Gender empowerment and energy access: evidence from seven countries

Gender equity is connected to modern energy services in many ways, but quantitative empirical work on these connections is limited. We examine the relationship between a multi-dimensional measure of women’s empowerment and access to improved cookstoves, clean fuels, and electricity. We use the World Bank Multi-Tier Framework survey datasets from seven countries that include almost 25 000 households in Africa and Asia. First, we apply principal component analysis to construct a household level empowerment index, using data on women’s education, credit access, social capital, mobility, and employment. Then, we use simple regression analysis to study the correlation between empowerment and energy access at the household level. We find a positive association between the women’s empowerment index and energy access variables, though this household pattern does not hold across all countries and contexts. While we do not claim that these relationships are causal, to our knowledge this is a fresh analysis of how the empowerment of women is differentially correlated with household energy access across geographies and technologies. Thus, our analysis provides a first step to further work aimed at clarifying gender-energy linkages.


Introduction
Lack of access to modern energy services plays an important role in gender-based cultural and economic inequalities. First, women typically spend more time obtaining and using cooking fuels in addition to performing other household chores [1]. Studies show that women who spend less time on such tasks can enjoy increased leisure time [2], income generation and productivity [3][4][5][6][7]. Further studies show that reductions in energy poverty are associated with greater employment opportunities and shares for women, as well as decision making about household expenditures [8,9]. Second, the health costs of solid fuel collection and use are large and disproportionately borne by women due to gendered cooking norms. For example, women spend more time breathing in air pollutants from cooking with solid fuels, with myriad health damages [10][11][12]. Carrying firewood regularly can also physically damage the body [13].
Although the unequal burdens of energy poverty are pervasive in many rural low and middle income regions, women's agency in household energy choices is diminished by cultural norms and the low market value that is often placed on women's labor [14]. This in turn likely diminishes the uptake of technologies that could reduce such burdens [15]. The men, who commonly make decisions on major purchases tend to under-invest in energy poverty-reducing technologies, maintaining asymmetric burdens on women [16,17]. Therefore, women's empowerment can influence the uptake and use of improved energy services; concurrently, the use of improved energy services can improve women's societal standing.
Previous studies have considered many aspects of the complex associations between women's empowerment and use of improved and clean energy technologies. Analysis have considered various levels of these associations ranging from the national to household levels. For example, Nguyen and Su (2021) consider how access to clean fuels and electricity in 51 countries contributes to various individual measures or proxies of empowerment, including employment in different jobs and education levels of women relative to those of men [8]. Other studies frame their analyses of energy-empowerment relationships in the opposite direction, i.e. analyzing how empowerment affects energy access. For example, a household level analysis by Choudhuri and Desai (2020) suggests that women's access to salaried work and control over household expenditure decisions increases use of clean fuel in India [9].
There is also considerable discussion in the literature of which measures of energy access and empowerment to use. Pachauri and Rao (2013), for example, detail the multi-dimensional nature of both energy services and empowerment, providing arguments both for and against inclusion of characteristics such as labor force participation, head of household status, and access to groups in the measurement of empowerment [18]. As noted in Burney et al (2017), who create a multi-dimensional empowerment index specific to rural Benin, the tradeoff between a singular general index and context-specific empowerment measures is that the former facilitates comparative measurements, but at the cost of loss of cultural and other local specificities [19].
In this context, a major goal of our study is to test for consistency in the relationships between empowerment and energy access variables across a wide range of low-income countries. Our specific contributions are first, to use data from the Multi-Tier Framework (MTF) surveys, a rich energy services dataset, to first construct a single empowerment index that includes both intra-household and intrasocietal measures of women's empowerment highlighted in the prior literature, subject to data limitations. Second, we analyze patterns of co-variation of basic measures of household energy/electricity technology access and women's empowerment, both pooled and across a set of seven relevant countries, where energy poverty and gender inequality both remain substantial. While we perform cross country comparisons, at its core this is a household level analysis, specifically to analyze how women's agency is related to household level energy decisions.
While we are not the first paper to use the MTF datasets to study energy access or empowerment, to our knowledge, prior work has not used these data to construct a household level index of empowerment, to examine multiple correlations between empowerment and different energy/electricity access or to consider how the household level relationship between these variables varies by geography and technology. Our analysis is not causalwe do not attempt to isolate how energy access leads to empowerment, or conversely how empowerment leads to enhanced energy access. Rather, the patterns we uncover across 25 000 households in seven countries represent associations between these variables, both in simple bivariate terms, and alsoin multivariate analyses-controlling for a set of other covariates.

Data
We use a new energy-focused, multi-country data set, the MTF survey, to examine the household level association between measures of women's empowerment and energy access. The MTF surveys were developed to support tracking of Sustainable Development Goal 7: Universal access to modern energy [20]. We leverage the household, community, and local institutional surveys. We focus on the countries for which complete data on a core set of empowerment and energy access variables are available (Cambodia, Ethiopia, Myanmar, Nepal, Rwanda, São Tomé and Príncipe (STP), and Zambia).

Empowerment measure
Empowerment is a multi-dimensional concept that incorporates economic, political, social and psychological dimensions [21]. The MTF household surveys include demographic characteristics of household members, measures of women's access to credit, women's membership in various community groups, and their mobility. We first examine different proxies for women's empowerment that have been proposed in the literature including: 1. An indicator for a female household head: 1 if the household head is a woman, and 0 otherwise; 2. The share of adult women in the household (defined as the ratio of the number of adult women to the total number of adults in the household); 3. An adult women's education measure (average of the total educational attainment, measured in years, by adult women in the household); 4. The share of women employed in paid work (defined as the ratio of adult women employed in paid work to the total number of adult women in the household); 5. Women's reported mobility (an aggregate measure of three sub-measures of mobility pertaining to visits to family or friends, markets or economic centers, and to locations outside their village); 6. An indicator for women's access to various social groups or activities: 1 if the main female survey respondent was a member of any of a set of key groups or activities, and 0 otherwise; and, 7. An indicator for credit access: 1 if the female respondent had her own bank account, 0.5 if she had a joint account, and 0 if she did not have a bank account. A more detailed description of these variables is provided in table B01 in supplementary material B. Table 1 presents the mean (across households) for each these seven indicators for the overall sample and in each country and reveals considerable variation within and across our pool of seven countries.
Across our sample, the share of female household heads is 25%, ranging from 20% (Nepal) to 30% (in STP). Average educational attainment for adult women ranges from 0.5 (Rwanda) to 3 (Myanmar) years. Likewise, women's participation in paid occupations differs within and across countries; it is the highest being in Cambodia (60%) and the lowest in Ethiopia (28%). Both Cambodia and Rwanda have more than 50% of women employed in paid work. The share of adult women in the household varies between 35% and 43%. Mobility is highest in Myanmar and lowest in Rwanda, whereas the share of women participating in social groups or activities is highest in Nepal (41.7%) and lowest in Rwanda (3%). Access to credit (bank accounts) is highest among women in Nepal and lowest in Myanmar. Figure 1 presents the respective z-scores for these variables, which helps to facilitate comparison of estimates from each country-specific model. Figure 1 shows that Rwanda and Zambia stand out as having relatively low levels of these individual measures of women's empowerment, while Nepal has the highest overall levels in the sample, mostly owing to high access to credit in the latter country. Nonetheless, the small number (of seven) countries in the sample limits how much we can say about cross-country differences. The estimates for rural and urban samples are presented in supplementary tables A1 and A2, respectively.
For our statistical analysis, we construct a household level women's empowerment index (E) using the seven measures of women's empowerment as follows. First, we standardize the values of the indicators by generating z-scores for each indicator using the pooled country sample means and standard deviations. Second, we employ principal component analysis to construct an empowerment index using these z-scores in the pooled dataset (details are provided in tables A3-A6 in supplementary material A). We use the value of the first component as the weight for constructing the index. Third, we normalize that value such that it ranges from 0 (minimum empowerment) to 100 (maximum empowerment). Figure 2 depicts the empowerment index aggregated by country. In the overall sample, STP has the highest empowerment index value (6.4), and Rwanda has the lowest value (4.1). This unsurprising result is in line with the breakdown of individual component variables shown in figure 1, though with some reweighting (table A6). For instance, STP has the highest share of female household heads, which is relatively heavily weighted, as well as the second highest value for the mobility measure. Meanwhile, Nepal has the highest score for credit access (which is less heavily weighted), and the second highest share of women in the household (which is in contrast more heavily weighted) and social group membership (which is less heavily weighted). Similarly,  Myanmar has the highest average educational attainment among women (relatively high weight), the highest score for mobility (relatively low weight), and the highest share of women in the household (relatively high weight). While for some of these inbetween countries, the relative weighting switches their relative rankings, Rwanda is clearly at one extreme. Its low overall index is likely because Rwanda has the lowest average female educational attainment, the lowest mobility measure, the lowest social group membership, low women's access to credit and a low share of women in the household. (See table A7 for empowerment index averages for the overall, rural, and urban samples in each country).

Energy technologies 2.3.1. Cooking energy technology
For cooking energy, we consider the primary fuel source and the main cookstove reportedly used by households to determine whether the household relies on an improved cookstove (ICS) or a clean fuel (clean cookstove, hereafter) for most of its cooking needs. We use the method described in Krishnapriya et al 2021 to classify cooking fuels as clean or polluting, and cookstoves as ICS or otherwise [1]. In particular, we construct two variables for use of ICS and clean cookstoves. ICS = 1 if the household primarily uses an ICS and 0 otherwise. Likewise, the clean cookstove = 1 if the household primarily uses a clean fuel, and 0 otherwise.
In table 2, we report the primary use of ICS and clean stoves in the overall sample. The share of households that use ICS is highest in Cambodia (65.2%), followed by Myanmar (63%). Ethiopia (10.3%) and Zambia (13.1%) have the lowest use. Similarly, Cambodia has the highest share of clean cookstove users (42.4), while Rwanda has less than one percent of households using a clean cookstove, followed by STP (1.2%), and Ethiopia (3.7%). Tables A9 and A10 present the ICS and clean cookstove use in the rural and urban sub-samples. As expected, in every country in our sample, the use of ICS and clean cookstoves is higher in urban regions. As in the overall sample, Cambodia and Myanmar have the two highest shares of rural households using ICS and clean cookstoves. Ethiopia has the smallest share of rural households using ICS and clean cookstoves. On the contrary, in the urban sample, the share of households using ICS is highest in Myanmar, while Cambodia has the highest share of households using clean cookstoves. Ethiopia has the smallest share of urban households using ICS and STP has the smallest share of urban households using clean cookstoves.

Electricity
For electricity, we use both access to a household connection (a binary indicator variable) as well as the average number of hours of electricity typically available to households in a day, which is a more nuanced measure of quality (summary statistics by country are shown in table 2). We also include connections to generators and solar technologies. We include the average number of hours of availability from the full set of sources used-national grid, a generator, and household energy devices (usually solar). All measures in all countries are significantly higher in the urban samples (tables A9 and A10).

Empirical strategy
We use ordinary least squares (OLS) regression to examine the linear associations between women's empowerment and energy access at the household level (equation (1)): We run a set of four bivariate regressions for each country, j, to check if empowerment at the household level, (Eij) is significantly correlated with four distinct measures of energy access (yij): (i) ICS technology, broadly defined to include ICS and clean cookstoves, (ii) Clean cooking, defined as 1 only if clean fuels are used, (iii) Any electricity connection (national grid, generator, solar energy), and (iv) Hours of access to national grid electricity.
uij is the error term. Thus, we obtain four different β values for each country. Then, we repeat these household level analyses for the rural and urban subsamples in each country. Finally, we run single and separate urban and rural regressions with the full pooled sample, which also includes country fixed effects.
Following the bivariate analysis, we estimate a multivariate model with controls (Xij) that include (i) an indicator for urban residence, (ii) household size, and (iii) household wealth: In all regression analyses, we cluster standard errors at the smallest available unit above the household. The units correspond to 'locality' in STP, 'sampling cluster' in Rwanda, 'district' in Zambia, 'Kebele' in Ethiopia, 'village' in Cambodia, 'township' in Myanmar, and 'village development committees (VDC)/Municipality' in Nepal). Figure 3 presents the two-variable OLS regression results for the overall sample in each country and in the pooled multi-country sample of households that relate the women's empowerment to each of the four energy access variables : ICS technology, clean cooking fuels, any electricity, and electricity hours. (Table In Rwanda, the correlations are negative and statistically significant, except for clean cookstoves, where the correlation is positive but insignificant. As shown in figure 1, high E in Rwanda is mainly driven by the high ratio of working adult women in many households. Importantly, such households tend to be income poor compared to households with a lower share of working women. Thus, higher E women could be poorer either because of strong gender norms in the region that constrain women from working unless it is deemed essential for survival [22], or because most such women work in the informal sector and belong to lower socio-economic strata [23].

Results
The corresponding regression estimates for the rural and urban samples are presented in figures 4 and 5 (tables A12 and A13, respectively, in supplementary material A). As in the overall sample, we find that E is mostly positively related to energy access measures in the pooled (rural and urban) samples. Country-wise correlation patterns remain generally similar, but several lose statistical significance, likely because these samples are smaller and statistical power is, therefore, lower. Some exceptions to the patterns also merit additional discussion. First, associations between E and clean cookstoves are not statistically significant in rural STP, nor in urban Cambodia. Second, the association between E and The figure presents standardized coefficients that help us to assess the relative importance of E as a predictor across different countries and models. The 'All' category represents the pooled data estimates that include country fixed effects. Source: MTF data.

Figure 5.
Linear association between different measures of energy access and empowerment index (E) in the urban sample, based on OLS bivariate regression estimates reported in the supplementary material A, table 12. The figure presents standardized coefficients that help us to assess the relative importance of E as a predictor across different countries and models. The 'All' category represents the pooled data estimates that include country fixed effects. Source: MTF data. electricity access is positive and statistically significant only in urban Cambodia.
Lastly, the association between E and the electricity hours is less precisely estimated (i.e. not statistically significant) in Cambodia and Zambia (both rural and urban), STP (rural) and Nepal (urban).
Finally, we examine the empowerment-energy correlations using a multivariate model that includes household size and wealth as covariates. The correlation patterns in these analyses are similar to those observed in the bivariate model (see tables A14-A16 in supplementary material A). Specifically, the signs on the coefficient on E in the pooled country sample remain similar and statistically significant. However, in the country-level results, the coefficient on E loses significance in some specifications.

Discussion and conclusion
Gender inequality in low-and-middle-income countries can lead to disproportionately high burdens of energy poverty (e.g. time and health burdens associated with reliance on solid fuels and unequal access to household labor saving devices) falling on women; in turn, poor access to clean energy technologies helps perpetuate this dynamic. This is partly because women often lack agency within the household, which likely slows household adoption of clean energy technologies. While previous studies have implied that technology uptake and women's empowerment are connected, there is limited quantitative research that systematically parses these links at the household level across different technologies, countries, cultures, and rural/urban divides. Therefore, in this paper, we construct an index of women's empowerment at the household level that uses information on their education, employment status, relative position in the household, mobility, access to social groups, and credit access. We then investigate how this index is related to energy access. In general, we find a positive association between women's empowerment and energy access at the household level.
This empowerment-access pattern does not hold across all countries and contexts, however. Rwanda presents the most striking divergent exception to the general pattern, partly because even though the country scores poorly on most aspects of women's agency, many women participate in the labor force. In the case of Rwanda, labor force participation may be a particularly bad measure of agency because it is correlated with poverty. Meanwhile, the positive associations are perhaps clearest in Ethiopia and Nepal, where empowerment scores are relatively higher. In Nepal, for example, it is entirely possible that more inclusive policies have led to more rural banks, more community groups, and therefore perhaps more empowered women. However, there are many facets of empowerment, including psychological, political, and social aspects, that are omitted from our index [21]. If those aspects (e.g. political leadership, modern banking) are systematically higher or lower in certain countries, these omissions could explain some of the variation we see in energy access. It is also possible that cross-country differences are driven by mechanisms that are not directly tied to women's agency per se, such as the relative cost of energy technologies. For example, in countries with high costs of energy technologies, even high empowerment may not be enough to enable access. Unfortunately in this case, without many more countries and more variation in relative energy costs, we cannot clearly establish why some high scoring countries on empowerment do better than others. But it does suggest that activating the virtuous cycle linking gender empowerment and energy access may require achievement of at least some threshold level of empowerment [21].
Our study has other limitations as well. First, gender is not restricted to, or defined by, biological sex. According to one commonly used definition, gender is 'the socially constructed roles, behaviors, expressions, and identities of girls, women, boys, men, and gender diverse persons' [24]. We focus on women mainly because the MTF datasets contain only a binary characterization of gender.
Second, we only provide exploratory and descriptive evidence of associations between empowerment and the uptake of improved energy technologies. We do not claim causality, unlike some literature in this domain. Moreover, we are skeptical about causal claims regarding empowermentenergy because of the bidirectional link between these variables [21]. As such, it is not surprising that several empirical puzzles (including the negative association in Rwanda) emerge from the data.
To be sure, we need more studies of the exceptions to the mostly positive associations of empowerment and energy access measures for some technologies and in some sub-samples. Such studies would need to address potential reasons for this exception such as (i) low statistical power (when the sub-sample is small and variation on key measures is insufficient), (ii) confounding (by other factors correlated with both empowerment and energy access), or (iii) other forms of misspecification of the complex dynamics between these variables. To some extent, by pooling the data across countries we improve the statistical power, but pooling the data does not necessarily address the other concerns. Additionally, the country fixed effects in the pooled sample control for confounding by country-specific macro-level and infrastructure variables. Yet, these analyses fail to illuminate which of those country fixed effects are the most important for energy access, which limits how much we can say about mechanisms by comparing across countries. Simply put, we need many more countries that vary (statistically speaking) along key dimensions of these higher order mechanisms to clearly establish which ones matter (institutional or other country level factors). Further, the strength and statistical significance of the energy-empowerment relationships also diminishes somewhat when controlling for household wealth and size. While this is likely because of the correlation between these controls and women's empowerment, it suggests that other important confounders might have been omitted.
Future work should attempt to unpack these complex causal connections, perhaps by deploying an array of methods such as (1) before/after comparisons of the impacts of exogenous improvements (via randomized trials or quasi-experiments) in empowerment, or (2) panel data measuring changes in energy access [25]. Such approaches are better able to account for confounding factors, for example, households' prior experience with different technologies. Qualitative methods can not only better characterize empowerment but can also shed light on complex mechanisms of change, especially when carefully mixed with the quantitative methods discussed above.
Third, while we use a range of empowerment and access measures, future work could consider a more complete set of metrics or alternative constructs representing either empowerment or access. While the empowerment aspects of the MTF data are somewhat limited, more could be done to model fuel stacking (the use of multiple sources that are both modern and polluting) with the rich MTF data set. It is now better appreciated that most households in low-income contexts 'stack' fuels for a variety of reasons related to affordability, reliability, convenience (of alternatives), and cultural tastes and norms. Exploring how gender empowerment figures into these dynamics would be a fruitful area for future research [26,27].
Despite these limitations, to our knowledge, we are the first to examine the associations between multiple aspects of women's empowerment and multiple energy access at the household level, using observations from such a broad range of low-income African and Asian countries. Beyond a pooled analysis of a common energy-empowerment correlation regardless of setting, we also examine how these associations vary both across and within countries and by technology. As such, we present a starting point for more refined future explorations of the energygender nexus.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.

Conflict of interest
No potential conflict of interest is reported by the authors.