2. Dynamic Panel Data Models

Table 4. Models to explain various forms of energy consumption in the EU-27 countries (2000–2023).

Variable	Coefficient (*, **, and *** for Significance at 10%, 5%, 1% Levels)
	Energy		Electricity		Δ Natural Gas		Diesel Gas		Δ Heat		Δ Ambient Heat
Energy in the previous year	0.912 ***	0.911 ***	-	-	-	-	-	-	-	-	-	-
Electricity in the previous year	-	-	1.010 ***	1.009 ***	-	-	-	-	-	-	-	-
Δ natural gas in the previous year	-	-	-	-	−0.060	−0.066	-	-	-	-	-	-
Diesel gas in the previous year	-	-	-	-	-	-	0.914 ***	0.922 ***	-	-	-	-
Δ heat in the previous year	-	-	-	-	-	-	-	-	−0.089	−0.084	-	-
Δ ambient heat in the previous year	-	-	-	-	-	-	-	-	-	-	−0.02	−0.017
GDP	−0.009	−0.005	0.007	0.005	0.006	0.006	−0.191 ***	−0.2 ***	−0.033	−0.003	0.007	0.007
Inflation	−0.017 *	−0.022 *	−0.002 *	−0.002 *	−0.002 *	−0.003 *	−0.027 *	−0.03 **	−0.004	−0.001	−0.002 *	−0.004 *
Unemployment	−0.014 **	−0.017 **	−0.030 ***	−0.031 ***	−0.011 *	−0.008 *	−0.092 *	−0.102 *	−0.03	−0.003	−0.034 *	−0.032 *
Gini	−0.004 *	-	−0.001	-	−0.004	-	−0.025 *	-	0.008	-	0.004	-
GPG	-	−0.001	-	0.002	-	0.021 *		0.048	-	0.022 *	-	−0.044 **
Urban	0.016 *	0.022 **	0.002 ***	0.002 ***	0.006 ***	0.008 ***	0.015 ***	0.011 **	0.004 **	0.003 *	0.004 **	0.004 **
FDI	0.001	0.005	0.008 **	0.008 **	0.005	0.008	0.025	0.014	0.015	0.018	0.011	0.013
Constant	0.624 ***	1.443 ***	0.035	0.038	0.075	0.021	2.412 ***	2.378 ***	0.093	0.015	0.047	0.017
Diagnostics
Number of instruments	233	233	72	72	61	61	72	72	61	61	61	61
AR(2) p-value	0.115	0.198	0.278	0.272	0.376	0.399	0.273	0.245	0.339	0.566	0.645	0.8765
Hansen p-value	0.886	0.673	0.443	0.421	0.663	0.678	0.322	0.315	0.382	0.445	0.232	0.328

Source: own calculations in Stata. Note: ***, **, * show significance at 1%, 5%, and 10%.

shows that the urban population increased total energy consumption per capita based on various sources and uses and electricity consumption. More economic growth reduced final energy consumption based on diesel and gas oil, while inflation and unemployment diminished electricity use and energy consumption, except energy use for heat. FDI contributed to more electricity consumption. The Gini index reduced the final energy consumption per capita and that based on gas oil and diesel. The gender pay gap contributed to less energy consumption based on natural gas and energy for ambient heat.

4.3. Causality Analysis

According to the results in

Table 5. The causality between income inequality and energy consumption (2000–2023).

Null Hypothesis	Computed Statistic	Conclusion
Gini is not Granger cause for energy	−3.89 ***	Bidirectional causality between Gini and energy
Energy is not Granger cause for Gini	2.33 ***
GPG is not Granger cause for energy	−6.89 ***	Bidirectional causality between GPG and energy
Energy is not cause for GPG	2.99 ***
Gini is not Granger cause for electricity	−2.86 ***	Bidirectional causality between Gini and energy
Electricity is not Granger cause for Gini	78.66 ***
Gini is not Granger cause for Δ natural gas	19.16 ***	Bidirectional causality between Gini and Δ natural gas
Δ natural gas is not Granger cause for Gini	−2.64 **
Gini is not Granger cause for diesel gas	−0.94	Causality from diesel gas to Gini
Diesel gas is not Granger cause for Gini	−2.66 ***
Gini is not Granger cause for Δ heat	19.22 ***	Bidirectional causality between Gini and Δ heat
Δ heat is not Granger cause for Gini	−10.17 ***
Gini is not Granger cause for Δ ambient heat	6.88 ***	Bidirectional causality between Gini and Δ ambient heat
Δ ambient heat is not Granger cause for Gini	−4.55 ***
GPG is not Granger cause for electricity	1.02	No causality between GPG and electricity
Electricity is not Granger cause for gender pay gap	−0.56
GPG is not Granger cause for Δ natural gas	−3.51 ***	Causality from GPG to Δ natural gas
Δ natural gas is not Granger cause for GPG	−1.09
GPG is not Granger cause for diesel gas	−2.56 ***	Bidirectional causality between GPG and diesel gas
Diesel gas is not Granger cause for GPG	2.88 ***
GPG is not Granger cause for Δ heat	−1.59 *	Bidirectional causality between GPG and Δ heat
Δ heat is not Granger cause for GPG	−2.48 **
GPG is not Granger cause for Δ ambient heat	−0.89	Causality from Δ ambient heat to GPG
Δ ambient heat is not Granger cause for GPG	2.98 ***

Source: own calculations in Stata. Note: ***, **, * show significance at 1%, 5%, and 10%.

based on the Juodis, Karavias, and Sarafidis (JKS) test, the bidirectional causality is supported in almost all the cases for the period 2000–2023, except a few cases: causality from diesel gas to Gini, from GPG to Δ natural gas, and from Δ ambient heat to GPG. No causality was detected between GPG and electricity.

Analysis of the Bayesian networks depicted in Figure 1 reveals a strong association between the Gini index and energy. Six out of the eight algorithms used identified the Gini index as a causal factor for energy. One algorithm (Hill Climbing) suggested the reverse causality, while another (Max-Min Parent Child) was inconclusive. A direct causal link from GPG to electricity consumption was observed in four instances, specifically using the Hill Climbing, Grow-Shrink, Incremental Association, and Fast Incremental Association methods. The remaining methods did not detect such a relationship. Panel data analysis corroborates the causal direction from the Gini index to energy. Based on the Bayesian factor, the Hill Climbing method appears to be the most effective in this context. These results suggest a bidirectional causal relationship between the Gini index and final energy consumption and a unidirectional relationship from gender pay gap to final energy use.

5. Discussion and Policy Proposals

A higher Gini index indicates greater income inequality. In such societies, a larger portion of the population may have lower disposable income, potentially leading to reduced overall energy demand, a shift toward essential energy use, and limited access to resources. Lower-income households may have less capacity to purchase energy-intensive goods and services or to invest in energy-efficient appliances. Consumption may be concentrated on basic needs like heating and cooking, with less discretionary spending on energy-intensive activities. In highly unequal societies, marginalized communities may face barriers to accessing reliable energy sources or adopting cleaner technologies [63].

A wider gender pay gap often reflects broader gender inequalities, which can impact energy consumption in more ways. When women earn less, overall household income may be lower, limiting energy expenditures. Traditional gender roles may influence energy use. For example, if women are primarily responsible for household chores, they may use less energy-intensive appliances or have less control over energy-related decisions. In households with limited resources, women and girls may be disproportionately affected by energy poverty, having less access to clean and affordable energy.

A 2022 United Nations report states that household air pollution, largely caused by burning inexpensive but environmentally damaging fuels for heating and cooking, leads to the premature deaths of 3.2 million people annually, primarily women and children. Rising energy costs are driving a resurgence in biomass fuel use, which, as the UN report also notes, has a disproportionate negative impact on women and girls, increasing their unpaid domestic work, harming their health, and jeopardizing their livelihoods. This, in turn, is likely to hinder their access to education and healthcare, further widening the gender pay gap [64]. Tindall et al. [65] suggest that women tend to exhibit more environmentally conscious and energy-efficient behaviors than men. Furthermore, they posit that women in decision-making roles are more likely to advocate for pro-environmental legislation and equitable compensation. This suggests that increasing women’s representation in leadership positions, particularly within the energy sector, could be beneficial for promoting both environmental sustainability and gender equality.

In addition to these findings for the energy use–income inequality nexus, the impact of control variables is also relevant. Cities concentrate populations, businesses, and industries, leading to a higher demand for energy in a smaller area. This includes energy for housing, transportation, commercial activities, and industrial production. Urban living can lead to increased energy consumption due to factors like higher use of private vehicles, increased reliance on energy-consuming appliances (air conditioning, electronics), and more frequent out-of-home activities (dining, entertainment) [66].

Economic growth often drives innovation and the adoption of more energy-efficient technologies. This can lead to reduced reliance on traditional fossil fuels like diesel and gas oil. As economies grow, there might be a shift toward less energy-intensive sectors (services, information technology) and away from heavy industries that rely heavily on diesel and gas oil. Growing economies may have the resources to invest in cleaner energy sources and implement policies that discourage the use of polluting fuels [40].

Inflation and unemployment can decrease disposable income, forcing individuals and businesses to cut back on non-essential spending, including energy consumption. Economic downturns often lead to reduced industrial production and commercial activity, which translates to lower energy demand. Energy use for heating is often considered a necessity, so it might be less affected by economic fluctuations compared with other forms of energy consumption [67].

Foreign direct investment often targets industrial sectors, which tend to be heavy consumers of electricity. FDI can contribute to the development of energy infrastructure, including power plants and distribution networks, leading to increased electricity generation and consumption.

Since the Gini index (income inequality) negatively correlates with energy consumption, particularly of gas oil and diesel, policies should focus on making energy efficiency improvements accessible to lower-income households. This could involve subsidies for energy-efficient appliances, home insulation programs, and financial assistance for switching to cleaner energy sources. This not only addresses energy consumption but also alleviates the energy burden on vulnerable populations. Addressing the root cause of the Gini index’s impact requires broader policies aimed at reducing income inequality. Progressive taxation, robust social safety nets, and investments in education and job training can help level the playing field and improve economic opportunities for marginalized communities, indirectly impacting their energy consumption patterns [68,69].

The link between the gender pay gap and decreased energy consumption suggests that empowering women economically can have broader societal benefits. Policies promoting equal pay for equal work, access to education and training for women in diverse fields (including STEM), and support for women-owned businesses are crucial [70,71]. This not only addresses gender inequality but also could influence energy consumption patterns in the long term.

Given the strong link between urbanization and increased energy consumption, urban planning must prioritize energy efficiency. This includes promoting compact, walkable cities, investing in public transportation, encouraging green building practices, and supporting the development of renewable energy sources within urban areas [72]. As FDI increases electricity consumption, investment in smart grid technologies is crucial. Smart grids can optimize energy distribution, reduce waste, and integrate renewable energy sources more effectively, mitigating the environmental impact of increased electricity demand [73].

While economic growth correlates with reduced reliance on diesel and gas oil, policies should actively incentivize investments in renewable energy technologies and energy efficiency across all sectors. This could include tax credits, subsidies, and research and development funding for clean energy solutions. Reducing reliance on specific fossil fuels is essential for long-term energy security and environmental sustainability. Policies should support the diversification of energy sources, including wind, solar, hydro, and other renewables [74].

Recognizing the impact of inflation and unemployment on energy consumption, policies should provide targeted support to vulnerable households during economic downturns. This could include energy assistance programs, unemployment benefits, and measures to stabilize energy prices [75,76,77].

The interconnectedness of these issues necessitates an integrated policy framework that considers the social, economic, and environmental dimensions of energy consumption. Policies should be designed to address multiple goals simultaneously, such as promoting gender equality, reducing income inequality, and transitioning to a cleaner energy future.

6. Conclusions

The current European landscape necessitates EU-wide solutions to address both the energy crisis and gender discrimination. Crucially, these solutions must consider the interplay between energy consumption and income inequality as measured by the Gini index and the gender pay gap. Women are disproportionately affected by the energy crisis due to the persistent gender pay gap and the disproportionate burden they carry in unpaid domestic and care work. The findings presented here suggest that governments should prioritize the development of gender-sensitive policies to mitigate energy poverty. The demonstrated causal link between income inequality measures and per capita final energy consumption underscores the need for an integrated policy framework. Energy consumption, in turn, can exacerbate income inequality, as reflected in both the Gini index and the gender pay gap. The increased energy demands of unpaid domestic work, often performed by women, can limit their access to employment and educational opportunities, negatively impacting their income potential. Therefore, ensuring equitable access to clean energy for both men and women is essential.

The Gini index, a measure of income inequality, presents a negative impact on final energy consumption per capita as well as energy consumption specifically from gas oil and diesel. On the other hand, the gender pay gap appears to be linked to decreased energy consumption in the realms of natural gas and ambient heating.

Urbanization appears to drive an increase in per capita energy consumption, total energy use, energy derived from various sources and applications, and electricity consumption. Conversely, economic growth seems to correlate with a reduction in final energy consumption that is reliant on diesel and gas oil. Both inflation and unemployment exhibit a negative correlation with electricity consumption and overall energy use, with the exception of energy utilized for heating purposes. FDI is associated with increased electricity consumption.

The ongoing conflict between Russia and Ukraine has dramatically reshaped the global energy markets, highlighting the vulnerability of many nations to energy price shocks and supply disruptions. This crisis underscores the urgency of diversifying energy sources, improving energy efficiency, and accelerating the transition to renewable energy. The conflict’s effects are likely to exacerbate existing inequalities, with low-income households and vulnerable populations bearing a disproportionate burden. The findings of this study, particularly regarding the impact of income inequality and the gender pay gap on energy consumption, become even more relevant in this context, as they can inform policies designed to mitigate the social and economic consequences of the energy crisis. For example, targeted support for low-income households to manage rising energy bills and investments in energy efficiency programs for vulnerable communities become even more critical in the wake of the conflict. Moreover, the conflict highlights the need for greater energy independence and resilience, which can be achieved through investments in renewable energy and decentralized energy systems. These investments can not only reduce reliance on volatile global fossil fuel markets but also create new economic opportunities and promote social equity.

This study is subject to some limitations. The period is short, especially for the variables related to final energy consumption in households by type, because of data availability for these indicators. The limited period does not allow us to consider an econometric model for each country. An overall image for all the EU-27 countries is provided, but a comparative analysis for old and new EU member states is not provided. Only a few control variables are included in the models. A few recommendations should be taken into account in future studies. For example, the period could be extended for only some indicators in the analysis (final energy consumption and electricity consumption). A separate analysis should be conducted for new EU member states and old EU member states to make comparisons and design the most suitable policies for each group of countries. Other variables might be considered in the models to replace GDP per capita (for example, human capital index).