Short-run impact of electricity on social capital: evidence from a rural electricity program

Household access to electricity in developing countries is increasing, partly due to United Nations Sustainable Development Goal 7, namely “affordable and clean energy." According to World Bank (2022), the number of people without access to electricity declined from 1.2 billion in 2010 to 759 million in 2019. In developing countries, electricity enables households to adopt technology for both home production and leisure. Supporting this conventional wisdom, the electricity impact literature suggests that electricity improves female labor participation (Dinkelman 2011; He 2019), contributes to environmental conservation by displacing carbon dioxide emissions (Dendup 2022), reduces household (or indoor) air pollution (Barron and Torero 2017) and increases household income or expenditure (Thomas et al. 2020a). In this study, we examine the impact of electricity on noneconomic outcome social capital, particularly focusing on trust, social interactions and engagement. Whether the provision of electricity is benign or detrimental to social capital is an empirical question, and this is the approach we take in this study.

Social capital is a broad concept that encompasses various aspects of social cohesion. In general, it includes interpersonal relationships such as commitment to associations, network formation, and trust within and between networks (Dasgupta 2005; Putnam 1995). Trust is crucial in enforcing social contracts and overcoming market failures when enforcing such contracts is challenging. Additionally, trust promotes collective action, while networks facilitate the flow of information. However, social capital and interpersonal relationships are susceptible to external shocks. One such external shock in rural communities is electricity provision (and the development of similar infrastructure). In rural developing countries, electricity access enables households to adopt simple technology such as television, radio, and phone. Such technology allows households to consume more media and information. Information that is customarily used to flow through private networks within communities or households may flow through television, radio, and telephones after electrification. Households may also be exposed to media related to crime, robbery, violence, generosity behaviors, government propaganda, etc., which may affect individuals’ perceived level of trust. However, the impact of electricity on social capital will be largely determined by the nature of the information (and related media).

In this study, we examine the impact of electricity on social capital using household-level data from Bhutan. In the context of Bhutan, understanding the impact of electricity on social capital serves as important policy feedback. For approximately the last half century, Bhutan had followed a unique economic development philosophy that is commonly called gross national happiness (GNH), which places a proper balance between economic development, environmental conservation, cultural preservation and good governance. In fact, community vitality is one of the nine domains that defines the economic development philosophy of Bhutan.¹ In addition, Bhutan achieved universal electricity coverage in 2015 (Dendup 2022); thus, understanding the impact of electricity on both economic and noneconomic outcomes will help in devising complementary policies and realigning development priorities. Furthermore, Bhutan is situated in the eastern part of the Himalayas and is characterized by a unique mountainous topology with elevations ranging from 200 to 7000 ms above sea level (asl), which has led residents to adopt a peculiar way of living, such as embracing nonmaterialistic ways of life. The overall concept of the GNH economic development policy is closely related to the United Nations Sustainable Development Goals; therefore, understanding the impact of electricity on social capital is equally important and relevant in other settings.

We use the 2012 Bhutan Living Standard Survey (BLSS) by leveraging the social capital module to construct outcome variables. For identification, we exploit variation in household electrification status. However, identifying the impact of electricity is challenging, as electricity provision confounds unobservables, including those of political importance (Dendup 2022; Dinkelman 2011). In addition, once electricity arrives in a community, connecting to the grid is also an endogenous household decision. To overcome the issue of endogenous electricity provision, we estimate a bivariate probit model using a plausibly exogenous land gradient as an instrument that affects who receives electricity provision. In mountainous regions with limited plains, such as Bhutan, the cost of expanding electricity provision increases with a gradient (Duflo and Pande 2007; Dendup 2022), which determines who receives electricity provision and access to related infrastructure development programs (Dendup 2022; Dinkelman 2011). Our first-stage results show that the land gradient is correlated with electricity provision.

Our results show that in the short run, the effect of electricity on social capital is not distinguishable from zero. However, we find evidence of the heterogeneous effect of electricity on our measures of social capital. The heterogeneous results suggest that female-headed and electrified households are positively correlated with social interactions and feelings of closeness in the community than the male-headed and unelectrified households. Similarly, a greater share of female and electrified households is positively correlated with social interactions. The heterogeneous results of trust also show that electrified and female-headed households are positively correlated with the trust that others will help during an emergency and in terms of borrowing and lending money. However, the results largely depend on the validity of the land gradient instrument. In observational studies such as ours, it is difficult to exclude the possibility of violation of instrument exogeneity. Hence, we also conduct sensitivity tests on our results, following Conley et al. (2012). The sensitivity test results show that our results remain consistent even under mild deviation from instrument exogeneity. In addition, we use a matching method as the alternative identification strategy to estimate the effect of electricity on social capital; the results are comparable with the bivariate probit results.

We also empirically explore the underlying mechanism to understand why electricity provision has no effect on social capital. The effect of electricity provision on social capital operates through the adoption of technology or the use of basic electrical appliances. We examine whether electricity provision induces households to adopt appliances; our results show that the effect of electricity on the adoption of television, radio, and phones is not distinguishable from zero in the short term. In Bhutan, the majority of rural electrification projects were implemented from 2008 to 2013. During the data collection periods of the 2012 BLSS, it is likely that households would have only recently obtained an electricity connection. On the other hand, the appliance adoption results show that the adoption of basic cooking appliances is positively correlated with electricity provision. However, our data show that the adoption of electrical cooking appliances does not completely replace the use of traditional cooking fuel firewood² Therefore, we interpret our results as indicating a short-term effect of electricity on social capital. Our result is consistent with findings from field experiments conducted in Kenya by Lee et al. (2020), who reported no effect of electricity on numerous outcomes in the short term.

This study contributes to both the electricity and social capital literature within the field of economics. In the electricity literature, our study builds upon previous research, particularly that of Dinkelman (2011), which examines the impact of electricity on female labor outcomes in South Africa using a land gradient as an instrument; the study reports a positive effect on female labor outcomes. Our study is also related to the existing electricity literature that examines the effect of electricity on health (Barron and Torero 2017), income, expenditure (Thomas et al. 2020b), firm output (Rud 2012), education (Lipscomb et al. 2013), and carbon emissions and firewood substitution (Dendup 2022). Similarly, a field experiment by Lee et al. (2020) reported significant effects on various energy-related outcomes in Kenya, including the use of electricity for lighting and expenditures on kerosene; however, the authors observed no significant impact on employment, health, or test scores. The current study enhances the extensive literature on the impact of electricity by offering evidence of the impact of electricity on social capital, which is a dimension that has thus far remained unclear in the existing electricity literature.

In the social capital literature, two main branches of literature have emerged. One strand delves into the effects of social capital on broadly defined indicators of welfare, for instance, the effect of social capital on health (Cao et al. 2022), crime (Buonanno et al. 2009; Stuart and Taylor 2021), human capital formation (Lee and Bell 2013; Agénor and Dinh 2015), income, gross domestic product (Knack and Keefer 1997; Tenzin et al. 2015), sustainable agriculture business (Lang et al. 2023), trade (Korovkin and Makarin 2023) and firm output (Miguel et al. 2005). Conversely, the other strand of social capital literature focuses on the impact of policies or events on social capital. Our study aligns with this latter category and is closely linked with the seminal work by Olken (2009), which examines the impact of radio and television on social capital in Indonesia, and Nunn and Wantchekon (2011), which investigates the impact of the slave trade on trust. In Olken (2009), trust (as an indicator of social capital) is measured in terms of general trust and trust in neighbors, people in the village, the government, the president, the village head, and the village parliament. Similarly, in Nunn and Wantchekon (2011), trust is measured in terms of trust in relatives, neighbors, local councils, intragroup, and intergroups. Our study contributes to these seminal works by introducing two additional measures of trust, namely trust in terms of (1) lending money and (2) taking care of children.

Similarly, in Olken (2009), social engagement is measured in terms of group membership and participation. Our study further contributes to the exploration of social engagement by providing new evidence of the influence of electricity on social capital on broad measures, including having friends, willingness to help in the community, a sense of togetherness, and general interactions. In addition, numerous studies have examined the effects of media technologies, such as Facebook on social interaction (Allcott et al. 2020), television on civic engagement (Durante et al. 2019), and mobile phones on social mobilization or coordination (Manacorda and Tesei 2020). In our study, instead of focusing on a single technology, we assess the impact of a key technology enabler, namely electricity. Given that electricity enables households to adopt various media and related technologies, our results contribute to understanding the overall effect of technology on our measures of social capital.

The paper is arranged as follows. Section 2 discusses social capital and related literature, Sect. 3 describes the data and background, and Sect. 4 describes the econometric model. The results are reported in Sect. 5. Section 6 describes mechanism results and Sect. 7 concludes the paper.

Social capital is a multidisciplinary concept defined in numerous ways. Many empirical studies have utilized the concept of social capital, which includes the basic features of institutions and organizations, such as trust. Some studies have associated social capital with social networks and social engagements. Additionally, social capital is understood as behavioral norms of reciprocity that may promote collective action. In general, social capital includes various features of social organizations and behavioral norms. For instance, according to Dasgupta (2005), social capital can be understood as an interpersonal social network that includes the components of social networks, trust, and reciprocity. On the other hand, according to Putnam (1995), social capital may be more concisely understood in terms of the features of institutions, i.e., trust and social networks. Therefore, in this study, we use broad definitions of social capital, including indicators of trust and network and social engagement. Trust is helpful in making social contracts more credible, facilitating economic transactions, and minimizing monitoring costs, both in public and private domains. Similarly, social networks serve as a medium for information that may be useful in addressing market failures (when the cost of information is high). External factors can affect social capital, particularly those that interfere with local norms and social networks. For instance, in Africa, the descendants of those who experienced the slave trade exhibit lower levels of trust (Nunn and Wantchekon 2011). Rural infrastructure development programs, such as electricity, can also affect social capital through technology adoption.

Numerous empirical studies have explored how information technology, or the diffusion of information, influences social capital. According to the framework proposed by Barbera and Jackson (2020), which is relevant in terms of explaining incentives or motivation for prosocial behavior, information and trust play vital roles in addressing coordination and collective action problems. The authors posit that the success of (collective) action relies on the proportion of individuals participating in the undertaking, which is influenced by anticipated rewards, associated costs, and beliefs about others’ engagement. The propensity of others to make an engagement decision is private, but information within networks helps to gauge others’ propensity to engage (in action). Private information both within and between networks and subsequent trust help individuals decide whether to contribute to activities that benefit society. The relation between social capital and electricity is evident once we stipulate that electricity is a technology enabler (in rural developing countries) and that technology plays a pivotal role in information diffusion, as well as in building trust in terms of others propensity to contribute to undertakings.

It is evident from the above-mentioned simple conceptual framework that technology adoption may be either benign or destructive, depending upon how such technology interacts with local norms, culture, interference with local communication systems and types of information. In theory, the adoption of media appliances such as television and radio could result in a tradeoff between social participation and time spent using television and radio. Past studies that have examined the effect of media technology have reported mixed results. A study by Olken (2009) in Indonesia reported that television and radio network expansion resulted in households making a trade-off between the time spent using television and the time spent using the radio with social participation. As a result, the same study reported that increased time spent using television and radio resulted in a reduction in social participation and reduced the level of self-reported trust. Similarly, Durante et al. (2019) reported that exposure to commercial television in Italy at a younger age is associated with a lower level of social participation. In another study, using data from the European Value Study and World Values Surveys from 36 OECD countries, (Campante et al. 2022) reported an insignificant correlation between television penetration rates and aggregate measures of social capital. However, the same study reported evidence of different effects on the different measures of social capital. While the authors reported no significant effect of television penetration rates on total numbers of members, the membership status of parties/unions or self-reported interest in politics, they did report a significant and negative correlation between political party activity and trust. The authors associated these results with the crowding out effect of entertainment content on political information. On the other hand, Bruni and Stanca (2008) reported a positive relationship between television viewing and time spent with family but a negative relationship with time spent with friends, colleagues and churches.

DellaVigna and Kaplan (2007) reported that voter participation increases due to television news, while Gentzkow (2006) reported that voter turnout decreased during the 1950-1990 period after the introduction of television in the USA, when television content was mostly dedicated to entertainment. Similarly, Strömberg (2004) reported that a greater percentage of radio listeners is associated with a greater percentage of voter turnout. Similarly, the effect of exposure to reconciliation messages via radio in Rwanda depicts a higher level of inter ethnicity trust (Blouin and Mukand 2019). The difference in effects across space, time and technology seems to be explained by the type of information disseminated. For instance, a study by Dendup and Arimura (2019) showed that exposure to health and conservation information via television is positively correlated with the use of electricity and gas as cooking fuel in Bhutan. Similarly, entertainment television has reduced the voter turnout in Sweden (Ellingsen and Hernæs 2018), suggesting that the effect of information technology depends on the type of information viewers are exposed to.

Furthermore, evidence suggests that mobile phone adoption helps in facilitating coordination. For instance, Manacorda and Tesei (2020) reported that text and voice messages increased the mobilization of anti-government protests in Africa between 1998 and 2012. Similarly, in Russia, the diffusion of independent television channels also has a significant negative effect on voter turnout, suggesting the successful coordination of boycotting an election (Enikolopov et al. 2011). In addition, many studies have focused on examining the effect of the internet on social capital and health. A field experiment conducted by Allcott et al. (2020) showed that those in the treatment group who deactivated Facebook for a month spent more time with family members than the control group. Similarly, Braghieri et al. (2022) reported a negative effect of staggered Facebook rollout across universities in the U.S. on mental health. The effect of the increase in the use of high-speed internet on mental health has a negative effect on younger cohorts, and the effect is even stronger for females (Golin 2022). The evidence of the effect of the internet on interactions and health suggests that the effect of electricity on social capital may also operate via its impact on mental health and social interactions.

Electricity provision also enables households to adopt simple (electrical) cooking appliances such as electric cookers and water boilers. The replacement of traditional cooking fuel through the adoption of cooking appliances reduces households’ dependence on firewood (which is one of the primary cooking fuels used in developing countries). Previous studies in Bhutan have reported that rural households tend to adopt such appliances (Dendup 2022). In developing countries such as Bhutan, firewood is traditionally collected in groups involving family members and friends from the neighborhood³ The replacement of traditional cooking fuel with electrical appliances is likely to reduce the number of traditional social interactions experienced during the process of firewood collection. Thus, the effect of electricity on social capital through the adoption of cooking appliances and the subsequent replacement of firewood collection can be determined by how much firewood is being replaced by cooking appliances and how the time saved from engaging in firewood collection affects other forms of social engagement.

Electricity infrastructure developments are seldom randomized, and unobservables confound the effect of electricity provision. In particular, there are two different levels of selection issues in electricity infrastructure provision. First, at the community or village level, whether the community receives electricity provision is determined by political importance (Dinkelman 2011; Dendup 2022). Second, once the community is electrified, whether households choose to connect to the grid is a household or individual-level decision (Dendup 2022; Litzow et al. 2019). However, selection at the household (or individual) level is of less concern in our case because our data show that only approximately 4% did not connect to electricity once a community was electrified.

To overcome endogenous electricity provisions due to unobserved omitted variables such as political importance, we estimate a bivariate probit model using the land gradient (at the subdistrict level) as an instrument. We estimate the following econometric model:where electricity\(_{ij}\) is a binary variable indicating whether household i in subdistrict j is connected to electricity, SC\(_{ij}\) is one of the measures of social capital of household i in subdistrict j, and gradient\(_j\) is an average land gradient of subdistrict j. In the electricity equation, \(\gamma \) captures the correlation between the land gradient and electricity provision. In the social capital equation, the coefficient of interest \(\beta \) captures the effect of electricity on (one of the measures of) social capital (SC\(_{ij}\)), and \(\delta _d\) is the district fixed effect. The vector X includes other controls, and \(\eta \) and \(\epsilon \) represent error terms of the electricity and social capital equations, respectively.

Following Nunn and Wantchekon (2011) and Olken (2009), the vector of control includes individual, household and community-level information measured at the subdistrict level. Individual-level control variables include age, age squared, gender, level of education of the head of household and whether the head of the household can read and write. In Bhutan, there is a monastic education system; thus, even if a head of household reported zero years of schooling, he or she may be able to read and write. To capture this effect, we also control for whether household heads can read or write. Household covariates include distance to the nearest market, distance to the nearest road, household size, share of female household members and per capita monthly household expenditure. We also include district fixed effects in our model to control for differences at the district level. In particular, district fixed effects control for differences in local culture and social norms specific to particular regions that may also affect social capital.

In our bivariate probit model, the vectors of the controls are very similar to those of Nunn and Wantchekon (2011) and Olken (2009). However, in our model, we do not control for living condition fixed effects to capture the effect of income or wealth, as in Nunn and Wantchekon (2011). Instead, in our model, we directly control for monthly capita household expenditures as a proxy for income. Similarly, Olken (2009) controlled for whether the subdistrict faces north, east or south. However, Bhutan has irregular terrain, with peaks, valleys, and slopes that obscure the sun’s position, making it difficult to determine direction based on landmarks or natural cues. Hence, we do not control for such fixed effects; however, district fixed effects capture such differences. In addition, both Nunn and Wantchekon (2011) and Olken (2009) controlled for ethnicity fragmentation and religion fixed effects. In the 2012 BLSS, such important information was not collected. We assess the impact of such important variables on parameter estimates in our study; we will return to this discussion in Sect. 5.1.

The results of bivariate probit models depend on the validity of the instrument. It is assumed that the instrument is sufficiently correlated with the endogenous variables (which can be verified from the electricity equation results). In addition, we assume that the instrument is exogenous and that the effect of the instrument on the outcome only operates through the endogenous variable, electricity provision. It is difficult to exclude the possibility of the violation of this assumption in observational studies such as this. At the same time, similar to other observational studies, we encounter the challenge of finding purely exogenous instruments. We conduct sensitivity tests for the main results under the mild deviation of the exogeneity assumption of the instrument, which we discuss in detail under Sect. 5.2.

The average partial effect of electricity on trust is reported in Table 2, social interaction and network density in Table 3, and social engagement in Table 4. All the results reported in Table 2 through Table 4 are estimated using a land gradient as an instrument; however, the results of the electricity equation are not reported for brevity purposes, except for the binary outcome variable indicating whether an individual trusts others in terms of lending and borrowing money, which is shown in Column 1 of Table 2. All the coefficients of the bivariate probit models, along with the coefficients of the electricity equation, are reported in the Appendix A in Table A1 through Table A3. The coefficient of the land gradient is negative and significant at the 1% significance level, indicating that as the land gradient increases, communities are less likely to receive electricity provision due to the higher cost of building electricity infrastructure, as discussed in Sect. 3. In addition, the results also show that the land gradient is sufficiently correlated with the endogenous variable of electricity provision.

The average partial effects of different measures of trust are reported in Columns 2 through 5 (of Table 2). In Columns 2 through 5 (of Table 2), the outcome variables are binary variables indicating whether respondents trust that others are willing to lend them money, whether households trust that their neighbors could help take of their children, whether households trust their neighbors and whether respondents trust that their neighbors will not exploit them. The effects of electricity on all the measures of trust are not distinguishable from zero except for the indicators of whether households trust their neighbors; this coefficient is negative and marginally significant at the 10% significance level.

In Columns 1 through 6 of Table 3, the average partial effects of different measures of social interactions and network density are reported. In Column 1, the outcome variable is the number of close friends.¹¹ In Column 2, the outcome variable is a binary variable indicating whether neighbors are always willing to help. In Column 3, the outcome variable is a binary variable indicating whether the community has a strong feeling of togetherness, and in Columns 4 through 6, the outcome variables are binary variables indicating whether in the month prior to the survey, the respondent (usually the head of the household) had met others in public places for leisure, whether others had visited the household and whether the respondent had visited others’ home. The effects of electricity on the indicators of social interaction and network density are not distinguishable from zero except for whether people had met in public places for leisure. The coefficient of the outcome variable of whether they had met in public places for leisure is positive and marginally significant at the 10% significance level, suggesting a positive correlation between electricity and social interactions. Similar to other South Asian countries, Bhutan is also a patriarchal society. The majority of household chores, including cooking, are performed by women. Traditional cooking with firewood is time-consuming, and women spend a substantial amount of time in the kitchen. Therefore, one possible explanation for the positive relationship between electricity and the time spent meeting others in public places for leisure could be attributed to the time saved from the adoption of electrical appliances, as shown in Sect. 6. We report the results of social engagement in Table 4. The level of social engagement in our study is captured by the willingness to contribute time and money for community activities and whether the household reported engaging in work for the community in the 12 months preceding the survey. The coefficient of electricity is not distinguishable from zero for all three measures of social engagement.¹²

Overall, these results suggest that the effect of electricity on social capital is not distinguishable from zero. This result is consistent with the field experiment results from Kenya by Lee et al. (2020), who showed that the effect of electricity on almost all household-level outcome variables is insignificant in the short run except for lighting and basic appliance adoption. The interpretation of our results requires a great deal of caution. As discussed in Sect. 3, the majority of rural electrification in Bhutan was implemented in 2008; thus, we suspect that when the 2012 BLSS was implemented, most households may have only recently obtained their electricity connection. Therefore, we interpret our results as the effect of electricity on social capital in the short run. However, one of the weaknesses of the BLSS data is that our data do not allow us to investigate the effect based on the number of months or years that households have been electrified. In addition, the instrumental variable results are local average treatment effect (LATE) (Angrist and Imbens 1995); thus, our results represent the effect of electricity on households affected by the instrumental variable and do not represent an average effect on the population.

Furthermore, we estimate the effect of electricity on social capital using the matching method. The details of the matching estimation procedure are discussed in Appendix A, and the results are reported in the Appendix in Fig. 1. The matching results are consistent with the bivariate probit results both in terms of the level of significance and the sign of the coefficients (except for two outcome variables).¹³

The bivariate probit results of no effect of electricity on social capital are consistent with the evidence from Kenya by Lee et al. (2020), who reported that electricity has no effect on the number of outcome variables in the short run. In this section, we examine why we do not observe an impact of electricity on social capital. Electricity provision in developing countries enables households to adopt and use technology for household production, media consumption and entertainment. As discussed in Sect. 2, the adoption and use of simple technology, such as television, radio, mobile and cooking appliances, is likely to affect our outcome variable social capital by interfering with local social norms and cultures. For instance, Jensen and Oster (2009) reported that media consumption in India has reduced the acceptance level of (self-reported) domestic violence against women. Similarly, the expansion of television and radio networks and the subsequent amount of time spent on television and radio use has reduced household participation in community activities in Indonesia (Olken 2009). Therefore, for electricity to affect our outcome variable social capital, households must adopt basic electrical appliances such as radio, television, phone, etc., that will enable them to consume media and information.

Similarly, in rural Bhutan, the majority of households traditionally use firewood for cooking; thus, the adoption of simple cooking appliances such as rice cookers, curry cookers and water boilers is likely to reduce the household dependence on firewood. Firewood collection is usually conducted in a group; thus, the adoption of such cooking appliances is likely to affect social interactions in rural Bhutan.¹⁶ In the BLSS data, information about the adoption of electrical appliances was collected. Using this appliance adoption information, in this section, we empirically examine whether electrified households adopted electrical appliances that would affect social capital.

To examine this mechanism, we estimate the following appliance ownership model:where we use the land gradient as an instrument for endogenous electricity provision. We also control for the same set of controls as that used in Eqs. 1A and 1B. We estimate six different appliance adoptions, including the adoption of television, radio, phone and three different basic cooking appliances. The effect of information and media consumption on social capital is largely determined by the types of television programs and information that households are exposed to. Depending on the type of information that the household is exposed to through appliance adoption, such as television, radio and phone, the levels of trust, social interaction and engagements are likely to be affected positively or negatively. Similarly, the effect of cooking appliance ownership on social capital will be largely determined by the amount of firewood displaced by the adoption of cooking appliances.

The results are reported in Table 10. The effects of electricity on the adoption of television, radio and phones are not distinguishable from zero. This result may suggest that one of the possible explanations for the lack of an effect of electricity on social capital is that the adoption of media and information-consuming appliances are not distinguishable from zero. Nevertheless, the results of the adoption of the three basic cooking appliances are positive and significant. However, our data show that the adoption of cooking appliances does not completely replace firewood consumption. For instance, 77% of the households that have adopted rice cookers also reported that they still use firewood, 75% of those that reported using curry cookers also reported still using firewood, and 72% of those who reported using water boilers also reported still using firewood. We also examine the heterogeneous effect of appliance adoption by female-headed households, the share of female household members, household size, household expenditure and level of education. The results are reported in Table 11. The results suggest that female-headed households are positively correlated with the adoption of television, radio and cooking appliances.

There are many situations where the social contract remains incomplete. In such situations, social capital can help in enforcing and establishing credible social contracts. However, social capital can be affected by external shocks such as electricity provision through appliance adoption and increased media consumption. In this study, we examine the effect of electricity provision on social capital, particularly self-reported levels of trust, social interactions and social engagements. We use nationally representative 2012 BLSS data. To identify the effect of electricity on social capital, we exploit the difference in the electrification status of rural households.

However, electricity provision and the development of the related infrastructure are often confounded by unobservable factors such as political importance and self-selection in obtaining connections. In this study, we address endogenous electricity provision using a plausibly exogenous land gradient as an instrument. The strong correlation between the land gradient and electricity provision suggests that electricity provision is correlated with the land gradient. However, in observational studies such as ours, it is challenging to find a purely exogenous instrument, and it is not possible to exclude the possible violation of instrument exogeneity. Therefore, we also conduct sensitivity tests of our results following Conley et al. (2012). Our results show that under a mild deviation from the exclusion restriction assumption, our results still lie within the lower and upper bounds of the sensitivity results of Conley et al. (2012). Furthermore, following the literature, we control for carefully chosen control variables. We also assess the effect of unobservables using selection on observables and it is less unlikely that the effect of electricity on social capital is driven by unobservables.

The bivariate probit results suggest that the effect of electricity on social capital is not distinguishable from zero for numerous measures of trust and social interaction and social engagements. However, we interpret our results as the short-term effect of electricity on social capital considering the implementation of the rural electrification program in Bhutan. In Bhutan, major works on rural electrification started in 2008; thus, it is likely that electrified households may have obtained their electricity connection immediately prior to the 2012 BLSS. As a result, households may not have had sufficient time to adjust their household budget to adopt appliances and related technology that would interfere with local norms and, subsequently, with our measures of social capital. Therefore, we can carefully conclude that in the short run, we do not find evidence of an effect of electricity on social capital. We also examine the effect of electricity on social capital using the matching method and find that the effect from nearest neighbor matching is consistent with the bivariate probit results.

We also examine the heterogeneous effect of electricity on social capital by female-headed households, share of female household members, household size, household expenditure (or income) and level of education of the head of household. Our results suggest that the impact of electricity on social capital is heterogeneous. Compared with unelectrified male-headed households, electrified and female-headed households are positively correlated with self-reported closeness in community and social interactions. Our results also show a heterogeneous effect of expenditure (or household income); however, we do not observe a heterogeneous effect by the level of education of the head of household.

We further explore the underlying mechanism that affects the effect of electricity on social capital. Electricity provision in developing countries enables households to adopt technology. For instance, electricity enables households to adopt information- and media-consuming appliances such as television, radios and phones. Similarly, providing electricity also enables households to adopt basic electrical cooking appliances such as rice cookers, curry cookers and water boilers. The adoption of such cooking appliances is likely to affect households’ dependence on firewood and subsequently impact their social interactions. Our results suggest that in the short run, media adoption and information-consuming appliances are not distinguishable from zero. On the other hand, our results show that the adoption of cooking electrical appliances is positively correlated with electricity provision. However, our data also show that electrical cooking appliances do not completely replace firewood consumption.

Full size image

Overall, our results show that in the short run, the effects of electricity on social capital, particularly self-reported trust, social interaction and social engagement, are not distinguishable from zero. Our results also show that in the short run, the effects of electricity on the adoption of media and information-consuming appliances are also not distinguishable from zero. On the other hand, we observe that households adopt basic electric cooking appliances; however, our data show that such appliances do not completely replace firewood. The results also show that the effect of electricity is heterogeneous. One of the limitations of our data is that our measures of social capital are self-reported. Thus, it is likely that our outcome variable may have also been affected by social desirability bias and other inherent issues that are typically related to self-reported outcome variables. In addition, Bhutan is characterized by different topographies and unique village-level characteristics. However, as the BLSS data do not provide village identifiers, we could not control for such differences in our study.