Rural electrification in India and RGGVY

When India gained independence in 1947, electrification was limited to a small number of urban areas (Kale Reference Kale2014). Over time, the need to electrify pumpsets to extract groundwater for the high-yield plant varieties introduced under the agricultural green revolution enabled rural electrification (Smith and Urpelainen Reference Smith and Urpelainen2016). And yet, according to the 2001 census of India, only 44% of rural households used electricity as their primary source of lighting.

In India, the governance of the electricity sector is a “concurrent” subject, with both the central and state governments having policy authority. While electricity distribution companies are controlled by state governments, except under privatisation in Odisha and the capital city of Delhi, the central government is also authorised to initiate schemes to promote rural electrification. This multilevel governance structure is essential to understanding how the RGGVY is implemented.

Launched in April 2005 by Dr. Manmohan Singh’s (Indian National Congress) central government, the goal of the RGGVY was to increase the rate of rural electrification across India. At the initiation of the RGGVY, the government announced that the “scheme has been launched to fulfill the commitment of the National Common Minimum Programme (NCMP) of completing the household electrification in next 5 years and modernising the rural electricity infrastructure.” ^{Footnote 2}

In the RGGVY, the central government provides a 90% capital subsidy for the electrification of villages; the remainder is provided by the national Rural Electrification Corporation as a soft loan. While the RGGVY is a village electrification scheme, households living below the official poverty line are guaranteed free electric connections. State governments can apply for the funds, provided they agree to implement the scheme according to the guidelines set by the central government. Until 2007, only habitations (subvillage units) with more than 300 people according to the 2001 Census were eligible; at that time, the population floor was decreased to 100 (Burlig and Preonas Reference Burlig and Preonas2016). While all villages qualify for the electrification subsidy regardless of socioeconomic status, RGGVY also gave households below the national poverty line a free household connection.

The actual selection of villages and habitations for RGGVY implementation is the responsibility of district-level administrative and technical officials, such as the Chair of District Council (Zilla Parishad), District Magistrate (or Collector) and Electricity Engineers. They work closely with their respective MLAs and Block Development Officers. In India, district-level officials enjoy a lot of discretion in implementing poverty alleviation programs introduced by the central government in Delhi. Focusing on the Employment Assurance Scheme (EAS), Corbridge et al. (Reference Corbridge, Williams, Srivastava and Véron2005) find that district-level officials often revise the details of implementation and collaborate with MLAs to screen eligible recipients for political reasons.

As a result, village selection under RGGVY similarly provides politicians ample opportunity to target the policy according to their own preferences. ^{Footnote 3} The formal rules specifically encourage RGGVY implementation to consider the special needs of the Dalits: “[b]asic infrastructure such as distribution transformer and distribution lines is provided in the inhabited locality as well as the Dalit basti/hamlet where it exists” (Ministry of Power 2013, 3).

The RGGVY has achieved its goal of rapid rural electrification (Aklin et al. Reference Aklin, Cheng, Urpelainen, Ganesan and Jain2016; Burlig and Preonas Reference Burlig and Preonas2016). By the end of the financial year 2012, the program had reached a total of 104,496 unelectrified villages and another 248,553 previously electrified villages for intensification – that is, more than one-half of all villages in India. Within villages, 19.5 million households living below the official poverty line were electrified. As Burlig and Preonas (Reference Burlig and Preonas2016) show, the RGGVY has not had large economic effects in targeted villages, but it has significantly increased electricity access.

In UP, a May 2012 assessment of program implementation showed that 27,770 nonelectrified villages were electrified and 2,982 electrified villages saw improvements through intensive electrification by the end of 2011 across 63 districts. Compared to a goal of 32,118 villages set by the state government, the achievement rate was thus 96%. At the same time, the number is very low relative to the total number of villages in the state: according to the 2011 Census of India, UP had 106,773 villages. Based on our analysis, 96,557 of these villages were inhabited (Indian census also includes abandoned villages in the total number). Thus, only about 31% of all UP villages saw RGGVY implementation by the time of the assessment even though almost all villages in the state perform poorly with regard to electrification: the average household electrification rate in a typical UP village in the census was only 23%. ^{Footnote 4}

Caste inequality in India

Scholars have documented persistent patterns of social bias in India against both religious minorities, such as Muslims and Christians (Hasan Reference Hasan2009), and against Dalits (SC) (Bayly Reference Bayly2001; Louis Reference Louis2003; Kapur et al. Reference Kapur, Prasad, Pritchett and Babu2010; Thorat and Newman Reference Thorat and Neuman2012). In the case of Dalit discrimination, some members of the upper caste groups consider their low-caste counterparts socially inferior. As Thorat and Lee (Reference Thorat and Lee2005, 4198) note, for example, shopkeepers belonging to upper castes may refuse to trade with Dalit customers “until they have hung cloth screens in a place to ‘protect’ themselves from the polluting presence of the ‘untouchables’.” Experimental audit studies from Indian labour markets show that Dalits often face obstacles in finding employment (Banerjee et al. Reference Banerjee, Bertrand, Datta and Mullainathan2009). These results highlight the fact that caste bias remains common in India, as higher castes often hold negative attitudes towards the lower castes and refuse to engage in social or economic exchange with them. Tolerance of discrimination remains widespread across large segments of the population (e.g. Shah et al. Reference Shah, Mander, Thorat, Deshpande and Baviskar2006; Kapur et al. Reference Kapur, Prasad, Pritchett and Babu2010; Ramaiah Reference Ramaiah2011).

Caste inequality can take at least three forms (Shah et al. Reference Shah, Mander, Thorat, Deshpande and Baviskar2006). First, physical exclusion, the most common form of untouchability, involves members of upper castes who refuse and avoid contact with those from the SC through residential segregation and denials of commercial exchanges and services, food sharing and water distribution (e.g. Bros and Couttenier Reference Bros and Couttenier2010) and access to public facilities. Next, untouchability also involves public humiliation, as dominant castes restrict the Dalits from using vehicles, smoking, or even holding their wedding processions. Finally, upper castes have also economically exploited the Dalits by blocking the Dalits from certain employment opportunities (e.g. Siddique Reference Siddique2011) and underpaying them. Here we examine these patterns of caste inequality in the case of the RGGVY, a large-scale public policy financed by the central government and implemented by the state governments.

Despite the government’s various legal and policy efforts to eradicate caste bias, untouchability remains widespread in India, especially in the rural areas. Even worse, recent years have seen the emergence of collective, organised violence targeting the SC who seek to assert their rights (Chauchard Reference Chauchard2017). In several cases, “these atrocities took place” because local elites aim to maintain their dominance and privilege in the face of increasing economic competition from the Dalits (e.g. Sharma Reference Sharma2015).

Although caste inequality in the society and the private sector has been carefully documented, including with experimental methods that can identify causal effects, it remains unclear whether a bias remains in public policy itself. Thorat and Lee (Reference Thorat and Lee2005, 4198) use survey data from 531 villages in five states to show that “patterns of exclusion and caste discrimination … afflict, if not overwhelm, the government India’s mid-day meal scheme.” However, it remains unclear whether such patterns reflect policy implementation by the government or, alternatively, the social behaviour of the people who use or prepare mid-day meals in Indian schools.

Solutions to the problem of caste inequality have proven elusive. In India, the most important national policies against caste bias are educational, employment and representation “reservations” (Jaffrelot Reference Jaffrelot2006; Bhavnani Reference Bhavnani2017), which guarantee both the SC and scheduled tribes (adivasis) a certain percentage of admissions to public institutions of higher learning, government jobs and seats in different political bodies from village councils (gram panchayats) to state legislative assemblies and the national parliament. However, it appears that these reservations have not effected a social revolution: the evidence on the socio-economic effects of reservations suggests that they are weak (Jensenius Reference Jensenius2015), though they have contributed to policy changes and political mobilisation (Pande Reference Pande2003; Jaffrelot Reference Jaffrelot2006).

Caste inequality in rural electrification

Rural electrification can contribute to caste inequality in UP through several channels. Because the RGGVY program design allows state governments, district-level officials and MLAs to decide on the targeting of RGGVY implementation at the village level, the national program has a lot of scope for inequality at the implementation stage despite formal rules that base implementation on need. Therefore, examining RGGVY offers an opportunity to detect bias in public policy implementation. In turn, bias in implementation feeds inequality when it is done in way that further weakens groups (such as SCs) that are already marginalised.

The goal of the RGGVY was to electrify as many villages as possible, and villages themselves did not have to do anything to request RGGVY implementation, and in fact lacked lobbying power to make demands (Prayas Energy Group 2011, 27). As per the 2006 National Rural Electrification Policy of India (Ministry of Power 2006), village heads (pradhans) and councils (gram panchayats) only had an advisory and supervisory role in RGGVY implementation. Field research conducted by Greenpeace, an environmental nongovernmental group, finds that fewer than 10% of villagers in Madhubani district, Bihar, were aware of RGGVY even though their village was electrified under the scheme (Greenpeace, n.d.). Indeed, a large-scale survey of six North Indian states reveals that fewer than 50% of village heads had heard of RGGVY in 2014, after a decade of implementation (Aklin et al. Reference Aklin, Cheng, Ganesan, Jain and Urpelainen2016). This basic lack of awareness suggests limited interest or ability in participating in the policy formulation process, which is not surprising given the limited authority of local government in this area.

This feature of the RGGVY is useful for us, as it ensures we cannot empirically conflate bias against lower castes with less vocal demands or less effective collective action by the SC. Given that the RGGVY is a top-down program that does not condition eligibility on active demands by villages, any bias against SC can be attributed to bias by government officials. Moreover, this feature of the RGGVY also ensures that a bias against SC in RGGVY implementation cannot be attributed to logrolling: given the RGGVY rules, communities cannot expect to gain access to other public services by forgoing electrification. The RGGVY is not part of a menu of options for villages, but rather a standalone program.

The first mechanism of bias in policy implementation is the choice of villages. According to RGGVY rules, states are allowed to choose the targeting of districts. With the average district in UP having about three million people, however, this administrative unit is too large for biased implementation: every district in UP has large numbers of both Dalit and non-Dalit communities. There is little scope for inequality in the selection of districts.

Within districts, however, local officials play a key role. The RGGVY rules state that the selection of villages remains in the hands of “the respective States/DISCOMs [electricity distribution companies] based on field survey while preparing Detailed Project Reports” (Ministry of Power 2013, 7). Because the distribution companies in UP are state-owned, they create ample opportunities for political and bureaucratic control (e.g. Wade Reference Wade1985; Iyer and Mani Reference Iyer and Mani2012). MLAs, who compete for office in constituencies that are smaller than districts, have both formal authority over the selection of local officials, such as the powerful Block Development Officers, and informal power through their popular support and networks of connections. A comprehensive assessment of the RGGVY itself in UP notes that (i) the local district-level planning leaves a lot to be desired and that (ii) state-level officials lack the capacity to properly assess the resulting plans (IRADe 2013, 54–55), suggesting ample scope for biased implementation by local politicians. We argue, therefore, that RGGVY’s inequality potential is found at the district and electoral constituency level. Due to their political power, connections and ability to shape the careers of bureaucrats, MLAs can influence village selection for the RGGVY within their own electoral constituency.

Indeed, Section 6 of the Rural Electrification Policy of 2006 further mandates the formation of a district committee for rural electrification, with inter alia elected representatives from the district as members (Ministry of Power 2006). Specifically, the committees are chaired by the district panchayat’s chair, the district planning committee’s chair or the district collector. Other members include district-level government agencies, consumer associations and “other important stakeholders.” This setup gives the committee chair, who must be an elected official or a high-level bureaucrat, ample scope to select his or her preferred members, as there are no specific rules to ensure broad representation. Moreover, since elections are in India hold only every five years, the elected chairpersons have a long period of time to direct rural electrification to their preferred direction.

The role of the committees is to “to coordinate and review the extension of electrification in each district, to review the quality of power supply and consumer satisfaction and to promote energy efficiency and its conservation” (Ministry of Power 2009, 35). By participating in this committee and influencing its work, the MLA has potential for shaping the allocation of electrification works across villages. As politicians and bureaucrats work together to select villages for RGGVY implementation, they can use census lists and other data sources to identify villages for preferential implementation at the expense of others. By using this information, the local administration can thus channel RGGVY resources to the electrification of non-Dalit communities – if it so prefers. ^{Footnote 5}

The logic of local inequality in village selection is also facilitated by the role of Members of State Assembly (MLAs) as “fixers” in their electoral constituencies (Chopra Reference Chopra1996; Jensenius Reference Jensenius2015). In India, the typical MLA actually spends very little time in the state capital in legislative debates and instead mostly focuses on serving the people in his or her constituency. Thus, the MLAs have a strong local presence and can influence village selection. Most MLAs spend their time in their own constituencies and thus have easy access to the district-level officials who are selecting villages for electrification.

Political mobilisation against caste inequality

Frustrated with the low pace of change, the Dalit population has mobilised politically against caste inequality (Mendelsohn and Vicziany Reference Mendelsohn and Vicziany1998; Duncan Reference Duncan1999; Jaffrelot Reference Jaffrelot2003; Jaoul Reference Jaoul2006; Jeffrey et al. Reference Jeffrey, Jeffery and Jeffery2008; Singh Reference Singh2017). Political parties are central players in Indian politics. They deliver goods for both the general public and particular interests (Chhibber and Nooruddin Reference Chhibber and Nooruddin2004; Dunning and Nilekani Reference Dunning and Nilekani2013; Kruks-Wisner Reference Kruks-Wisner2018). And they tend to respond to their core supporters’ demands, which makes representation crucial to obtain favourable public policies. This is particularly true for parties that are closely associated to a particular ethnic or social group (Chandra Reference Chandra2007). Inequality-reducing policies, in turn, necessitate strong electoral mobilisation.

Political activists such as Kanshiram and Mayawati have created their own political party, the BSP, and achieved electoral success in UP (Duncan Reference Duncan1999). While the developmental achievements of the BSP remain unclear, scholarship shows that it has empowered the Dalit population to become politically active and vocal (Duncan Reference Duncan1999). Describing the experience of one Dalit village with Mayawati’s program of constructing statues of Dr. Ambedkar, Jaoul (Reference Jaoul2006, 198) notes that “[t]o these villagers, installing a statue was a daring act that cashed in on the new power equation. It gave shape to their new status, enacting a political change that would otherwise remain beyond the realm of local reality.” On the other hand, Mehrotra (Reference Mehrotra2006, 4261) notes that “[s]ymbolic acts of defiance of the established ‘manuvadi’ order have indeed been dominant in UP, without much tangible benefits for the poor and the oppressed to show for it.”

The ideology and strategy of the BSP shed light on why and how the party’s politicians might be able to reduce inequality: “Although the BSP believes in total transformation, i.e., destruction rather than reform of the Hindu social and political order, this revolution is to take place not through social upheaval, but the ballot box … the first past the post system makes it possible for [Dalits] to come to power, and thereby seize power from within” (Pai Reference Pai and Lal2001, 62). Where the BSP wins an electoral seat, the local MLA can act to reduce inequality in several ways.

First, as noted above, the selection of villages within districts is a local process. In his or her own electoral constituency, a BSP MLA can exercise an influence on the selection of villages through contacts with the district committees that officially select villages. The MLA can also influence the selection of households within gram panchayats – India’s rural local governments, often comprising several villages – and thus ensuring that villages/habitations with Dalits are adequately covered. Finally, the MLA can also monitor the rural electrification process and report perceived bias to the state government.

A BSP MLA can also support rural electrification through complementary programs. One of the BSP’s key programs for Dalit empowerment has been the Ambedkar Villages Program (AVP), which channels state development funds into villages with a high percentage of Dalit population (Pai Reference Pai2004, 1145–1146). A BSP MLA can use AVP resources to support rural electrification and thus ensure that RGGVY implementation in Dalit villages succeeds.

Against these predictions, there are also reasons to be skeptical about the effect of electing particular MLAs. Corrupt, inexperienced, isolated or ineffective MLAs will hardly be able to counterbalance the effects of discrimination. These forces could neutralise the benign effect of representation.

During our study period, the UP state government was at times led by BSP. The BSP’s leading politician, Mayawati, had intermittently served as the state chief minister in the 1990s and then again from May 2002 to August 2003, after which BSP lost power because its coalition partner, the national Bharatiya Janata Party (BJP), withdrew. In 2007, BSP won an absolute majority in the state elections with a new electoral strategy that invited non-Dalits to support the BSP.

The BSP’s attempt to form coalitions with other parties raises questions about the party’s commitment to lower-caste empowerment. As Pai (Reference Pai and Lal2001, 63) explains, however, “[t]he BSP has tried to justify its alliances with upper caste parties since the mid-1990s, as not constituting a shift from its path of Dalit justice and upliftment but as short-term strategic alliances … capturing political power by any means is both necessary and justified in the case of a Dalit party in UP as without it social transformation is impossible.” Indeed, in the 2009 general election of India, the only caste group in UP that voted for BSP in the majority were the SC, both the jatav caste, which has historically been the key BSP constituency since its establishment, and other subcastes in the SC (Pai Reference Pai2014, 157). Despite the BSP’s alliances with non-Dalit political forces, the core voters of the party during our study period were the SC – all other social groups in UP favoured other political parties.

In sum, we expect policy discrimination that may or may not be mitigated by political mobilisation. Before turning to the analysis, we highlight some of the scope conditions for our argument. Our theoretical framework applies to situations in which (a) group antipathy is widespread, (b) one group is politically weaker than others, (c) there are few mechanisms available to the disadvantaged group to correct uneven policy implementation (e.g. easy access to courts or media). The effect of political mobilisation itself is contingent on the efficacy of parliamentary action, the willingness of politicians to monitor and exert pressure if they disagree with policy implementation. We believe that this model can be adapted to study other policy areas.

Dependent variable: RGGVY implementation

The primary dependent variable is a binary indicator for RGGVY implementation in a village. The Rural Electrification Corporation of India monitors the implementation of the RGGVY, and we use their master database as of October 2014. ^{Footnote 7} The database is a list of villages that have been electrified under the RGGVY. While the database does not contain comprehensive or reliable data for the exact timing of village electrification, it provides us with a crosssection of RGGVY implementation three years after the 2011 Census of India. For every village in UP, we thus know whether RGGVY was implemented in it within approximately a decade of the initiation of the program, between April 2005 and October 2014.

To quantify the consequences of caste inequality, we also exploit village-level data on the percentage of households with grid electricity access. We thus repeat our main analysis but replace the RGGVY implementation indicator with the share of electrified households. We expect this share to be lower when the percentage of SC people is high at the village level. ^{Footnote 8}

Note that access to the grid does not mean that households benefit from reliable electricity. Several studies have highlighted problems related to low hours of electricity actually available (e.g. Aklin et al. Reference Aklin, Cheng, Urpelainen, Ganesan and Jain2016). As a result, access to the grid does not necessarily mean that households benefit from abundant electricity. This being said, we believe that modelling connections (rather than, say, hours of electricity) make sense give our research interest. Electricity supply (i.e. hours) mostly varies across districts and subdistricts, as opposed to villages, because officials have no direct control over whether any particular village under a certain feeder receives electricity or not. Connections to the grid, on the other hand, are more immediately controllable by officials. In fact, analysing hours of electricity offers us the opportunity to run a placebo test: we would not expect the share of SC to affect supply for the reasons we stated. And indeed, this is what we find (Table A29 in Supplementary Material). ^{Footnote 9}

Explanatory variable: scheduled caste population

The primary explanatory variable is based on the SC population of the village. The 2011 Census of India provides both the total population and the SC population of every village in UP, and we can thus compute the percentage of SC population in the village. The results remain very similar with Census data from 2001 (Table A34 in Supplementary Material). Figure A3 in Supplementary Material demonstrates the considerable variation across villages in SC population. The average SC percentage in our dataset is 24.6%. In Figure 1, we show the geographic distribution of RGGVY implementation (share of villages electrified under RGGVY) and SC share (0–1 scale) across UP electoral constituencies. In the empirical analysis, we use constituency fixed effects to ensure that geography does not bias our estimates. In Table A35 in Supplementary Material, we show that the results remain qualitatively the same when allowing the share of SC to be quadratic.

Figure 1. Geographic distribution of RGGVY implementation and SC share by electoral constituency.

Control variables and split samples

We control for a select set of variables that could influence both the SC population percentage and the probability of RGGVY implementation. Section A1 in Supplementary Material presents the variables in greater detail. For each control variable, we also explain why it is not a “bad control” in that it would be influenced by SC population (Angrist and Pischke Reference Angrist and Pischke2009). For summary statistics, see Table A2 in Supplementary Material.

We begin with the logarithmised distance between the village and the nearest town. Because all towns in UP are now electrified, the cost and ease of implementing rural electrification works in nearby villages is much lower than the cost of such works in faraway villages. Next, we control for earlier village-level electrification status, as per the 2001 Census of India. Obviously, the status of village electrification in 2001 is a strong predictor for the need for RGGVY. We also control for the logarithmised population of the village. Larger villages tend to have higher electrification rates to begin with, so they may not need the RGGVY. Because RGGVY implementation requires infrastructure, we also control for the presence of a paved road. We include electoral constituency fixed effects in some models. These fixed effects allow us to compare villages close to each other and sharing similar political histories. It also helps us rule out competing explanations for our findings, such as those based on UP’s location with respect to national electricity sources.

Finally, we control for wealth-related confounders in Table A30 in Supplementary Material. One possible source of bias could be that wealthier villages, which tend to have lower SC shares, either have less need for electrification or are electrified first for economic gains. Poverty could thus be simultaneously correlated to policy implementation and SC presence, making our main estimates spurious. While controlling for wealth is difficult because of possible posttreatment bias, we can use data from the pre-RGGVY period to this effect. From the 2001 census, we include an index of wealth built on the share of the population that owns assets such as TVs, cars, radios and so forth (Filmer and Pritchett Reference Filmer and Pritchett2001). We also include other wealth proxies such as the village literacy rate (%), the number of cooperative banks and the (logarithmised) area of irrigated land. We also control for average nighttime luminosity from 1995 to 2004 using NOAA satellite data (further disaggregation by hamlet is not feasible due to data constraints). Our results remain virtually identical, suggesting that wealth is not biasing our results.

For a more stringent test of caste inequality in policy implementation, we also split the sample by 2001 village electrification status. We first examine whether among villages electrified in 2001, the size of the SC population predicts RGGVY implementation; we then conduct the analysis for villages that were not electrified in 2001. Relatedly, we discuss the relationship between village electrification and SC population before the RGGVY begins in Section A2 in Supplementary Material. We show that while villages without Dalits had slightly lower electrification rates pre-RGGVY, this was almost entirely driven by a number of small villages without any Dalits.

Results from the full sample

In Table 1, we show the results for the full sample of all UP villages. The upper panel shows the pooled regressions; the lower panel shows the estimates with constituency fixed effects.

Table 1. Dependent variable: RGGVY (if present, RGGVY = 100). The standard errors are clustered by constituency

Standard errors in parentheses.

*p < 0.10, **p < 0.05, ***p < 0.01.

The negative correlation between the SC percentage and RGGVY implementation at the village level is large and robust. Depending on the model, increasing the SC population by 10 percentage points (1/2 standard deviation) reduces the probability of RGGVY implementation by about two percentage points. Comparing a village populated by Dalits to one without any Dalits, the difference is thus about 20 percentage points – a massive difference, when only 31% of all villages in UP saw RGGVY implementation. The result also cannot be attributed to differences in 2001 village electrification status, as controlling for this variable makes no difference whatsoever.

When we repeat the estimation for the major regions of the state (Western, Central, Eastern and Bundelkhand), we see evidence of inequality in all four areas (Section A8 in Supplementary Material). This not only demonstrates the robustness of the main result but also shows that major differences in socio-economic characteristics do not seem to eradicate inequality. In fact, the relatively wealthy western parts and the poor Bundelkhand region have the strongest negative associations between SC share and the likelihood of RGGVY implementation.

Split samples

Table 2 shows the estimates separately for villages that were or were not electrified according to the 2001 Census of India. All models are estimated with constituency fixed effects. As the table shows, the result does not depend on prior electrification status. Regardless of whether a village was electrified in 2001 or not, increasing SC population by ten percentage points reduces the probability of RGGVY implementation by about two percentage points. Importantly, this lack of difference allows us to rule out the possibility that the positive association between Dalit population and village electrification in the 2001 Census of India would confound our estimates.

Table 2. Dependent variable: RGGVY (if present, RGGVY = 100). All models estimated with constituency fixed effects. The standard errors are clustered by constituency

Standard errors in parentheses.

*p < 0.10, **p < 0.05, ***p < 0.01.

Consequences of caste inequality

In Table 3, we summarise the consequences of caste inequality by replacing RGGVY implementation with the percentage of electrified households in the 2011 Census of India. The various models reveal the cost of caste inequality to the SC population: as the SC share of a village increases by ten percentage points, the village electrification rate decreases by about 1 percentage point. Thus, a comparison of a Dalit village to one without any Dalits would thus show a difference of 10 percentage points.

Table 3. Dependent variable: household electrification in 2011 (0–100 percent). The standard errors are clustered by constituency

Standard errors in parentheses.

*p < 0.10, **p < 0.05, ***p < 0.01.

We confirm these findings with survey data from Aklin et al. (Reference Aklin, Cheng, Urpelainen, Ganesan and Jain2016). Table A45 in Supplementary Material shows that Dalit households are 15 percentage points less likely to have grid electricity connections, again consistent with the notion that the lack of RGGVY implementation is hurting Dalit households.

Sample and model

The basic unit is now a village election (elections were held in 2002 and 2007) and the outcome a binary indicator for RGGVY implementation. The treatment is assigned at the constituency-election level: did a BSP candidate win against a non-BSP candidate by a narrow margin? We map villages into constituencies by using official delimitation lists from Jensenius (Reference Jensenius2015) that allow us to link every village in UP to a unique constituency. ^{Footnote 10}

We estimate models at the 1, 2 and 5% margin of victory. Larger bandwidths benefit from stronger statistical power, but at the cost of more potential for bias. The Imbens and Kalyanaraman (Reference Imbens and Kalyanaraman2012) test of optimal bandwidth suggest that the best margin is about 3.2, and the results also hold with this bandwidth. For discontinuity plots, see Figures A7–A9 in Supplementary Material; for additional bandwidths, see Table A20 in Supplementary Material.

Let $i$ denote a village, $j$ a constituency and $k$ an election period (either after 2002 or after 2007 elections). The estimation equation can be written as follows:

(2) $${Y_{ijk}} = \alpha + {\beta _1}{\rm{BS}}{{\rm{P}}_{jk}} + {\beta _2}{\rm{S}}{{\rm{C}}_{ij}} + {\beta _3}{\rm{BSP}}*{\rm{S}}{{\rm{C}}_{ijk}} + X_{ijk}^{'}\gamma + {{\rm{\epsilon}}_{ijk}},$$

where ${Y_{ijk}}$ is again the dependent variable, SC the relevant population percentage and BSP an indicator for the party’s electoral victory. The vector of control variables now includes the forcing variable, that is, the margin of victory of BSP victory (negative when BSP barely loses) in percentages of valid votes. ^{Footnote 11}

Standard errors are conservatively clustered by constituency throughout. Because there were two elections (2002 and 2007) between the 2001 and 2011 Censuses in UP, each village appears in the dataset twice, once for each election. ^{Footnote 12} The clustering of standard errors ensures that this procedure does not result in double-counting of observation and thus the artificial deflation of standard errors (e.g. Folke Reference Folke2014). ^{Footnote 13}

Explanatory variable

The primary explanatory variable is a BSP victory in close elections. Because BSP portrays itself as the party of the Dalits and other oppressed people, it is the primary mode of political mobilisation against caste inequality in UP. Because BSP victory can be considered quasirandom in close elections, we can use it to identify the effect of this kind of political mobilisation on inequality.

In some models, we interact the treatment with the SC percentage in a village. If BSP truly protects SC populations, then electing a BSP MLA should have a particularly large positive effect on RGGVY implementation in villages with a large SC population.

We also estimate models with a triple interaction between BSP win, SC share and an indicator for an SC-reserved constituency. This test allows us to investigate whether a BSP win has differential effects when the candidates must be from the SC. The difference might be negative if SC candidates from non-BSP are sympathetic to the plight of the lower castes; or might be positive if it ensures that the BSP MLA is actually himself or herself from a SC.

Figure A4 in Supplementary Material shows the margin of victory for each constituency election in the RDD sample (5% margin of victory). Table A9 in Supplementary Material, in turn, summarises the sample for the RDD analysis. In total, we have 235 close constituency elections when the sample is restricted to a 5% margin of victory. Finally, Table A10 in Supplementary Material compares BSP and non-BSP MLAs. As the table shows, both the candidate and constituency characteristics are mostly similar. The only exception – an unsurprising one – is that BSP MLAs tend to come from SC-reserved constituencies.

Tests of the identifying assumptions are found in Section A3 in Supplementary Material. We find balance over pretreatment covariates across samples. Similarly, following McCrary (Reference McCrary2008), we find no suspicious discontinuity, alleviating concerns about electoral fraud and other irregularities.