Development from Representation? A Study of Quotas for Scheduled Castes in India

Development from Representation? A Study of Quotas for Scheduled Castes in India Draft Manuscript. Please do not cite. Francesca Refsum Jensenius U.C. Berkeley, Department of Political Science Abstract An important intervention in the political representational system in India is the extensive quota system for Scheduled Castes (SCs) in the national and state-level legislatures, which has been in place since independence. A series of papers about quotas at the village level in India have shown that who governs might affect the type of public goods provided to the citizens, but the evidence presented in these papers is contradictory and not always clearly identified. Data scarcity has also made it difficult to identify the same relationship at the state level. The selection of the seats that were reserved 1974-2008 was based on 1971 census data. This paper introduces a new dataset, merging Indian census data from 1971 and 2001 with political data at the state assembly constituency level. The data covers about 3000 political constituencies in the 15 largest Indian states in 1971 and 2001. Matching constituencies on the same variables and same data used to select the reserved seats in 1971, the paper compares the development trajectories of places with and without reservations. Development is operationalized as the literacy rate and the percent non-workers among SCs and non-scs at the constituency level. The paper concludes that in the aggregate there is no constituency-level effect of representation on the literacy rate or the percent non-workers among SCs or non-scs. frj@berkeley.edu 1

1 Introduction Since the Indian constitution came into force in 1950, Scheduled Castes (SCs) and Scheduled Tribes (STs) have been guaranteed political representation in the Lok Sabha (the lower house in parliament) and in state legislative assemblies through political quotas ( reservations ). According to the Indian constituent assembly debates, SCs/STs were given electoral quotas apparently and clearly on grounds of their economic, social and educational backwardness (CAD, 1999, vol 3, p. 308). India s first Prime Minister, Jawaharlal Nehru, was principally against group-wise representation, but argued in favor of reserved seats for SCs/STs because this was about helping backward groups in the country (CAD, 1999, vol. 3, p. 331). The motivation was to improve the political access and the socio-economic standing of the most deprived groups in the country. The reservation policy was originally meant to last for only 10 years, but was extended every time it expired. During the meeting about extending the policy in 1969, the Swatantra Party MP M.R. Masani objected: No proof has been given by the hon. Minister or anybody else to show that this reservation has in practice led to concrete advance and benefits for this class (Lok Sabha Debates, December 8 1969, p. 299-300). His comment was ignored, and more than 40 years later we still have not clear response to his objection. Existing literature about local level quotas in India, suggests that politicians tend to act in the interest of their own group, and that we should expect to see a linkage between representation for SCs and development for the SC community. The median voter theorem, on the other hand, posits that politicians in a plurality voting system have to cater to the interests of the median voter in their constituency. In practical terms this usually means the majority group. The way the quota system is designed in India, SCs are almost always a minority in reserved, as well as in general, constituencies. Since SC politicians face the same political incentives as politicians in general constituencies, we 2

should therefore not expect to see any difference in the development levels in general and reserved constituencies. This paper explores two indicators of development, the literacy rate and the percent non-workers, among non-scs and SCs, in general and SC-reserved constituencies. In order to do so, the paper makes two original contributions. First, I describe the process by which reserved seats were selected, based on archival work in the Election Commission record room. Second, I introduce a new dataset, merging census data and political data from 1971 and 2001. 1 Exploiting the knowledge about the selection of reserved seats and this new data, I match variables on the exact same measures used to select the location of the reserved seats in the 1970s. Since the balance on observables of the matches is excellent I assume balance on unobservables and therefore compare the growth in literacy and the percent non-workers among non-scs and SCs in general and reserved constituencies. It is found that in the aggregate, there is no constituency level effect of SC reservations on the literacy rate nor on the percent non-workers among SCs or non-scs. I argue that the design of the quota system, as well as the influence of political parties on candidate selection, prevent SC politicians from acting differently than other politicians. However, I also suggest a few alternative explanations for this finding and outline how I plan to explore these alternative explanations in other papers. The findings in this paper do not suggest that reservations have not had an important emotional or political effect, but rather that there is a disjuncture between the design of the quota system and the common expectations of its effects. This means that the quota system is widely criticized for something it cannot be expected to deliver. 2 Quotas and representation Pitkin (1972) changed the debate about political representation by distinguishing be- 1 This dataset has been developed in collaboration with Dr. Rikhil Bhavanani. 3

tween its four different aspects: formalistic, descriptive, substantive and symbolic. The study of formalistic representation focuses on the rules that determine who is elected to represent. Descriptive representation captures the extent to which politicians resemble the people being represented. Symbolic representation is about the emotional response invoked in the represented group by their politician, while substantive representation is about whether the politician acts in the interest of those represented. This paper is part of my dissertation in which I explore the linkage between descriptive and substantive representation in India, where the descriptive representation is ensured through electoral quotas rather than through an electoral majority. Some 100 countries in the world use some form of political quotas. Quotas are a form of social planning aimed at guaranteeing equality of results, rather than facilitating equality of opportunity (Klausen and Maier, 2001). They can be implemented as aspirant quotas (a minimum number of the group is required among pre-candidates in parties), candidate quotas (a minimum number of the group must be fielded as candidates) and reserved seats (an elected political position can only be held by an individual belonging to the group) (e.g. see Htun, 2004; Dahlerup, 2006; Matland, 2006). All of these types of quotas can be voluntary or mandatory through the constitution or through electoral laws. The most common target groups (intended beneficiaries) for quotas are women, ethnic minorities such as indigenous groups and racial minorities, people with physical disabilities, and also lower caste groups. 2.1 Reservations in India today India is a constitutional democracy with a parliamentary system of government. The Indian Constitution grants adult suffrage (Article 326), and there are currently more than 670 million registered voters. The 552 members of the lower house and the more than 4000 members of state assemblies are elected from single-member electoral districts (referred to as political constituencies) using a plurality voting system. The use 4

of single-member districts makes the threshold to political entry high, favoring the representation of larger groups in society. As a counterweight to the majoritarian electoral system, Scheduled Castes (SCs) have constitutionally guaranteed reserved seats in the Lok Sabha (the lower houses of the national parliament) and State legislatures, in proportion to their population in each state (on average 16%). 2 In reserved constituencies, only individuals belonging to a SC can run for election, while the whole electorate votes. Table 1: Assembly seats reserved for SCs in 15 Indian state assemblies 1974-2008 State Assembly SC Percentage seats seats SC seats Andhra Pradesh 294 39 13.3 Bihar 324 47 14.5 Gujarat 182 13 7.1 Haryana 90 17 18.9 Himachal Pradesh 68 16 23.5 Karnataka 224 33 14.7 Kerala 140 12 8.6 Madhya Pradesh 320 43 13.4 Maharashtra 288 17 5.9 Orissa 147 22 15.0 Punjab 117 29 24.8 Rajasthan 200 32 16.0 Tamil Nadu 234 42 17.9 Uttar Pradesh 425 90 21.2 West Bengal 294 59 20.1 Every ten years, after the decennial census, the Election Commission of India is supposed to appoint a Delimitation Commission to redraw political boundaries. At the same time, they are supposed to select which political seats are to be reserved for SCs and STs. At the time of the delimitation following the 1971 census, the population 2 Scheduled Tribes (STs) have about 7% of the seats in the country, but these reserved seats are quite concentrated with some states having a high proportion of ST seats. In this paper I focus exclusively on the SC seats 5

size of the constituencies was meant to be the same, but it soon became evident that the high population growth in certain parts of India would lead to a dramatic change in the number of political seats granted to each state. To avoid encouraging higher birthrates with more political representation, the government of India chose to freeze the boundaries of the political constituencies, and, with it, the geographic location of the reserved seats. A new delimitation was first initiated after the 2001 census, and the new constituency boundaries were used for the first time in 2008. The result was that political constituencies remained the same from 1974 until 2008, allowing us to study places that have been continuously reserved for more than 30 years. Table 1 summarizes the number of state assembly seats that were reserved for SCs in the 15 largest states according to the delimitation in the 1970s. There were two selection criteria according to the Delimitation Act of 1972: (1) that the proportion of SCs should be high in selected constituencies and (2) that the reserved constituencies should be geographically spread out within the state (Gazette of India, December 30 1972). In practice this meant that states and then districts were assigned a reserved seats on the basis of the proportion of SCs in their population. If a district was eligible for a reserved seat, the constituency with the highest proportion of SCs within the district was assigned to be reserved. If a district was eligible for more than one seat, they were usually distributed between subdistricts or blocks. 3 The result of this selection process is great variations in the proportion of SCs in reserved constituencies, and several examples of constituencies with high proportions of SCs that were not reserved. In fact, the percentage of SCs living in reserved constituencies in India between 1974 and 2008 ranged from 4% to 66.5% (in Bihar and West Bengal respectively), while there were also general (non-reserved) constituencies where SCs constitute up to about 50% of the population. 3 The source of this information is the state-wise folders in the Election Commission Record room related to the Delimitation process, accessed February 2011. 6

3 Theorizing development outcomes 3.1 Can MLAs effect development? It is an often repeated truism that political quotas for SCs were implemented in order to uplift the SC community. With upliftment people generally mean socio-economic development. But can Members of Legislative Assemblies (MLAs) effect development? In 2010-11 I conducted interviews with politicians, bureaucrats and political activists in several Indian states about the work of MLAs and SC MLAs in particular. 4 According to the conversations with my respondents, there are at least four ways in which MLAs can increase development in their constituencies: A politician can work in the legislature to pass policies that are beneficial for their constituents. Generally, though, there is minimal discussion of bills in the state legislatures in India, and MLAs tend to vote according to party line. This therefore seems to be perceived as the least important way in which an MLA can affect development patterns in their constituency. A politician can benefit a group by functioning as a fixer who facilitates access to developmental schemes, speed up the bureaucratic processes, and influence job transfers. 5 One common example given by my interview respondents was that MLAs tend to push the administration to hire people from their own networks in the public services or try to have them transferred into more favorable positions. One top-level bureaucrat I talked to estimated that half of the interaction between MLAs and his/her constituents were about transfers. He gave an example that under the sitting government in his state, all the school principles appointed by the education ministry were from the same caste as the minister. 4 I conducted about 70 interviews with MLAs, pradhans, IAS officers and activists. The bulk of the interviews were done in Delhi, Shimla, Lucknow, Meerut, Varanasi and Bangalore. The interviews were semi-structured, lasting between 15 minutes and 3 hours each. In Shimla, Varanasi and Delhi the interviews were conducted alone, while in the rest of the locations I worked with a collaborator or research assistant. 5 Beteille (2009) finds in a survey of teachers that 70-85 % think that connections are essential for getting a favorable job transfer, suggesting that the role of politician as fixer plays a major role in Indian politics. 7

Since 1993, Indian MLAs have had access to so-called MLA-LADS, development funding that can be distributed at the MLA s discretion. Most of the MLAs I talked to made it clear that they only spend this money in areas where people have voted for them in large numbers. A few respondents spoke of an emotional effect in areas with reservations, where parents from an SC community might be more likely to educate their children in reserved constituencies because of an increased belief in social mobility. 3.2 Literature about quotas in India The interest in quotas in India has increased over the last few years. The focus of this interest has been the local level (panchayat) electoral quotas that were implemented after the Constitutional (73rd Amendment) Act came into force in 1993. Several excellent studies of women s quotas show that women s quotas at the local level lead to an increase in the provision of public and private goods. 6 With data from West Bengal and Rajasthan, Chattopadhyay and Duflo (2004a,b) show that women politicians tend to invest more in goods that women express an interest in. Extending this study, Duflo and Topalova (2004) show that villages with women s quotas tend to see less local level corruption and have a better supply of drinking water facilities. Bardhan et al. (2009) find evidence from West Bengal that places where the pradhan post was reserved for SCs/STs, saw an increase in benefits to the village as a whole, as well as an increase in the goods targeted to female-headed households and the group of the pradhan. These findings support the idea that there might be a linkage between descriptive and substantive representation in reserved constituencies. The authors also find evidence for a decrease in the targeting of SCs/STs in panchayats reserved for women, and attribute these findings to women having less political experience. Thus, from this existing literature we can conclude that politicians might try to benefit their 6 Here public goods should be understood as non-excludable goods provided by the government, such as roads and irrigation. Private goods on the other hand are excludable goods such as employment schemes. 8

own group and that experience in politics is key to the effective provision of goods. 7 Very few attempts have been made to discern the effects of quotas at the state or national level. This is mainly because of a lack of data at the political constituency level, since most data are collected for administrative units (districts, blocks and villages) which are geographically different from political units. Thus, studies of the state level quota system have focused on effects of political outcomes for which data is compatible, such as the effect on party strategies (Jaffrelot, 2003) and voter turnout (McMillan, 2005). However, these studies are not adequately controlling for the structural differences between constituencies chosen to be reserved and general. Some attempts have also been made to study the state level quotas with the state as the unit of analysis, which in fact becomes a study of the effect of the proportion of SCs in the state. Clots-Figueras (2005) tries to get around this by studying the proportions of female legislators in general and reserved constituencies (16 cases 1967-1999) and then looks at policy outcome at the state level. She finds evidence that states with a high proportion of women in SC seats tend to invest more in low level education and irrigation and be more likely to vote to improve women s legal rights. Pande (2003) estimates the effect of the proportion of SC quotas on economic transfers to SCs/STs, using changes in the population and changes in the number of reserved seats as a way to identify the changes in the transfers made to SCs/STs. This study finds that the transfers to SCs/STs increase with SC/ST representation, suggesting that there indeed is a relationship between descriptive and substantive representation in the Indian case. However, since virtually no changes were made in constituency borders since 1976, this study is mainly based on data from the 1960s and early 1970s. Also, since the outcome variables are at the state-level allocation of resources and it is 7 There are also several studies of how people attitudes change as they have experienced living in a panchayats reserved for women. Several articles suggest that the quotas have a long-lasting political effect, as they increase the representation for women even after the quotas are gone(bhavnani, 2009; Beaman et al., 2008). Duflo and Topalova (2004) also show that although women leaders seem to be doing a better job than their male counterparts, villagers are less satisfied with their leadership. 9

well-known that a limited amount of such funds actually reach the citizens, it is unclear whether SC/ST quotas have had a development effect on the ground. It should be noted that while the literature about quotas tends to expect more development for SCs in SC constituencies, a common statement among the respondents in my interviews was that there probably has been less development in reserved constituencies. The reasons given for this were mainly that SC politicians were not as experienced as other politicians, were less educated, had fewer contacts, and that they channelled money mainly to the SC minority thereby pulling down the rest of the population. 3.3 The median voter theorem and Ambedkar The incentives for political actors have long concerned scholars of political science. In their simplest form, economic models posit that political parties are office seeking and therefore have an incentive to converge on the policy position of the median voter (Downs, 1957). Looking at the individuals within the party organization, scholarship about the motivation of members of the American Congress suggests that they are motivated by reelection, public goods provision, career prospects, their party gaining majority in Congress, and their own reputation among other politicians (e.g Fenno, 1978; Kingdon, 1989; Cox and McCubbins, 2005). However, many follow Mayhew (1974) in making the simplifying assumption that politicians mainly are office seeking. This is not because it is thought to be the only interest driving the behavior of politicians, but rather because it provides considerable traction to attempts of understanding real-life political dynamics. Since both India and the US have a plurality voting system in single-member districts and politicians therefore face a similar incentive structure, similar expectations might apply. The interesting difference between the US and Indian cases is that the electoral quota system creates many political units in which SCs compete for election 10

in constituencies that have a very small proportion of SCs. Thus, candidates can only win an election by appealing to the majority group in their electorate, and parties will only field candidates who they think will adhere to party line. Thus, we should expect to see that both SC and non-sc candidates have an incentive to try to attract votes from the largest groups in their constituency. Im fact, even in only some politicians act this way they will win the election and prevent the others from getting to power. Thus, if SCs do not constitute a large group in the constituency,politicians have no incentive to try to attract their votes. This argument is precisely the one made by the SC leader Dr. Bhim Rao Ambedkar at the time when the quota system was developed. After The British Prime Minister Ramsay MacDonald announces the Communal Award in 1932, in which SCs were granted separate electorates, Mahatma Gandhi went on hunger strike to protest against it. The British refused to change the Award without the consent of Ambedkar, and he was pressured into signing the Poona Pact. In the pact, SCs gave up the claim for separate electorates, in return for 151 reserved seats in provincial assemblies and 18% of the seats in the central legislature, elected in joint electorates (Poona Pact 1932). Ambedkar saw this agreement as a failure, because the elected SCs were no longer elected by and answerable to an untouchable electorate : the result is that the legislatures of the minority elected to the reserved seat instead of being a champion of the minority is really a slave of the majority (quoted in Samujh, 2005, p. 59). In his book Mr. Gandhi and the Emancipation of the untouchables, Ambedkar (1943, p. 24-25) argues that Congress had wanted joint electorates precisely in order to control politicians elected from reserved seats: [S]eparate electorate does not permit the Hindus to capture the seats reserved for the Untouchables. On the other hand the joint electorate does. [... If] there is a joint electorate in these constituencies the representative of the Untouchables would be only a nominal representative and not a real 11

representative, for no Untouchable who did not agree to be a nominee of the Hindus and a tool in their hands could be elected in a joint electorate in which the Untouchable voter was out numbered in ratio of 1 to 24 or in some cases 1 to 49. 4 Hypotheses and identification strategy From the interviews I conducted with bureaucrats, politicians and activists, it seems clear that many believe that MLAs can effect the development in their constituency, and that they can channel resources to a specific group if they want to. It is not certain, however, that they would want to channel funding to their own caste group. Three main hypothesis come out of interviews and the existing literature: From the existing literature on India we have reason to expect a positive relationship between SC quotas and the substantive representation of SCs. If SC politicians systematically have channelled resources and favors to their own caste group through 30 years, we should expect to find a higher level of development for SCs living in reserved constituencies than in general constituencies. If SC politicians respond to the political incentive of catering to the majority group in order to be re-elected, as suggested by the median voter literature and arguments made by the SC leader Dr. Ambedkar, we should expect to find no systematic difference in the behavior of politicians, and consequently no difference in the level of development for SCs or for non-scs. If SC politicians are systematically less effective than other politicians, we should expect to see less development in reserved constituencies than in general constituencies. 12

4.1 Identification strategy The main challenge in this study is that the assignment of quotas in the 1970s was not random. Non-random assignment to treatment means that the differences we observe in reserved and non-reserved constituencies today could be the result of confounding variables related to the selection mechanism. In this paper I try to get around this challenge by identifying how the selection of reserved seats took place and then match constituencies on the selection mechanism. Since the selection mechanism for reserved seats is known, and since there is considerable overlap in the distributions of reserved and general seats, a matching model can ensure good balance on observable and unobservable confounding variables. To get as close as possible to the selection mechanism, I chose to match constituencies on district and then do a closest neighbor match on the proportion of SCs in a constituency. 5 Data collection and operationalization 5.1 Getting the baseline data The units of analysis in this paper is the political constituency, and the baseline data is Indian census data from 1971. There were two major challenges to merging political data and 1971 census data. First, the data was not electronically available. Second, the geographical units of the census are different from political constituencies. 8 The first problem was solved by scanning the books with block-level data (the administrative level below district, usually called tehsil or taluk) for the 15 largest Indian states in 1971. The scanned copies where OCRs in order to extract the information, and the electronic files where then manually cleaned for errors. 9 The second problem was solved 8 I collaborated with Dr Rikhil Bhavnani in solving these problems and developing the data. 9 The OCR left many errors and the cleaning work consisted of setting up logical tests for all the data such as male+female=total, rural+urban=total, sum(all blocks in a district)=district etc. Where these tests were negative, the numbers were checked with the original books and corrected. In cases where the numbers in the original books were clearly wrong, we followed the original books. I am very grateful to the 13

by manually writing a merging code for the political data and the census data. The estimated census values for each constituency are based on block level data weighted by population. 10 The result of this data work are estimates for all the census variables from the Primary Census Abstract (PCA) and the SC/ST PCA for each of the political constituencies in the 15 largest Indian states. Creating population-weighted estimates means that there are some inaccuracies in the data, but these inaccuracies should not be systematic. We there believe that the estimates are unbiased in expectation. Thus when used in analyses as part of a large sample, the estimates can be treated as unbiased. 5.2 Operationalizing and acquiring outcome variables For more recent years, some data is publicly available electronically at a disaggregated level, but the problem remains that most data is collected at the level of administrative units, which differ from political constituencies. In this case we chose to use GISmapping to merge the 2001 census with political units. 11 In this case the merging was approximately 20 research assistants who have helped out with this tedious work. 10 We started out with block level census data and the Delimitation report of 1976 specifying which blocks fall into each constituency. We had the total population of the constituencies from Election Commission documents from the 1970s and the population for each block in the census data. The merging files were created on the basis of mathematical calculations of the proportion of the population of a block that falls into a constituency. For example, if the Delimitation report listed that Constituency x (population 150000) consisted of all of block A and part of Block B (each having a population of 100000), then the census values for Constituency x = Values for Block A + Half the values for Block B. In some cases, two or more constituencies consisted of parts of the same two blocks. In these cases the exact proportions could not be calculated. We solved these cases in two different ways. In most cases, we used the notes about the exact population proportions referred to in the notes of the Delimitation Commission from the 1970s. These notes were accessed February 2011 in the record room of the Election Commission in New Delhi. These records were soon thereafter transferred to the National Archive in New Delhi where they are now available. In the few cases where we could not find written sources among the records of the Delimitation commission, we used estimates of the population size of villages in that region as well as the number of villages in the constituency, to estimate how much of a block fell in each constituency. The assumptions made are listed in the data documentation which will be released with this data. 11 This was actually the only solution we could find. The large surveys, such as NSS, do not have indicators for political constituencies and do not release the indicators for geographic location below the district level. Between 1971 and today, the district and village borders have also changed extensively, making manual merging on the basis of the 1976 Delimitation report almost impossible. GIS mapping is therefore the only 14

done from block-level data by area-weighting. Since there is no reason why inaccuracies should be correlated with the outcome variables, we believe that the estimates can be treated as unbiased in expectation. The census includes variables such as population size, occupation, numbers of literate and non-literate, as well as non-workers. In this paper the goal was to explore the extent to which SCs are uplifted by living in a constituency reserved for SCs. From the variables available in the census I chose to focus on two sets of variables in this paper. The fist variable is literacy rate, because schooling often is referred to as the main route to upliftment for SCs in India and because and MLA does have the power to affect literacy rates through checking up on whether teachers show up at schools, whether SC children get access to the classroom and by channeling funds into books, uniforms or scholarships. The second variable is the percent non-workers in a constituency. Having a stable income is crucial to development for a family, and because MLAs do have the power to affect who gets hired. In order to separate out the effects of the potential targeting of benefits to SCs in SC constituencies, I chose to look at the literacy rate and percent non-workers among SCs and non-scs separately. The final dataset used in this paper has data from 1971 and 2001 for 2627 general seats and 511 SC seats. 12 6 Empirical Analysis Most generalizations about quotas in India have been made by comparing reserved and general constituencies at the present time. In this case the outcome variables we are looking at are the literacy rate and the percent non-workers among SCs and practicable merging method. The only large dataset with development indicators which releases GIS maps is the census of India. The last census released before all the political boundaries were redrawn in 2008 was the census 2001. 12 All constituencies reserved for STs have been excluded from the data and there are also still some missing values in the case of urban areas, where the GIS-mapping did not work and where we are still trying to create manual matches based on detailed ward-level data. 15

non-scs according to the 2001 census. Comparing the raw data shows there is a small difference in the overall literacy rates between reserved and general constituencies. In general constituencies the average literacy rate was 54.8%, as compared to 52.6% in reserved constituencies. unpaired t-test, p 0.001). The difference in means is statistically significant (Welch s There is no overall difference, however, in the number of non-workers in reserved and general constituencies. In both types of constituencies there is an average percent of 59% non-workers (Welch s unpaired t-test, p=0.86). The difference in literacy rate is larger if we look at at the non-scs and SCs separately. Figure 1 shows the distribution of the literacy rates among non-scs and SCs in general constituencies and constituencies reserved for SCs. In this case, there is a statistically significant difference of about one percentage point difference in the average literacy rates for non-scs (p=0.05). Among SCs living in reserved constituencies, the average literacy rate is 44%, while the average literacy rate among SCs living in general constituencies i 46.9%. This difference is highly statistically significant (p 0.001). Figure 1: Distribution of the literacy rate among non-scs and SCs in general and reserved constituencies in 2001 Non SC population SC population Percentage 20 40 60 80 Percentage 20 40 60 80 GEN SC GEN SC Constituency type Constituency type 16

Figure 2: Distribution of the literacy rate among non-scs and SCs in general and reserved constituencies in 1971 Non SC population SC population Percentage 0 20 40 60 80 Percentage 0 20 40 60 80 GEN SC GEN SC Constituency type Constituency type One should be careful about drawing conclusions from looking at this data from 2001. As discussed above, constituencies were chosen to be reserved in the 1970s on the basis of having a high proportion of SCs in the population. The literacy rates were not only much lower across the board, but also lower in reserved constituencies than in general constituencies. Figure 2 shows the distribution of literacy rates among non-scs and SCs in general and reserved constituencies according to the 1971 census. Among non-scs, the literacy rates in general and reserved constituencies was on average 31.1% and 29.2% respectively. Among SCs, an average of 16.1% were literate in general constituencies versus 13.2% in reserved constituencies. Both of these difference in means are highly statistically significant. The figures show that the literacy rates in reserved constituencies started out lower than in general constituencies in 1971 and remained lower in 2001. How have the constituencies developed in comparison to each other? Figure 3 shows that the literacy rate for SCs was lower than for non-scs in both general and reserved constituencies, 17

but that this gap diminished over time in both types of constituencies. Figure 4 shows the same pattern for non-workers, which has been remarkably similar from 1971 till 2001. Figure 3: Change in average literacy rate among non-scs and SCs in general and reserved constituencies 1971 to 2001 General constituencies SC constituencies Percent 10 20 30 40 50 60 Non SCs SCs Percent 10 20 30 40 50 60 1971 2001 Year 1971 2001 Year 18

Figure 4: Change in the percent non-workers among non-scs and SCs in general and reserved constituencies 1971 to 2001 General constituencies SC constituencies Percent 50 55 60 65 70 75 Non SCs SCs Percent 50 55 60 65 70 75 1971 2001 Year 1971 2001 Year 19

6.1 Matching model From the descriptive statistics presented above, it looks like the development in literacy rates and the percent non-workers has been comparable in reserved and general constituencies, both among SCs and non-scs, but these descriptive statistics are still bias by the difference in starting point. In order to eliminate this bias, I matched constituencies on the variables used to select the seats to be reserved in the first place in order to eliminate this bias. Each of the reserved constituencies in the sample was matched to the general constituency in the same district. The matched constituency was the one with a proportion of SCs in the population closest that that of the reserved constituency. In most cases this meant that within every district with a reserved constituency, the constituency with the highest proportion of SCs was matched to the one with the second highest proportion of SCs. 13 By matching on the selection mechanism of reserved seats, I assume that the matched pairs can be treated as exchangeable, that they are as if randomly assigned to be reserved or general. This means that I assume that there is balance on all possible confounding variables. This assumption seems plausible given the fact that we see a much improved balance on the proportion SCs in a constituencies and almost perfect balance is achieved on the observed variables in the dataset (although I did not match on these variables, only on the percent of SCs in a constituency). Figure 5 shows the balance before and after matching. The difference after matching is still much smaller than before. Before matching the average percentage of SCs in general constituencies is 14.1% compared to 23.1% in the reserved constituencies. After matching the general constituencies have an average of 21.1% of SCs, while the percentage in reserved constituencies remains the same. That there is still a difference is not surprising, since reserved seats were selected because they had the highest proportion of SCs in the district, but the balance 13 In cases were there were more than one reserved seat in a district, the reserved seats could be the rules of the delimitation committee not be right next to each other. In those cases the matches are sometimes the third highest or fourth highest proportion. 20

is much improved. Figure 5: Balance on the proportion of SCs in general and reserved constituencies, before and after matching Before matching After matching Density 0.00 0.02 0.04 0.06 0.08 General seats SC seats Density 0.00 0.02 0.04 0.06 0.08 General seats SC seats 0 10 20 30 40 50 60 0 10 20 30 40 50 60 Percent SC in constituency Percent SC in constituency Table 2 shows the balance statistics before and after matching for a selection of observed variables. The table reports p-values from a t-test and a bootstrap Kolmogorov- Smirnov (KS) test. 14 P-values close to 1 suggest that the average value in the group of general and reserved constituencies are statistically insignificant from 0, in other words that they are very similar to each other. The matched pairs consist of 480 reserved constituencies and 480 general constituencies that are very similar in all ways, expect that the reserved constituencies still have 2% more SCs in the population on average. Table 3 reports the matching estimates for six different outcome variables. The first 14 These are the default tests provided by the function MatchBalance in the Matching package developed for R by Jas Sekhon (2011). A two-sample t-test is used before matching, since there is no reason to assume the same variance in the two samples before matching, while a paired t-test is used after matching. The KS is a nonparametric test of the difference in one-dimensional probability distributions. 21

Table 2: Difference in means for treated and control and Balance output from matches Covariate Before matching After matching t p-value KS p-value t p-value KS p-value Percent ST 0.28 0.11 0.18 0.77 Non SC non-workers 0.46 0.36 0.46 0.58 SC non-workers 0.29 0.14 0.47 0.84 Literacy rate non-sc 0 0.04 0.16 0.38 Literacy rate SCs 0 0 0.55 0.82 Agricultural labor non-sc 0 0 0.84 0.96 Agricultural labor SC 0.91 0.88 0.59 0.96 reported variable is the difference in the change in the non-sc literacy rate in reserved and general constituencies 1971-2001. In other words, the estimate takes the growth in literacy 1971-2001 in reserved constituencies (which was 25.72 percentage points) and subtracts the growth in literacy 1971-2001 in general constituencies (which was 26.03 percentage points). The difference of 0.31 percentage points suggests that on average the literacy rate among non-scs in reserved constituencies grew 0.31 percentage points more non-scs in general constituencies. This difference is, however, both substantively small and statistically insignificant. The second variable I present in the table is the difference in the growth in literacy among SCs in reserved and general constituencies. Once again we see an estimate that is close to 0 and statistically indistinguishable from 0. The third estimate compares the difference in the growth for non-scs and SCs in the two types of constituencies. In this case I subtract the growth of non-scs from the growth of SCs in general constituencies, do the same in reserved constituencies and then compare these two numbers. The three last variables repeats the same procedure for non-workers among non-scs and among SCs in reserved and general constituencies. As is clear in the table, none of the variables are substantively nor statistically significant. During the 30 years from 1971 and 2001 there is simply no difference in the rate of growth in literacy for SCs or non-scs or in the change in number of non-workers in general and reserved constituencies. 22

Table 3: Matching estimates of the difference in literacy rates and percent non-workers in general and reserved constituencies Variable Estimate St. error P-value 1 Non-SC literacy 0.31 0.25 0.21 2 SC-literacy 0.04 0.23 0.85 3 Difference in percent literacy 0.27 0.26 0.3 4 Non-SC non-workers 0.21 0.23 0.37 5 SC non-workers 0.18 0.26 0.48 6 Difference in percent non-workers 0.03 0.18 0.88 N = 2797 and number of matched pairs is 480 7 Robustness checks While I have only presented variations on two outcome variables in this paper, the difference in development trajectories is in fact statistically insignificant from 0 on all the variables in the census data, including the number of people involved in farming and industry, the differences between men and women and the changes in population size. The zero-findings are also robust to looking at the actual values in 2001 instead of the differences in differences, but I chose to focus on the first differences in order to control for any remaining unbalance in the outcome variables. There are many ways of checking the robustness of these findings. In this section I will briefly present two such robustness checks. First, since all the political boundaries were reorganized in the early 1970s, there is no way of exactly checking whether a constituency was reserved or general the last time around. However, it is possible to make an approximation. Comparing the delimitation reports of 1967 and 1976, it is possible to create fuzzy matches of constituencies. A fuzzy match is the constituency in 1967 containing the largest part of the 1976 constituency. Using the fuzzy matches, I reran the matching models comparing only constituencies that used to be general and then became general or reserved. In this case the number of matched pairs is reduced to 108 because of missing values and because of reducing the samples to former general constituencies. The balance of observables is quite similar to the one reported earlier, 23

and as can be seen in table 4, none of the outcome variables come out substantively or statistically significant. Table 4: Matching estimates of the difference in literacy rates and percent non-workers in general and reserved constituencies, looking only at constituencies that used to be general according to fuzzy matching across delimitations Variable Estimate St. error P-value 1 Non-SC literacy -0.29 0.61 0.64 2 SC-literacy -0.4 0.48 0.41 3 Difference in percent literacy 0.11 0.57 0.84 4 Non-SC non-workers 0.12 0.57 0.83 5 SC non-workers 0.22 0.65 0.73 6 Difference in percent non-workers -0.1 0.4 0.8 N = 2033 and number of matched pairs is 108 Finally, an important robustness check is to look at variation by states. In this case, there is more variation in the estimates, but the sample sizes are also smaller rendering almost all of the estimates insignificant. In fact, there is only one estimate across the states and variables that is statistically significant at a 0.05 level. According to the estimate, the growth in literacy rate among non-scs in Bihar was 1.2 percentage points higher in reserved constituencies than in general constituencies. Appendix A shows the state-wise estimates for the first differences of literacy rates and percent non-workers. 8 Discussion The empirical section shows that there is no substantively large or statistically significant difference in the rate of growth in literacy or in the percent non-workers among SCs or non-scs in general and reserved constituencies. On the basis of this finding we can reject the hypothesis that SC politicians particularly work to promote development for SCs in their constituencies. We can also reject the hypothesis that SC politicians are useless and have a negative impact on development. The hypothesis that is strengthened is that SC politicians act in a similar way to other politicians 24

and follow party line. Since SC politicians run for election in reserved constituencies with joint electorates, and since political parties dominated by non-scs are the ones choosing who to field as political candidates, SC politicians cannot rely on votes only from their own community in order to win elections. In line with the median voter theorem, and the fears of among others Dr. Ambedkar, SC politicians might be tied down by electoral incentives in exactly the same way as other politicians. Stories about biased transfers and preferential treatment might be based on politicians attempts to please different parts of their electorates, but might not be large enough scale to result in structural changes. There are some alternative explanations for this findings. First, it is possible that MLAs impact some types of development, but simply do not make an impact on the literacy rate in their constituencies. One should think that 30 years of targeted benefits, biased transfers and special attention to a specific community should become visible in something like the literacy rate. Also, 30 years of having a less-connected, less-skilled politician should become visible in an aggregate measure. It is possible, however, that a more direct measure of development is needed in order to capture this bias. Other papers related to my dissertation work will look at the public goods such as roads, water and electricity provisions, in order to explore this topic further. Second, SC politicians might try to benefit their own group, but have little power to do so. During my interviews with politicians I often heard complaints that the bureaucracy only listened to the MLAs from the party in power. If this is the case we might see that places that have had MLAs belonging to the ruling party will have developed more than places that have had MLAs in opposition. In that case SC politicians are like other politicians in that they are powerless once they are in opposition. These ideas are explored in a forthcoming paper by Jensenius and Bhavnani. Third, it is quite possible that some SC politicians would like to work for their community, while others do not. The power of political parties in India is strong, and it is possible that parties have chosen to field only candidates that follow the party 25

line and act in a similar way to other politicians. Some of the evidence in interviews suggests this. If this is the case, however, it goes back to the main hypothesis in this paper, that something in the institutional design prevents a substantive representation of SCs. Finally, it is possible that SC politicians have been important in influencing policy choices overall that benefit the whole state they live in, but that these benefits are not visible at the constituency level. This, however, is hard to measure, and goes against all the anecdotal evidence of the importance of MLAs as fixers at the local level. 9 Conclusion Since independence, India has had the world s largest quota system, with 16% of all the political seats in the Lok Sabha (lower house) and in state legislative assemblies reserved for Schedule Castes (SCs) and 7% reserved for Scheduled Tribes (STs). It is commonly heard that quotes as meant to help uplift the SC community. This was the stated purpose of implementing the quotas during the Constituent Assembly debates and it is an oft repeated argument among bureaucrats and politicians in India. Activists, bureaucrats and opinion makers seem to assume that SC politicians will try to enhance development for their own community by targeting benefits and pushing for advantageous transfers. Recent scholarship about quotas in India has also suggested that politicians tend to work for their own group and that development patterns change when quotas are put in place. At the same time, it is often heard among politicians, bureaucrats and activists that SC politicians are less talented, less educated and less connected than other politicians, and that there therefore is less overall development in places with reservations. In this paper, I have looked at two indicators for development, the literacy rate and the percent non-workers, among non-scs and SCs, in reserved and general constituencies. Although politicians cannot directly influence literacy rate, 30 years worth 26

of targeted transfers, hiring or encouragement could become visible in aggregate development indicators. If this bias indeed existed, we should expect to see a higher rate of growth in the literacy rate of SCs living in reserved constituencies than of those living in general constituencies and a lower percent non-workers among SCs in reserved constituencies. Similarly, if SC politicians generally were less skilled than other politicians, we should expect to find an overall lower rate of growth in literacy in places with reservations. Neither of these claims found support in the empirical data. In fact, once controlling for the bias of a different starting point, there was no substantive large or statistically significant different in the rate of growth in literacy for SCs or non-scs in reserved and general constituencies. This finding supports the hypothesis that the design of the quota system is such that SC politicians are incentivized to work for everyone in the constituency. Finding no correlation between reservations and development does not mean that quotas do not serve an important role in guaranteeing the political presence of a marginalized group, or that it has not influenced social discrimination. However, the general expectation of the quotas in India seem to be that they should effect development. Thus, there seems to be a disjuncture between the incentive structures created by the design of the quota system and the expectation of the what the system is to achieve. 27