Socioeconomic Impacts of Renewable Energy Plants Through the Lens of the Triple Bottom Line

Abstract

Investing in renewable energy is a key driver for achieving the objectives outlined in the 2015 Paris Agreement. In this context, Brazil has stood out, and this study investigates the socioeconomic impacts of different types of renewable energy projects across Brazilian municipalities. The analysis focuses on projects installed after 2010, from which investments in non-conventional sources have grown substantially in the country. The applied methodology combines Propensity Score Matching and Difference-in-Differences techniques to analyze the average impacts and source-specific effects of hydro, wind, and photovoltaic solar projects on GDP per capita and formal employment. The results indicate an average positive effect of 16.8% on per capita income, with wind power having the greatest impact, and 6.7% on formal employment, where hydropower stands out. Therefore, this work provides valuable insights for policymakers and companies, who can use the findings to make decisions and direct investments based on the various dimensions of the Triple Bottom Line.

1. Introduction

In the context of the 2030 Agenda established in 2015, the United Nations (UN) proposed the sustainable development goals (SDGs) [1] to ensure global growth based on the Triple Bottom Line (TBL) framework, i.e., social, environmental, and economic aspects [2]. As a result, this has made the sustainable development paradigm of prime eminence for governments and policymakers [2]. However, according to the 2024 SDG Report [3], the last progress assessment reveals that the world is severely off-track to achieve the 2030 Agenda. Out of 135 targets with trend data and additional insights from custodian agencies, only 17% are progressing as expected to be completed by 2030 [3]. As per the report, we must double down on those areas that can unlock transformative progress across the goals. Critical transitions around energy, food, and digital connectivity, underpinned by expanded access to social protection, decent jobs, and education and skills for the future, are essential for rewiring economies to combat the several planetary crises and to reduce inequalities within and between countries [1,3].

In this regard, the achievement of several SDGs heavily relies on investments in renewable energy sources (RESs), as access to RESs is a critical vehicle that drives sustainable development through the provision of basic needs, the realization of productive activities, and the protection of the local environment, thus generating improvements in livelihood conditions [1,4]. Da Silva et al. [5] emphasize that facilitating access to electricity directly combats a chain of adversities that affect the quality of life of communities and perpetuate the cycle of poverty. Moreover, since renewable energy consumption is instrumental in reducing CO2 emissions, policymakers should prioritize investments in clean energy, focusing on green growth and a green economy [6].

Although still insufficient, investments in renewable energies have increased worldwide in recent decades [7]. Latin America has followed this trend, with Brazil standing out, as its electrical matrix is predominantly composed of RESs (84.8%), and there is still significant potential to be explored [8,9]. Non-conventional sources, such as wind and photovoltaic solar (PV), distinguish themselves, having experienced the most growth in the country’s electrical grid over the last decade, according to the National Electric System Operator (ONS) [10]. Factors such as declining technology costs, government support, the demand for energy matrix diversification, and environmental commitments explain the exponential growth of non-conventional renewable sources in this period [11].

On the other hand, despite abundant natural resources available to support the country’s development, Brazil exhibits significant socioeconomic disparities, ranking as one of the most unequal countries in the world. According to the World Inequality Lab (WIL) [12], the poorest half of Brazil’s population holds less than 1% of the country’s wealth. Additionally, the minimum wage in Brazil is the second lowest in a list of 31 countries compiled by the Organization for Economic Co-operation and Development (OECD) [13].

It is also noteworthy that many Brazilian regions with high levels of underdevelopment and poverty, such as several states in the Northeast, also possess a vast availability of renewable natural resources to produce energy [14,15,16,17]. In light of this, it is valid to question how investments in the renewable energy industry can contribute to economic leverage and social change in countries with natural characteristics like Brazil, especially in regions with high poverty.

Assessing the impacts obtained by investments in RESs from different perspectives, such as environmental, social, and economic, is fundamental to society. Studies focused on this theme are crucial to supporting decision-making processes and guiding public policies, particularly those related to incentives for the sector’s development. They are also fundamental for evaluating how investments in a given technology contribute to achieving a country’s sustainable development goals.

In this sense, many works have focused on evaluating the impacts of projects related to sustainable development worldwide [4,18,19,20,21,22], as detailed in Section 2. But, although several works have analyzed their economic benefits, particularly from the perspective of profits and revenues, and their environmental benefits, mainly from the standpoint of greenhouse gas (GHG) emission reductions, the analysis from the social or socioeconomic perspective based on an econometric approach is less addressed [4,18].

It is important to note that socioeconomic change can be reflected through metrics that capture key aspects of economic performance and social well-being. Commonly used indicators include GDP per capita, employment levels, the Human Development Index (HDI), and the Gini Index [4,20,22]. These indexes provide tangible evidence of changes in living standards and economic activity over time. In assessing the impacts of renewable energy projects, they serve as proxies for identifying potential spillover effects in the regions where such plants are installed.

In Brazil, many studies on the socioeconomic impacts of RES projects—such as [4,19]—have not considered plants installed after 2010. However, most wind and PV power plants in the country became operational only in the subsequent decade [10]. More recent analyses, such as [22], have still not covered the latest socioeconomic data (from 2020 onward) or isolated the effects of different renewable technologies. Thus, systematically and widely assessing socioeconomic changes from renewable plants in Brazilian municipalities becomes essential, especially after the country’s significant penetration of non-conventional plants over the last decade.

Despite already having most of its electrical matrix composed of renewable sources, Brazil still has vast space and availability of renewable natural resources that can be harnessed for energy production [14,15]. Therefore, if the socioeconomic impacts are deemed relevant, they may inform decisions made by sector stakeholders. Public policies can be implemented to strategically deploy infrastructure in areas with renewable production potential but lacking the necessary installations. Tax incentive programs can be designed to channel investments in power plants to specific regions, aiming to improve their social indicators. Companies in the sector, concerned about their environmental and socioeconomic impacts, can leverage this information to improve their environmental, social, and governance (ESG) ratings. The work [23] highlights that organizations face increasing challenges to stay in the market, and they must adopt practices that not only are economically profitable but also impact the environmental and social dimensions.

In light of these points, the primary goal of this study is to assess if and how introducing renewable energy plants can deliver socioeconomic benefits in Brazilian municipalities. Using a data-driven methodology, it is investigated whether and to what extent the three main renewable sources in the country (hydro, wind, and PV) impact the GDP per capita and the employed population in cities that host these types of power plants.

The main innovations of this work, in relation to the existing literature, include the quantification of the impacts of renewable energy plants on the indicators aforementioned between 2010 and 2020—a period marked by a significant expansion of wind and photovoltaic power plants, which remains underexplored in the literature. In addition to estimating the aggregated average effects of renewable plants, the study also identifies the individual impacts of the three main renewable sources used in the country, providing more specific evidence. Finally, counterfactual scenarios are introduced, offering practical tools to probabilistically assess the potential evolution of socioeconomic indicators in municipalities with potential for renewable energy production.

This study is structured as follows. Section 2 presents the theoretical framework, the main gaps found, and, based on this, the research question. Section 3 comprises the material and methods. Section 4 presents the results, discussion, and innovation. Finally, Section 5 brings the conclusions and proposes future studies.

2. Literature Overview: Assessing Socioeconomic Impacts of Renewable Energy Projects

Evaluating the impacts of projects related to sustainable development worldwide has been the focus of many studies in the literature. Many of them have focused on projects classified as Clean Development Mechanisms (CDMs) [4,18,19,20,21], which are flexible mechanisms created by the Kyoto Protocol to reduce greenhouse gas (GHG) emissions or carbon capture [24]. There are clear and strict rules for approving projects under the CDMs, which are not limited to renewable and alternative energy plants but also involve projects related to energy conservation or reforestation [24].

More frequently, the focus has been on evaluating the economic benefits for investors or the environmental gains from GHG emission reductions, with social or socioeconomic perspectives less investigated [4,18]. When addressing the social dimension (the main focus of this study), the works can be classified into qualitative or quantitative, based on the analytical methods employed. Qualitative methods are generally based on expected effects from an ex ante analysis perspective, primarily derived from the Project Design Document (PDD), which details the projects before their construction [4,18]. Two works [20,21] conducted qualitative investigations focusing on CDM projects implemented in Latin American countries, and both underlined that they only analyzed the intended co-benefits for the sustainable development of the CDMs. It was not verified whether the projects delivered these expected co-benefits.

On the other hand, quantitative empirical methods are employed to measure the impacts after the project implementation (ex post analysis) [4,18]. The goal is to quantify the socioeconomic impacts ex post, generally utilizing statistical methods based on observational data to compare the evolution of response variables.

Table 1. Summary of some quantitative studies evaluating the socioeconomic impacts of RESs.

Study	Analysis Period	Methodology	Project Types	Study Area	Source-Specific Analysis	Year of Publication
[4]	2000–2010	Matching + DID	CDM: hydro, biomass, and other non-energy	Brazil	Yes	2019
[19]	2000–2010	DID	CDM: hydro, biomass, wind, and other non-energy	Brazil	Yes	2020
[25]	Simulated	I–O	Concentrated solar power	Brazil	Yes	2020
[26]	2004–2016	Matching + DID	Wind	Brazil	Yes	2020
[22]	2002–2018	Matching + DID	PROINFA	Brazil	No	2024
[27]	1990–2018	Various panel approaches	Any energy source	BRICS	No	2022
[18]	2002–2011	Matching + DID	CDM: biomass, wind, and solar	China	Yes	2019
[28]	2014–2022	I–O	Renewable energies in general	Portugal	No	2025

summarizes selected quantitative studies relevant to our context, including their analysis period, methodology, project types, study area, and whether they estimated isolated effects by renewable source (‘Yes’ in the column Source-specific analysis) or only aggregated impacts (‘No’).

The Difference-in-Differences (DID) technique is the most applied for assessing program implementation or treatment in a particular population of interest [4,18,19,22,26,29]. The main idea is to compare the change in social benefits connected to projects in places that adopted them (treatment group) with the change in social benefits in locations that did not (control group), using at least two time periods, one pre-treatment and one post-treatment [18].

Using a DID approach, Mori-Clement [4] applied the method to assess the statistical significance of differences in socioeconomic indicators between Brazilian municipalities that received or did not receive CDM projects, using data for the years 2000 (pre-installation) and 2010 (post-installation) regarding 204 projects. Results show significant but small effects of CDM projects on employment and income benefits. It is emphasized that the author did not consider the impacts of wind and solar PV sources in their work. Other studies considered complex equations with various explanatory variables to obtain realistic approximations of the DID estimator, aiming to measure the actual treatment effect on the variables of interest, such as [18,19,22,26].

In [18], the authors assessed the effects of CDM project implementation in China on indices such as local income and labor demand, using data from the first two decades of this century. They found that the biomass-based CDM projects significantly contribute to income improvement and employment generation in rural communities in China. The results also revealed that wind energy-based CDM projects have the potential to increase income and the share of labor force in the primary industry in rural areas. In [19], the method was applied to examine the impacts of 147 CDM projects on inequality and poverty indices in Brazil using data from 2000 to 2010. Although the authors analyzed source-specific effects, they did not address the PV source and, due to the period analyzed (2000–2010), were limited to a small sample of wind projects in the Brazilian context.

Gonçalves & Chagas [26] applied DID to estimate the impact of wind farms on the Brazilian labor market, suggesting that wind farms increase employment in the industry, agriculture and construction. However, the study is limited to the wind source and job creation analysis. Other research [22] employed DID to estimate the social effects of a Brazilian incentive program for alternative electricity sources (PROINFA). In general, the results indicated that the program contributed to a 10% rise in per capita income in municipalities, a 13.8% growth in formal workers, and a 0.39% increase in capital expenditure. However, the authors did not examine whether distinct energy sources impact the indices differently.

Although DID is widely used to analyze the long-term effects of policy implementation, there is a concern that it may suffer from two sources of bias. The first is related to the assumption of parallel trends between the treatment and control groups: an average change in the comparison (or control) group represents the counterfactual change in the treatment group if there were no treatment [4,30]. Another bias could arise if the project sites are not randomly assigned but determined by various geographical, political, and socioeconomic factors [18]. Although these assumptions present some difficulties in being tested, there are some alternatives for mitigating potential biases in applying the method, such as combining DID with matching approaches to pair treatment group sites with control group sites with similar observed attributes [18]. In this sense, for example, refs. [18,22,26] combined DID with propensity score matching (PSM). At the same time, ref. [4] utilized a variant of PSM known as kernel matching, and both achieved more robustness and reliability of DID results.

Other studies apply different data-driven methods, such as [25], which used data from a solar thermal power plant in Brazil to formulate local content scenarios using industrial analysis. These scenarios were assessed using interregional Input–Output (I-O) analysis to estimate the labor and economic output generated by installing and operating this facility at the regional level. Santos [28] also applied an I-O model to assess the socioeconomic impacts of renewable energy policies in Portugal, concluding that the energy transition has positively influenced employment trends, gender equity, and financial resilience. Raghutla & Chittedi [27] examined the empirical impact of access to electricity on economic development in the BRICS countries using panel econometric approaches. The results confirmed cointegration between economic growth, access to electricity, education, employment, inflation, and income.

In this context, some gaps are identified in analyzing the literature, mainly focusing on Brazil:

• Most wind and PV projects began operation after 2010 in Brazil, but studies evaluating the impacts of renewable energy plants installed after this year are limited in sample size or in the range of renewable sources considered;
• There are no evaluations of the impacts on socioeconomic indices recently measured, from 2020 onward, which would reflect the greater maturity of renewable plants installed in the country over the past decade;
• Considering the periods mentioned in the previous gaps, there is a lack of studies quantifying ex post isolated impacts of different renewable sources, such as hydro, wind, and PV, which together account for approximately 78% of Brazil’s electrical matrix [8]. It is essential for society planning to understand whether (and how) different green technologies can contribute to its development.

Based on the previous gaps and discussions, the following research questions are raised, which we aim to address in this study. Has the installation of renewable power plants after 2010 been impacting socioeconomic indices in Brazilian municipalities, such as GDP per capita and the employed population? If so, what is the magnitude of these effects? Do the three main renewable sources in the country (hydro, wind, and PV) impact the indices of interest differently, or are there no significant differences between the individual effects of the plants?

3. Materials and Methods

Figure 1 summarizes the step-by-step process implemented to achieve the objectives of this research, comprising stages of data pre-processing, processing, and post-processing. Although there is a logical sequence, note that the process is interactive, meaning it is possible to return to previous steps if the results are unsatisfactory at any point.

The applied methodology combines DID and PSM techniques. As presented in Section 2, this combination is widely used in studies whose objectives involve quantitatively evaluating the impacts of programs, technologies, and public or private policies on variables of interest. The details of each stage are provided subsequently.

3.1. Data Pre-Processing

The first step of data pre-processing consists of defining and collecting data on the study’s individuals or units of analysis. Rosenbaum & Rubin [31] classify the data used in longitudinal studies into two main groups: treatment variables (e.g., participation in a program) and outcome variables (e.g., income, employment), on which the treatment impacts are to be measured.

In this study, the analysis units are the 5570 Brazilian municipalities, and the treatment refers to installing hydro, wind, and PV plants. The outcome variables are GDP per capita and employed population, which are purportedly impacted by the intervention. Other potential response variables, such as the HDI and the Gini index, were unavailable in public Brazilian municipal-level datasets for the post-treatment period. Proceeding, GDP per capita is one of the indicators of a society’s standard of living, reflecting the average production of goods and services by its population [32]. The employed population is a fundamental variable for assessing a region’s economic and social performance. It reflects the labor market’s ability to absorb the available workforce, directly influencing household well-being and income levels [33].

In addition to treatment and outcome variables, this work considers a third group of data, named control variables (covariates), which comprise geographical, political, and economic information about the municipalities. These variables are used in the matching stage and as other possible explanatory variables in the DID model and were selected based on the literature [4,18,19] and the availability in public Brazilian databases. Additionally, such controls were also chosen because they allow for the adjustment of observable characteristics that have the potential to influence both the outcome and the likelihood of receiving the treatment, helping to isolate the causal effect of the intervention.

Table 2. Summary of variables.

Variable	Description	Source
Treatment Variables
Type of plant	Plant source: hydro, wind, or PV	[34]
Municipality	Municipality of installation	[34]
Start date	Operation start date	[34]
Installed capacity	Maximum production capacity of the plant	[34]
Covariates
Latitude	Latitude	[35]
Longitude	Longitude	[35]
Altitude	Altitude	[35]
Area	Area	[36]
Population	Population	[37]
Rural population rate	% of rural population	[37]
Illiteracy rate	% of population aged 15 years or older illiterate	[38]
Industrial GDP rate	% of industrial production in total GDP	[39]
Industrial production	Industrial production in monetary units	[39]
Bolsa Família spending	Total Bolsa Família transfer in monetary units	[40]
Outcome Variables
GDP per capita	GDP per capita	[41]
Employed population	Population in formal jobs	[42]

summarizes the data qualitatively, categorizing them into the three groups previously mentioned. Information about the data source is also provided.

In more detail, the municipalities’ covariates are related to characteristics that can be classified into three main groups: (i) geographical (latitude, longitude, altitude, and area), (ii) social demographic (population, rural population rate, and illiteracy rate), and (iii) economic (industrial GDP rate, industrial production, and Bolsa Família spending). The Bolsa Família is a Brazilian Federal social welfare program that provides direct income transfers to socially vulnerable families under certain conditions [40].

In the second step, it is essential to treat missing data and evaluate the application of techniques such as normalization or standardization, aiming to mitigate the influence of dimensionality on modeling. Additionally, as the data span over time, variables involving financial resources (industrial production, Bolsa Familia spending, and GDP per capita) must be adjusted for the cumulative inflation of the period [43].

The third step involves dividing the units of interest into treatment and control groups, and a time reference must be defined to establish pre- and post-intervention periods. To achieve this, the following premises are considered:

• The year 2010 was defined as the pre-treatment period;
• The year 2020 was defined as the post-treatment period;
• To be part of the treatment group, the municipality must have had plants (of any type) installed between the pre-treatment and post-treatment years;
• Municipalities that had any plant installed before the pre-treatment period were removed from the database, as they were ineligible for both control and treatment groups since the variables of interest could be impacted by the previously installed plants;
• Thus, after allocating the municipalities to the treatment group according to Premise 3 and eliminating those that meet the criterion of Premise 4, all other Brazilian municipalities are eligible for the control group.

The next step is analyzing the variables statistically, including applying hypothesis tests, such as the Kolmogorov–Smirnov test [44] to compare the variables’ distributions between the treatment and control groups before the intervention. This step is important to assess the magnitude of the difference between the groups in the pre-treatment period. Treatment and control need to be similar before the intervention to avoid future bias in the final results. However, the next step aims to address this issue.

3.2. Data Processing

The primary data processing steps comprise (i) matching the analysis units and (ii) modeling the longitudinal data, which are detailed, respectively, in Section 3.2.1 and Section 3.2.2.

3.2.1. Propensity Score Matching

The most commonly used method for matching units of analysis is the PSM, a quasi-experimental sampling technique proposed first by [31], that produces a control group whose distribution of a vector of covariates (X) is similar to that of the treated group. The idea is to find a similar control unit to compare with a treated unit, thus reducing selection bias and improving the balance between the treated and control groups [4,45]. Note that a potential control group was created based on Premise 5, which was presented earlier, for the division of the municipalities. The final control group is the result of this step.

To match treatment and control units based on X is equivalent to matching them using a propensity score, $p\left(X\right)$, which gives the probability of receiving treatment ($T=1)$ given the pre-treatment value of X, that is, $p\left(X\right)=Pr(T=1\mid X)$ [18]. Table 2 presents the covariates used to match the municipalities in this study. The method has two main stages:

• Propensity score calculation: estimating the probability (propensity score) of each individual being treated (or not) based on the pre-treatment covariates [46]. The technique most used is logistic regression [31,46], represented in (1).

  $$\mathit{p}\left(\mathit{X}\right)={\displaystyle \frac{1}{1+e^{-({\beta }_0+{\beta }_1{x}_1+{\beta }_2{x}_2+\dots +{\beta }_n{x}_n)}}}$$

  (1)

  where $\mathit{p}\left(\mathit{X}\right)$ is the propensity score, ${\beta }_0$ is the intercept, ${\beta }_i$ are the regression coefficients, and $x_i$ are the covariates. Although logistic regression is the benchmark method used in this step [31,46], other techniques can be applied, constituting PSM variations in propensity score calculation. For instance, logistic regression can be used for probit regression or machine learning algorithms such as decision trees, random forests, and neural networks [47,48,49]. Comparatively, logistic regression is a well-established method that provides a straightforward and interpretable way to estimate the propensity score, as well as being computationally efficient and able to handle large datasets [46,50]. So, this technique was selected for this work.
• Matching: encompasses selecting pairs of treated and untreated individuals with similar propensity scores previously computed, ensuring that the treatment and control samples are similar regarding the observed variables [46]. To match the treatment and control groups, finally obtaining this last, the nearest neighbor method is applied, broadly recognized for its simplicity of implementation and effectiveness [49,50,51]. Other variations of PSM include using different matching estimators, such as radius, kernel, stratified, and Mahalonobis [4,18,49,50].

After the pair selection, it is essential to certify the balance of the observed variables between the treatment and control samples [50]. In this sense, the standardized percentage bias (SPB) is used to compare the difference between treated and control samples regarding each considered variable, standardized by the group variation. The SPB formula is given by Equation (2).

$$SPB={\displaystyle \frac{{\overline{X}}_T-{\overline{X}}_C}{\sqrt{\frac{s_T^2+s_C^2}{2}}}}\times 100$$

(2)

where:

• ${\overline{X}}_T$ is the variable mean in the treated group;
• ${\overline{X}}_C$ is the variable mean in the control group;
• $s_T$ is the variable standard deviation in the treated group;
• $s_C$ is the variable standard deviation in the control group.

Additionally, the Kolmogorov–Smirnov test must be applied again to verify if the differences between the variables of the treatment group and the final control group have been reduced.

3.2.2. Differences-in-Differences

The DID method is widely applied when one wants to evaluate the impact of a program or treatment on an outcome variable over a population of individuals, such as the approval of a law, enactment of a policy, or implementation of a large-scale program [4,18,19,29,30].

In this study, it is considered a DID estimator combined with a mix of fixed effects that allows the control for unobservable municipality characteristics that may influence the variable of interest, such as local policies and other municipality-specific factors [18,52,53]. As highlighted by [53], using fixed effects helps eliminate bias that can arise from unobserved characteristics, especially when there is significant heterogeneity and unobserved variation between units of analysis, such as in this work. The DID structure can be represented by Equation (3).

$$Y_{it}={\beta }_0+{\beta }_1{G}_i+\gamma t+\delta G_{i}t+\sum {\beta }_j{X}_{it}+{\alpha }_i+{\epsilon }_{it}$$

(3)

where:

• Each municipality is indexed by the letter $i=1,\dots ,N$, where N is the total number of municipalities;
• There are two different periods, one pre-treatment, where $t=0$, and one post-treatment, where $t=1$;
• $Y_{it}$ is the outcome variable of municipality i at time t;
• $G_i$ indicates the group of municipality i, i.e., if $G_i=0$, it belongs to the control group, and, if $G_i=1$, to the treatment group;
• $X_{it}$ is a vector of control variables (covariates) that possibly affect the outcome variable of municipality i at time t;
• ${\epsilon }_{it}$ corresponds to a random variable that represents the model error for municipality i at time t;
• ${\alpha }_i$ is the fixed effect of municipality i, used to capture and control for unobserved characteristics that are constant over time.

Thus, the coefficients can be interpreted as follows:

• ${\beta }_0$: intercept or constant term;
• ${\beta }_1$: specific effect of the treatment group, to account for the permanent average differences between the control and treatment;
• $\gamma$: common time trend for the control and treatment groups;
• $\delta$: known as the DID estimator, representing the interaction effect between the treatment and time, i.e., the true treatment effect, in which our main interest lies;
• ${\beta }_j$: effect of each covariate, where $j=2,\dots ,n+1$, considering n covariates.

It is worth noting that the usual application of DID does not consider control variables X; in this work, they are supposed to increase the results’ robustness and confidence.

Alternative model specifications with different treatment and control variables are considered. In more detail, four models are estimated to analyze the effects of RESs from various perspectives for each outcome variable—GDP per capita and employed population. One model evaluates the average impacts of installing any RESs, without distinction by source, and therefore includes all municipalities of the treatment group. The other three models aim to investigate the individual effects of each renewable energy analyzed—hydro, wind, and PV.

3.3. Data Post-Processing

In this stage, the final results are assessed and interpreted. The outputs of all estimated models are presented and evaluated, considering statistical techniques such as parameter significance tests, confidence intervals, coefficients of determination, and standard errors of the parameters. The implications of the results are also related. Furthermore, assessing whether the previously defined objectives have been achieved is crucial. Revisiting earlier stages may be required to address any issues identified.

4. Results and Discussion

All methodology steps were developed using the R programming language [54]. The scripts and data used are available to the public as highlighted in Data Availability Statement.

4.1. Data Pre-Processing

After the data collection, six of the 5570 municipalities (0.11%) were eliminated from the study due to missing data. The treatment variables encompass Brazil’s three most relevant types of renewable energy plants (hydro, wind, and PV) and were used to divide the control and treatment groups, applying the premises presented in Section 3.1.

Thus, 244 units were allocated to the treatment group and 4720 to the potential control group. Figure 2 details the number of municipalities that received new projects by RES and region during the treatment period, and for the first time. Brazil is divided into five geographical regions: Northeast (NE), Southeast (SE), South (S), Midwest (MW), and North (N). Most of the Brazilian municipalities that received wind and PV installations are in the Northeast, 85% and 71%, respectively, highlighting the region concerning these RESs. Meanwhile, for the hydro source, the majority of the municipalities are in the South (47%) and Southeast (31%).

The average installed production capacities per municipality and source were 280.9 MW (wind), 173.6 MW (PV), and 89.6 MW (hydro). The hydro number is comparatively lower. Despite remaining the predominant renewable source in Brazil, hydropower’s share has been declining in the electrical matrix of the country (from over 70% in 2014 to around 51% in 2024) [8]. Recent investments have focused on small- and medium-sized plants, driven by environmental sustainability concerns, as large projects require storage reservoirs that significantly impact the environment and local communities [55]. Proceeding with the treatment group analysis, only three municipalities, all located in the Northeast region, received more than one type of RES, specifically wind and PV. None received the three types of projects.

Table 3. Descriptive statistics of the covariates during the pre-treatment period.

		Potential Control (4720 obs)		Treatment (244 obs)
Variable	Unit	Mean	SD	Mean	SD
Latitude **	Decimal degrees	−45.99	6.49	−46.11	6.55
Longitude ***	Decimal degrees	−15.84	8.31	−17.84	8.73
Altitude ***	m	398.64	281.64	488.41	324.49
Area ***	km2	1452.99	5405.64	1586.49	3661.76
Population *	Number	30,040	134,597	31,753	58,656
Rural population rate	%	0.369	0.219	0.396	0.230
Illiteracy rate **	%	0.170	0.100	0.151	0.092
Industrial GDP rate **	%	0.124	0.127	0.134	0.123
Industrial production ***	Million BRL	134.101	872.709	115.214	408.661
Bolsa Família spending	Million BRL	2.532	6.622	2.798	4.404

Note: (1) * (p < 0.1), ** (p < 0.05), and *** (p < 0.01) indicate that the distributions of treatment and control groups are different with statistical significance according to the Kolmogorov–Smirnov test.

presents the covariates’ descriptive statistics during the pre-treatment period. According to the Kolmogorov–Smirnov test [44], most of the variables exhibit different distributions between the two groups, underscoring the need for applying the matching method.

The outcome variables analyzed in this study are ‘GDP per capita’ and ‘Employed population’. It should be noted that there is a delay in the release of the 2020 Brazilian demographic census, which is the source of other socioeconomic indicators covering the entire national territory. Therefore, this study is limited to the indicators available in the post-treatment period. Since the research aims to compare the social benefits before and after the plants’ installation, a statistical analysis of the outcome variables is conducted after balancing the control and treatment groups in Section 4.2, encompassing both periods, 2010 and 2020.

4.2. Propensity Score Matching

Figure 3 summarizes the results of the PSM application, presenting the standardized percentage bias of the covariates and propensity score before and after matching. Most variables’ bias was significantly reduced after balancing, aligning more closely with the vertical zero line. The bias reduction in the propensity score is the most notable, indicating that the matching effectively balanced the groups. Thus, after the PSM application, the control group included 244 municipalities, the same size as the treatment group.

Table 4. Descriptive statistics of the covariates during the pre-treatment period after matching.

		Control Group (244 obs)		Treatment Group (244 obs)
Variable	Unit	Mean	SD	Mean	SD
Latitude *	Decimal degrees	−45.64	5.60	−46.11	6.55
Longitude *	Decimal degrees	−17.40	8.11	−17.84	8.73
Altitude ***	m	495.82	309.06	488.41	324.49
Area *	km2	1534.32	8462.00	1586.49	3661.76
Population ***	Number	23,072	57,174	31,753	58,656
Rural population rate ***	%	0.394	0.210	0.396	0.230
Illiteracy rate ***	%	0.153	0.091	0.151	0.092
Industrial GDP rate ***	%	0.127	0.119	0.134	0.123
Industrial production **	Million BRL	64.092	228.565	115.214	408.661
Bolsa Família spending ***	Million BRL	2.298	4.417	2.798	4.404

Note: (1) * (p > 0.01), ** (p > 0.05), and *** (p > 0.1) indicate that the distributions of treatment and control groups are the same with statistical significance according to the Kolmogorov–Smirnov test.

presents covariates’ descriptive statistics during the pre-treatment period after matching. According to the Kolmogorov–Smirnov test [44], now all of the distribution variables can be statistically considered the same.

Table 5. Pre-treatment and post-treatment means of the outcome variables.

		Control Group (244 obs)		Treatment Group (244 obs)
Variable	Unit	2010	2020	2010	2020
GDP per capita	BRL *	18,360.91	23,662.96	20,668.66	30,712.21
Employed population	Number	3835	4178	6702	7457

Note: * monetary variables are in Brazilian Real (BRL). To convert to USD, use the 2020 average exchange rate of 5.16 BRL/USD [56].

presents the means of the outcome variables after matching for pre- and post-treatment periods. The GDP per capita increased by about 29% in the control group, while, in the municipalities that received RES plants, the growth was approximately 49%. Although the increase was less pronounced in the employed population, it was higher in the treated sample, with a rise of 11.3% compared with 8.9% in the control group. Therefore, the DID method, whose results are presented in Section 4.3, aims to estimate whether and to what extent these gains are due to the treatment and other possible effects.

4.3. Difference-in-Differences Models

As defined in Section 3.2.2, different DID models were estimated for each outcome variable to assess the general and specific impacts of the renewable sources on the evaluated indicators.

Table 6. GDP per capita models.

	Model
Variable	(1)	(2)	(3)	(4)
RES	3476.71 ***
	(1257.46)
Wind		5745.78 **
		(2714.23)
PV			3754.06
			(2570.14)
Hydro				3030.71 **
				(1236.35)
Time	6310.75 ***	5493.88 ***	6494.99 ***	5880.17 ***
	(836.00)	(1116.63)	(1816.50)	(841.54)
Bolsa Família spending	395.52 **	541.45	319.59 *	332.98
	(162.03)	(456.23)	(185.91)	(236.59)
Industrial production	30.66 ***	40.08 **	21.26 ***	29.05 **
	(9.84)	(17.99)	(7.85)	(13.00)
N	976	280	84	600
Adj. $R^2$	0.732	0.480	0.798	0.810

Note: (1) * (p < 0.1), ** (p < 0.05), and *** (p < 0.01) indicate that the coefficient is statistically significant (t-test). (2) Robust standard errors between parentheses.

presents the results for GDP per capita. There are four models in total. The first one considers installing any RES plant as the treatment, encompassing all municipalities from the treatment and control groups, each with 244 units of analysis.

The other models consider each renewable source individually as the treatment, i.e., wind (model 2), PV (model 3), and hydro (model 4), each with, respectively, 70, 21, and 150 treated municipalities and, consequently, the same numbers as the control. The three municipalities with more than one plant type were considered only in model (1) since the sample is insufficient to estimate a fifth model considering only places with more than one RES.

In Table 6, each renewable variable (RES, wind, PV, and hydro) corresponds to the DID estimator of the model; in other words, the effect of the treatment over time. Notably, without distinguishing the source type (model 1), installing a RES plant has a positive and significant effect on GDP per capita, contributing to an increase of 3476.71 BRL, on average, compared with untreated municipalities. Considering that the monthly minimum wage in the country was 1039 BRL in 2020 [57], the average impact on GDP per capita is significant, amounting to more than three times that value. Moreover, it represents an average increase of 16.8% compared with the pre-treatment period.

Each source’s specific effects are positive, highlighting the wind plants, whose average effect is 5745.78 BRL more on GDP per capita. Although more prominent than RES plants in general (model 1), the average effect of PV plants has a relatively high standard error, making the coefficient insignificant according to the t-test [58]. However, it is essential to emphasize that the sample for the photovoltaic model is relatively smaller than the others, including only 21 treated municipalities; therefore, a larger sample is needed for more robust results. The DID coefficients of the other three models are statistically significant based on the t-test for various significance levels, as indicated in Table 6. Furthermore, confidence intervals can be constructed from the standard errors provided in parentheses. For example, the values [1414.48; 5538.94] constitute the 90% confidence interval for the DID estimator of model (1), which allows us to infer that the average treatment effect lies between these values with this confidence level.

The positive effects are explained through various channels. Developing renewable energy projects requires private and public investments, enhancing local infrastructure and stimulating the municipal economy. These investments increase the capacity to produce goods and services, directly contributing to the rise in GDP per capita. Growth occurs not only through energy generation but also due to the positive impact the installation of these plants can have on other economic sectors, such as real estate, commerce, and service provision, which also benefit from improvements in energy infrastructure.

The most prominent impact of wind projects can be interpreted by the larger average size of wind plants compared with other sources, in terms of installed capacity, as presented in Section 4.1. In the past decade, wind energy was the fastest-growing source in the country [59]. The larger the installed capacity of power plants is, the more energy they tend to produce, resulting in a greater impact on GDP per capita. Large land areas are also leased for wind project installation, significantly affecting the local real estate sector [60].

The remaining coefficients of the models represent the common time trend for the control and treatment groups and two covariates: Bolsa Família spending and industrial production, both in millions of reais (BRL). These covariates were selected because they could influence the response variable [61]. The common time trend and industrial production are positive and statistically significant in all models. Although Bolsa Família spending presents positive values in all cases, its coefficients are not always statistically significant, as seen in the wind and hydro cases. This can be explained by the variability in the benefit distribution among the municipalities in these samples.

Regarding the adjusted $R^2$, the models generally exhibit substantial values for this metric, with most close to 0.80, indicating a significant proportion of GDP per capita variability explained by the selected models. Finally, we emphasize that the fixed effects account for the intercept by absorbing it within the entity-specific effects. This approach allows for the control of unobserved heterogeneity by capturing the intercept variations unique to each municipality, thereby improving the accuracy of the estimated coefficients for the other variables.

The other indicator analyzed is the employed population. The employment opportunities of renewable plants can be classified into three groups: (i) direct, involving activities related to the plant’s construction, operation, and maintenance; (ii) indirect, created in the supply chain and support services, such as suppliers of materials, transportation, and logistics services; and (iii) induced, created due to the increase in demand for local goods and services [62]. It is important to highlight that the aim is to analyze aggregated impacts rather than the specific effects on each type of employment.

The models of the employed population are presented in

Table 7. Employed population models.

	Model
Variable	(1)	(2)	(3)
RES	451.84 **
	(231.10)
Wind		353.74
		(249.44)
Hydro			622.74 *
			(341.94)
Time	447.30 *	144.19	492.63
	(265.97)	(176.71)	(468.61)
Bolsa Família spending	48.72	13.67	142.41
	(110.84)	(55.48)	(254.07)
Industrial production		1.64
		(1.30)
N	976	280	600
Adj. $R^2$	0.984	0.994	0.983

Note: (1) * (p < 0.1) and ** (p < 0.05) indicate that the coefficient is statistically significant (t-test). (2) Robust standard errors between parentheses.

. Similarly to GDP per capita, they were estimated with fixed effects for the same reasons as the GDP case. The first (1) considers all treated municipalities without distinguishing by source; the remaining were estimated for each plant. However, the model calculated for the photovoltaic project’s sample is not presented, as it yielded nonsensical coefficients, occasionally negative, due to relatively lower growth in municipalities that received PV plants compared with the specific control group for this source. It is noteworthy that, in addition to the relatively small photovoltaic sample, it was necessary to analyze these cases individually to elucidate the particular reasons for the observed discrepancy, which is beyond the scope of this study.

Analyzing the results, municipalities receiving any RES projects experienced an average increase of 451.84 in the employed population, which corresponds to a 6.7% rise. Specifically, the impact on municipalities receiving wind plants is an average increase of 353.74 formal jobs compared with the control group. The highlight is for the hydro source, where the DID estimator indicates an average of 622.74 more people employed. It is emphasized that the DID estimators for models (1) and (3) are statistically significant. However, the same cannot be said for the wind model, whose standard error for the treatment effect over time is relatively high.

In short, implementing renewable energy projects can increase municipalities’ labor demand. In this sense, the construction/installation phase is the most labor-intensive, offering temporary and diverse employment opportunities [60,63]. In contrast, the operation and maintenance phases typically experience reduced labor demand, focusing on technical and often automated activities [63].

Regarding specific effects, we observe a relatively greater impact from hydropower. One of the possible reasons for this is that, in general, the construction cycle for hydroelectric plants tends to be longer than for wind power, particularly due to complex activities such as building dams, canals, transmission infrastructure, and site adaptation [64,65]. It is important to emphasize that these estimated effects represent the average impacts of the plants. However, a case-by-case analysis would be needed to capture the specificities of each municipality, which falls outside the goals of this research.

The other variables considered are the same as those in the GDP per capita models. The common time trend for the control and treatment groups has a positive and statistically significant effect in the general model (1). Furthermore, although positive, the coefficient for the Bolsa Família variable does not reach statistical significance. Meanwhile, industrial production exhibited a negligible coefficient in models (1) and (3), reducing the adjusted $R^2$; thus, it is not presented for these cases. Regarding the adjusted $R^2$, the metric is notably high, consistently above 0.98, demonstrating a significant proportion of the variability of the dependent variable explained by the models.

4.4. Implications and Innovations

All the research questions formulated at the end of the Literature Overview (Section 2) are answered with the results of this work. In summary, without distinguishing between the plant types, the average effects on the social benefits evaluated are positive and significant: 16.8% on GDP per capita and 6.7% on formal employment. In the most recent study found in the literature analyzing Brazilian data up to the time of this work, Nunes et al. [22] estimated a 10% rise in per capita income and a 13.8% increase in the number of formal workers, focusing on projects under the PROINFA program. While the results are also positive, the differences can be explained by different sets of projects and periods. That is, we include plants beyond those linked to PROINFA and consider indicators measured from 2020 onwards, reflecting the greater maturity of renewable plants installed in the country over the past decade.

Moreover, although [22] estimated overall average impacts, the authors did not calculate the specific effects of different renewable plants. In this context, the present work highlights the impact of wind farms on GDP per capita and hydropower on the employed population, as well as the different channels that explain the outcomes. As for the PV source, a larger sample is necessary to obtain more generalized and robust results. Nevertheless, in this context, ref. [25] estimated that a solar thermal power plant installed in Brazil could generate between 45 and 61 jobs (direct, indirect, and induced) per MW installed.

Compared with other works focusing on Brazil, ref. [26] suggests that wind farm implementation may be associated with a 27% increase in employment and a 34% increase in wages, figures more optimistic than those found in the present study and in [22]. The work [4] estimated socioeconomic impacts of hydro projects and, according to the analyzed sample, the author did not find significant effects on labor/income. At the same time, ref. [19] pointed out that hydro projects are associated with a reduction of 1.6% in the percentage of individuals in poverty. However, this finding is not directly comparable to our results, since we are not evaluating this indicator in the present research. The authors also estimated that wind farms are associated with a reduction of 2.1% in unemployment and a slight decrease of 0.5% in income derived from work (opposite to the direction expected).

It is important to underscore that the demand for labor in renewable energy plants generally decreases after the start of operations and, even post-installation, often requires highly specialized and qualified labor, frequently sourced externally—particularly in many rural cities in Brazil [63]. This highlights the need for more investment in local workforce qualification and better distribution of benefits from the renewable energy industry.

In this sense, research in this area aims to stimulate sustainable development and support the implementation of public policies encouraging local industry, as it quantitatively demonstrates its potential. Brazil, an underdeveloped and less industrialized country, has great potential to develop a high-tech production chain linked to the non-conventional energy sector. Based on the insights from this study, considering the specific development objectives of each location, public policies and tax incentives can be used to strategically implement infrastructure and install power plants in municipalities with renewable energy potential but lacking the required facilities. However, compensatory policies should also be implemented to ensure that the benefits generated are more equitably distributed among the local population, including creating professional qualification programs and reducing energy costs for the community. Still, from the results of this research, companies in the sector can leverage the presented results to demonstrate different impacts that can be generated by their actions, improving their ESG ratings and showing commitment to social responsibility and long-term sustainability.

Furthermore, this work’s results can be used to construct probabilistic counterfactual scenarios related to socioeconomic benefits in municipalities that have not received renewable power plants but have the potential to do so. Figure 4 and Figure 5 present the cases of GDP per capita and employed population, respectively. Considering the control group, both show their respective index real evolution, the average counterfactual scenario, and the 90% confidence interval of the counterfactual scenarios. The intervals were constructed using the robust standard errors of the DID estimators, as exemplified by the GDP per capita case from model (1) in Section 4.3. In more detail, to simulate the counterfactual scenarios for each case, the mean values of the DID estimator, along with their respective lower and upper confidence interval bounds, were added to the actual values of the control group in the post-treatment period. Notably, the real values are mostly below the counterfactual confidence intervals. Therefore, these probabilistic forecasting scenarios can be used to estimate, on average, the evolution of the indices had such municipalities received investments in renewable plants.

Regarding the main innovations of this study, we can list:

• The impacts on the GDP per capita and employed population resulting from investments in renewable energy plants in Brazilian municipalities were quantified after the installation of these projects, analyzing the evolution of these indices between 2010 and 2020. It is important to note that most of Brazil’s wind and solar power plants were installed in the last decade, and there is a lack of studies evaluating their impacts on indices from 2020 onward. This study, therefore, provides relevant results on the effects of the significant integration of non-conventional plants over the last decade on recently measured indices.
• Not only were the overall average effects of renewable plants (without distinction by source) on the indicators estimated, but also the individual effects of the three primary renewable sources in the country (hydro, wind, and PV), which provide insights and results not found in the literature.
• A robust observational data method was applied in a generalized and systematic manner to all Brazilian municipalities. As of the publication of this research, no studies have been identified that simultaneously address the specific levels of coverage—spatial, temporal, and renewable types—considered in this study.
• Counterfactual scenarios were introduced, providing practical tools to assess how socioeconomic indicators could probabilistically evolve in municipalities with potential for renewable energy production.

Notably, potential endogeneity issues were addressed by using DID with fixed effects, controlling for unobserved factors of the municipalities that could be correlated with both the treatment and the dependent variables. Additionally, using different covariates, common time trends, and group-specific effects in the DID models further reduces the risk of bias due to omitted variables.

Finally, potential limitations include the small sample size for solar plants, which hampers the robustness of results for this source compared with others. Another point is that this study focuses on Brazil’s specific regions and characteristics. Nevertheless, the methodological framework can be applied in other national contexts—with the necessary adaptations—to assess the consistency of socioeconomic impacts across different renewable energy strategies and institutional settings. In this sense, the availability of public data on the three required groups of variables (treatment, covariates, and outcome) is a key prerequisite. Furthermore, although the model incorporates various mechanisms to address endogeneity issues, the response variables may still be influenced by uncontrolled external factors, such as economic crises or changes in the labor market. Nonetheless, it is important to highlight that the purpose is to estimate the general average effects and those specific to each type of plant. Examining the particularities of each municipality lies beyond the scope of this work.

5. Conclusions

This study aimed to quantitatively measure the effects of Brazil’s three most important RES plants (hydro, wind, and PV) on socioeconomic indices, considering data from all the country’s municipalities. A data-driven methodology based on PSM combined with DID was implemented. The methodology was designed to approximate the real effects robustly. In this sense, the PSM method was applied before DID to create comparable groups (treatment and control) with similar characteristics. Several variables, including geographic, demographic, and socioeconomic factors, were used in the modeling. Then, DID was employed to measure changes in the indicators before and after the treatment, isolating the impacts of the intervention.

The results are promising and generally indicate positive effects on the evaluated indices. Therefore, the findings have significant public impacts, and policymakers can develop strategies to stimulate local industry and attract investments. In addition, the findings offer valuable insights for energy companies committed to social responsibility. Furthermore, the research contributes to several Sustainable Development Goals (SDGs) established by the United Nations, such as SDGs 7 (Affordable and Clean Energy), 8 (Decent Work and Economic Growth), 10 (Reduced Inequality), and 13 (Climate Action).

Finally, we propose for future research: (i) considering the impacts of other RESs, such as biomass; (ii) evaluating impacts on other social variables, such as the Municipal Human Development Index (MHDI) and the Gini Index, as they are released; (iii) applying different matching methods, such as radius, kernel, or stratified techniques; (iv) analyzing specific effects on each type of employment (direct, indirect, and induced); (v) applying statistical resampling techniques, such as bootstrapping, to expand the photovoltaic sample and improve the generalizability of results; and (vi) developing an indicator based on the TBL to assist the decision-making process regarding the prioritization of regions for allocating resources related to the renewable plants’ installation.