World Development Vol. 38, No. 2, pp. 195–204, 2010 Ó 2009 Elsevier Ltd. All rights reserved 0305-750X/$ - see front matter www.elsevier.com/locate/worlddev doi:10.1016/j.worlddev.2009.10.013 Political Market Characteristics and the Provision of Educational Infrastructure in North India BENJAMIN CROST University of California, Berkeley, USA and UMA S. KAMBHAMPATI * University of Reading, UK Summary. — In this paper, we are concerned with the provision of schools in rural North India, particularly with whether such provision is determined by the demographic and economic characteristics of the region or whether local democracy also plays a role. We find that the probability that a governing party loses an election has a positive effect on the provision of schooling infrastructure, while the margin of victory of the governing party has a negative effect. Political reservation for members of the Scheduled Castes (SCs) has a positive effect on schooling infrastructure in villages with a large SC population, but a negative effect overall. Ó 2009 Elsevier Ltd. All rights reserved. Key words — schooling, India, public goods, political economy, political competition, caste cess in increasing literacy (male literacy increased by 11.8 percentage points and female literacy by 15 percentage points since the beginning of the 1990s (Borooah & Iyer, 2005)), India’s performance relative to other countries, particularly China, has been very poor. Thus, the adult literacy rate in India was only 61% in 2003, as opposed to almost 91% in China and virtually 100% in the UK. The difference between the male and female literacy rate was also highest in India, with 73.4% for males and 47.8% for females, while it was 95.1% for males and 86.5% for females in China in 2003. But a closer look at the data shows that educational attainment varies considerably across states within the country (see Table 2). For example, literacy rates were 49.2% in Bihar, 58.1% in Uttar Pradesh, but 90.9% in Kerala in 2001. To what extent are these correlated with the socio-economic conditions of these two states? Table 1 indicates that these two states are outliers in terms of many of these characteristics. Both have very low per capita state domestic product levels, with Bihar at Rs. 5,108 in 2000–1 and UP at Rs. 9,721. This compares unfavorably with a per capita income of Rs. 16,373 in Andhra Pradesh, Rs. 16,072 in West Bengal, and Rs. 25,048 in Punjab (a very prosperous state). Similarly, the proportion of population living below the poverty line in 1999–2000 was 42.6 in Bihar and 31.15 in UP compared to an all India average of 26. The low level of economic development is also reflected in the states’ per capita consumption of electricity (Bihar (146.7) and UP (179.1) compared to an all India average of 364.5) and their credit–deposit ratios (23.2 in Bihar, 30.9 in UP and an average of 56.0 for India overall) in 1999–2000. Bihar and UP have also lagged with respect to investments in education and health as compared to other states in India. Table 3 reveals that while Kerala has fewer primary schools per 1,000 people, its primary schools are accessible to a higher proportion of its population. Its schools also seem to be better 1. INTRODUCTION The positive effects of education on economic and social development have been extensively documented and include increased productivity, accelerated adoption of new technologies, reduced child and adult mortality, increased gender equality, and increased accountability of governance institutions. Since the returns to education and health are at least partly public in nature there is an economic rationale for public provision or subsidization of education to offset private under-investment. Indeed, in many developing countries the state plays a large role in the provision of education. If properly allocated, public expenditures could help to overcome market failures in these sectors but evidence seems to indicate that such market failures are simply reinforced by political market imperfections (Keefer & Khemani, 2005). In this paper, we are concerned with the provision of rural schools in two states in Northern India—Bihar and Uttar Pradesh. Our main question is the extent to which the allocation of schools in these states is related to political factors as opposed to the demographic and economic characteristics of the region. Is school allocation merely an outcome of the size of the village and its prosperity or does its ethnic composition also play a role? What role does local democracy (in terms of voter involvement, reservation of seats for minority communities, etc.) play in this allocation? In trying to answer these questions, the paper contributes to the growing political economy literature on the public provision of goods and services and on elite capture. The analysis is at village level and therefore allows us to consider the issues at a more disaggregated level than previous studies (e.g., Banerjee & Somanathan, 2007; Betancourt & Gleason, 2000). In India the public sector has made large investments in education in the past decade. Thus, public expenditure on education in India as a proportion of total government expenditure was 12.7% in 2002, which compares well with 12.8% in China in 1990 and 11.5% in the UK. While there has been some suc- * 195 Final revision accepted: December 3, 2008. 196 WORLD DEVELOPMENT Table 1. Socio-economic indicators for a selection of Indian states Source: (1) Per capita electricity consumption: Government of India (2002c). (2) For urban population and population below the poverty line: Government of India (2002b). (3) Reserve Bank of India (2001). (4) Per capita net state domestic product: Government of India, 2002a. States Ratio of urban Population (%, 1999–2000) Per capita consumption of electricity (1999–2000) Per Capita net state domestic product at current prices (2000–01, Rs) Credit–deposit ratio (utilization) 2000 % Living below poverty line (1999–2000) 27.1 10.5 37.4 26.0 42.4 33.9 20.8 28.0 27.8 434.0 146.7 840.9 315.0 571.6 924.1 179.1 206.9 364.5 16,373 5,108 19,228 19,463 23,726 25,048 9,721 16,072 NA 65.5 23.2 53.6 41.7 83.4 40.9 30.9 44.9 56.0 15.77 42.6 14.07 12.72 25.02 6.16 31.15 27.02 26.1 Andhra Pradesh Bihar Gujarat Kerala Maharastra Punjab Uttar Pradesh West Bengal India Table 2. Variation in educational attainment by state—2001. Source: Government of India (2001). State Uttar Pradesh Bihar Kerala Andhra Pradesh Gujarat West Bengal Literacy (%) Literacy SC (%) Literacy ST (%) Age specific enrollment rates (6–11 years) (%) Age specific enrollment rates (11–14 years) (%) Male Female All Male Female All Male Female All Male Female All Male Female All 70.9 62.2 94.2 70.9 76.5 77.6 43.9 35.2 87.9 51.2 55.6 60.2 58.1 49.2 90.9 61.1 66.4 69.2 59.0 38.7 87.2 60.6 79.2 69.1 28.3 14.1 76.4 39.8 51.2 44.5 44.5 26.9 81.7 50.3 65.6 57.1 46.7 37.6 70.2 46.1 58.1 56.6 18.3 13.3 57.3 24.5 34.6 27.9 33.0 25.9 65.7 35.4 46.5 42.4 79.9 98.2 87.7 105.2 137.9 110.9 50.3 60.5 86.5 102.9 113.6 103.3 65.7 79.9 87.1 104.1 126.2 107.2 48.1 40.7 99.8 53.5 73.0 59.7 25.2 20.7 94.8 44.3 59.5 44.4 37.4 31.3 97.3 49.0 66.5 52.2 Table 3. Inputs into education by state Source: UNDP (2001). Variable All India average Uttar Pradesh Bihar Kerala 85.13 85.5 45 29 5.75 66.8 73.15 58 55 4.44 87.44 77.51 52 37 4.88 85.81 84.07 32 30 2.19 Population with primary schools (up to 0.5 kms) in 1978 Population with primary schools (up to 0.5 kms) in 1993 Pupil–teacher ratio in primary school (1992–93) Pupil–teacher ratio in secondary school (1992–93) Number of primary schools per 1,000 population (1992–93) equipped, with pupil–teacher ratios of 32 as opposed to 50 or above in Bihar and UP. Dreze and Gazdar (1996) found that almost two-thirds of the teachers employed in the sample schools that they visited in the state of Uttar Pradesh were absent at the time of the visits. However, problems of this kind are not confined to Bihar and UP. Similar patterns have been identified at the all India level by Chaudhury, Hammer, Kremer, Muralidharan, and Halsey Rogers (2006) who found that one quarter of government primary school teachers in India were absent from school and only about half of the teachers present were actually teaching when enumerators arrived at the schools. Dreze and Sen (1996) argue that the contrast between UP and Kerala is caused by different political incentives with regard to the provision of social services in the two states. Keefer and Khemani (2005), extend this point, arguing that while the quality of public services is hard to target, it is clear that the quality and effectiveness of public spending are higher in Kerala. They argue that since the formal political and legal institutions in the two states are the same, the differences must arise from the “dynamics of political competition (i.e., the availability of information to voters, the extent of social polarization, and the credibility of political promises) rather than in differences in the political institutions themselves” (p. 18). Thus, in UP, the Congress Party dominated electoral politics and did not confront vigorous competition from other credible and well-organized alternatives. While some caste and ethnicity-based political parties did emerge—like the Bharatiya Janata Party, the Bahujan Samaj Party, and the Samajwadi Party—these parties competed on explicitly clientelist platforms and so were not judged even by the electorate on their performance with respect to the provision of broad public services like schools and health services. In this paper, we empirically analyze whether such political factors have influenced the provision of schools and schooling infrastructure in UP and Bihar. We argue that, despite these states being at one end of the socio-economic spectrum, many of the problems they face are similar in kind (if not degree) to those elsewhere in India. Therefore, while the results in the paper pertain particularly to two states, they are relevant to many other less developed states and regions in India. 2. LITERATURE The question of how public goods are allocated by the public sector in developing countries has recently received increasing attention (see Keefer and Khemani (2005) for a review). POLITICAL MARKET CHARACTERISTICS AND THE PROVISION OF EDUCATIONAL INFRASTRUCTURE IN NORTH INDIA 197 Much of the literature has concentrated on why public provision in developing countries is rarely as good as might be expected. Is it because too much centralization of decisionmaking and resource allocation results in governments being unaware of the needs of citizens at a local level? Such centralization of service delivery might also result in cost padding, service diversion, limited responsiveness to local needs, limited access, and higher prices (Bardhan & Mookherjee, 2006). On the other hand, the decentralization of resource allocation decisions also has its problems, in particular, the “greater capture of these programs by local elites” (Bardhan & Mookherjee, 2000). Thus, with limited political contestability of local elections, leaders may be susceptible to capture by special interest groups, slacken effort to improve public services, or may be incompetent, without facing any risk of losing their positions. In this case, “accountability, efficiency, and equity in service delivery may worsen under decentralization” (p. 102). The literature is generally divided over this issue. While Lieten (1996) and Matthew and Nayak (1996) find that some local village groups in India were subverted by local elites, Galasso and Ravallion (2005), find that intra-district targeting failures in schooling programs in Bangladesh are less severe than inter-district failures, leading them to conclude that local government is more redistributive than central government. Bardhan and Mookherjee (2000) argue that special interest groups find it harder to co-ordinate at the national level, which decreases their influence and makes centralized decision-making more effective. They also argue that the level of homogeneity of party loyalties, differences in intra-district inequality, and levels of political awareness may shape the effect of decentralization. Keefer and Khemani (2005) argue that broad public services are most vulnerable to political market imperfections (which include lack of information among voters, social fragmentation of voters, and lack of credibility of political promises), since it is hard for citizens to evaluate the quality of services or identify those responsible for improvements. It is therefore easier for politicians to promise narrowly targeted public goods where their responsibility is more visible and verifiable. In our context, this would imply that politicians would promise to build new schools rather than to improve the quality of existing schools. Keefer and Khemani (2005) also argue that social services suffer when societies are ethnically divided because people are more likely to vote for candidates that they identify with, regardless of their performance record. Probably fueled by the availability of good data and the large regional variation in socio-economic and political conditions, India has become a favorite subject of study in this area of research. Besley and Burgess (2002) have examined the effect of newspaper circulation on the responsiveness of Indian states to shocks in food production. Chattopadhyay and Duflo (2004) and Pande (2003) analyze the effect of mandated political representation on investment in publicly provided goods at the state and village levels, respectively. Pande (2003) finds that mandated reservation in state legislatures increased public sector job quotas for Scheduled Castes (SCs) but decreased resource allocation to education. In addition, Besley, Pande, and Rao (2007) in a study of public resource allocation in South India find that voters are aware of political discretion in policy making and use their electoral clout to ensure that resources are directed toward themselves. They also find that low caste households are significantly more likely to be allocated the Below Poverty Line (BPL) cards when the village Pradhan (or Head) shares their group identity. They therefore argue that politicians’ group identity and self-inter- est are important in the allocation of public resources. Banerjee and Somanathan (2007) and Betancourt and Gleason (2000) have examined the differences in allocation of several public goods (among them, the number of teachers and schools) between Indian districts. Betancourt and Gleason (2000) find that districts with a high proportion of Muslims have significantly fewer teachers and districts with a higher Scheduled Caste and Muslim population receive lower public inputs into education and health in India. Banerjee and Somanathan (2007) find inter alia that fewer villages have middle and high schools in districts that are more fragmented along the lines of caste and in districts with a high proportion of Muslims and Scheduled Tribes. While Betancourt and Gleason (2000) and Banerjee and Somanathan (2007) use the district as their level of analysis, much of the literature suggests the incentives for, or against, school provision work at a highly localized level. It is therefore possible that aggregating to the district level hides distributional aspects that may be important for our understanding of the issue. Consequently, we take a micro-approach and analyze the allocation of schools across villages in the two North Indian states of Uttar Pradesh and Bihar. In particular, we want to understand which village characteristics determine the presence of schools and teachers, and the quality of schooling infrastructure. 3. INSTITUTIONAL BACKGROUND The Indian education system is divided into pre-primary, primary, middle, and secondary schools. Within this, primary schools include children between 6 and 11 years, while middle schools include children between 11 and 14 years (though there is some variability across states regarding whether 11year olds are in primary or middle schools). The number of primary and upper primary schools in India increased from 0.223 million in 1950–51 to 0.775 million in 1996–97. Until 1976, education was largely the responsibility of the states, with the Central Government only specifying standards and co-ordinating technical and higher educations. However, since 1976, education has become the joint responsibility of the states and the center. The 8th five year plan put forward the target of universalizing elementary education in India along three lines—universal access, universal retention, and universal achievement. In its National Policy on Education (first put forward in 1986 and revised in 1992), the Government of India committed itself to increasing spending on education to 6% of GDP by 2000. Part of this expenditure was meant to be on improving access to schools, on improving the quality of schools, and on improving enrollment in schools. This culminated in the launch of the District Primary Education Program (DPEP) in 1994 by the Government of India in collaboration with the World Bank, European Commission, the Government of Netherlands, UNICEF, and the Department for International Development (UK). These external donors provided 85% of funding for this program, while 15% came from the state governments. Phase I concentrated on seven states—Assam, Haryana, Karnataka, Kerala, Maharashtra, Tamil Nadu, and Madhya Pradesh. It therefore excluded both our sample states, which were included in later phases in the DPEP (UP in Phase II and Bihar in Phase III started in 1997). Having said this, the precursor to the DPEP in Bihar was the Bihar Education Project (BEP), which was set up in 1992 in seven districts. There were also a number of other 198 WORLD DEVELOPMENT initiatives and schemes at both the state and the central levels, including the National Literacy Campaign launched in 1988, Operation Blackboard (1986), the Mid-day Meal Scheme (1982), and the Shiksha Karmi Program (1987). These would have had some effect on the supply side outcomes that we see in the districts that we are analyzing in this paper. They are however very difficult to identify at the village level in our dataset because there is no information about whether the specific village benefits under these schemes or not. While five states—Andhra Pradesh, Bihar, UP, Maharashtra, and Rajasthan—were especially backward on the educational front in 1991, Bihar became the worst off state in India in educational terms by 2001 (Schmid, 2007). What determines the allocation of schools across villages in India? It seems to be a combination of the ability of the village to lobby for resources and the district and state governments’ incentives to allocate resources to it. Thus, we would expect village characteristics as well as district level political institutions to play a role. Betancourt and Gleason (2000) identify three main stylized facts in this decision-making framework. First, Indian states are the main decision makers in the allocation of resources to health and education at a local level (Thakur, 1995, Chapter 3), since they control both the finances and the administrative bureaucracies devoted to these activities. Second, elected members of the state legislature play key roles in the allocation decisions. Third, the district is the next most important administrative level after the state, with district officers who have jurisdiction over the district. While the broad decisions over resource allocation are made by the states, they are implemented by the districts, which also monitor the provision of public services. As Betancourt and Gleason (2000) conclude, “local outcomes with respect to the availability of health and education inputs at the district level are determined by decision makers in each state, who are politicians elected by constituencies to the state legislatures, and they do so through the mediation of state bureaucracies and subject to constraints imposed by the center,” (p. 2171). The state distributes the education budget across districts in line with the number of schools, the number of teachers, enrollment rates, and other such factors. In addition, there are off-budget items like the mid-day meal scheme for which almost all funding comes from the central government based on the number of enrollments. The state government only provides the premises in which the scheme operates. pends on the individuals’ prospects in the labor market and this might reflect their returns to education (the evidence regarding differential returns to education by social group is mixed, see Das & Dutta, 2008; Madeshwaran & Attewell, 2007; Unni, 2001). The elasticity of income with respect to the public goods is denoted by l (>0, <1), and f denotes the level of the public goods provided in the village. 2 Following Bardhan and Mookherjee (2006) we assume that all individuals have concave and homothetic utility functions over their incomes uc ¼ y 1q c : 1q ð2Þ We further assume that the officials in charge of allocating the public goods do not maximize the aggregate utility of the villages, but rather assign them different welfare weights xv. These weights are determined by the characteristics of the village, particularly by the political influence of its inhabitants. In our sample, these weights may depend upon the proportion of upper castes, Scheduled Castes (SCs), or Muslims in the village, or on the wealth of its inhabitants. For different theories of how local elites can “capture” the process of political decision-making see for instance Grossman and Helpman (1996) or Bardhan and Mookherjee (2000). The demographic weights of the classes within P a village are denoted by bcv, which we normalize so that cv bcv ¼ 1. If we assume that, once a public good is present in the village, nobody can beP excluded from its use, the whole village receives utility uv ¼ c bcv ucv from the public goods. This need not, however, be true. There are a number of cultural and social factors that may, for instance, prevent Scheduled Caste children from attending schools to which upper caste children go. In some villages, they may be physically prevented from accessing public goods. In this case the village utility is just the sum of the weighted utilities of all classes that benefit from the school. Consequently, utility received from schools will be smaller, the larger the fraction of excluded classes in the village. Given the village utility levels, officials choose the allocation to maximize " # " # X X X X X ðhv Ac fvl Þ1q xv uv ¼ xv bcv ucv ¼ xv bcv 1q v v cv v cv X fv ¼ f : subject to the constraint v 4. THEORETICAL MODEL We analyze the provision of public services in rural India by building on a model proposed by Bardhan and Mookherjee (2006), which considers the allocation of direct government transfers to villages and individuals. We begin by dividing the population into classes, which differ in their potential to benefit from the public goods; schooling in our case. 1 The classes can be interpreted as representing ethnic, religious, or economic sub-groups of the population. This categorization is not meant to be exclusive; the same household could fall into several classes, for example, it could be Scheduled Caste as well as landless. The income generated by a member of class c is y c ¼ hv Ac fvl ; ð1Þ where hv is the village-specific productivity of the public goods, in our case, for example, the productivity of human capital manifest in the difference between skilled and unskilled wages. Ac is the class-specific productivity of the public goods and de- The relative allocations of the public goods to two villages, a and b, can then be described as 1 0P 11lð1qÞ 1q 1 1lð1qÞ 1lð1qÞ bca A1q c fa xa ha B ca C ¼ : ð3Þ @P A fb xb hb bcb A1q c cb We further assume that the welfare weight of the village is itself a function of the demographic composition of the village (b), the demographic composition of the district (B), and the political factors at both the village and district levels (P) xv ¼ xðb; B; P Þ: As pointed out by Bardhan and Mookherjee (2006), it is impossible to directly infer the relative welfare weights of the villages from their relative allocations of the public goods. For example, a low allocation of the goods can arise due either to the lower welfare weight assigned to the villagers (i.e., their political disempowerment), or to their lower returns to the POLITICAL MARKET CHARACTERISTICS AND THE PROVISION OF EDUCATIONAL INFRASTRUCTURE IN NORTH INDIA 199 public goods (and, as a consequence, their lower preference for it). The evidence that the members of disadvantaged groups (Scheduled Castes, Muslims, and women (see Kingdon, 1998) have lower returns to education and tend to prefer direct income transfers over public goods (see Pande, 2003) shows that this is indeed a problem in the Indian context. Nevertheless, it is possible to make some empirical inference on how the political process allocates public goods, by analyzing the welfare weights in detail. Particularly, we are interested v v in the signs of the partial cross-derivatives, @b@x and @b@x , cv @Bv cv @P v which describe how changes in district level demographic composition and political variables affect the political welfare weights of different classes. The north of India (and particularly Bihar) is well known for the high salience of caste in the political process. There are influential lobby groups, which represent caste interests and try to maximize government transfers to their members. In such a setting, it is likely that the political influence of a caste in a particular district rises with the number of its members. We would therefore expect the welfare weights of a particular caste to be higher in a district in which it represents a larger share of the population v (i.e., @b@x > 0). We would also expect the welfare weight of cv @Bv Scheduled Castes to increase through political reservation of v parliamentary seats for their members (i.e., @b@x > 0; see cv @P v Pande (2003) for a possible mechanism). 5. EMPIRICAL ESTIMATION As indicated earlier, we are primarily concerned with the allocation of public schools across villages in India. We apply the theoretical framework outlined above using the data from the village section of the Uttar Pradesh and Bihar Survey of Living Conditions, which was conducted by the World Bank in 1997–98. The survey collected detailed information on 120 villages located in 25 districts of the two states. For our analysis we counted government and government-aided schools as public. The villages in the sample have a total of 124 public primary schools. Twenty-seven villages have no public primary school and 93 have one or more; the maximum being five in one village. Twenty-seven of the villages have a public middle school and the other 93 have none. Unfortunately, for a few villages in our sample, the data on variables of interest is missing. The unskilled wage is missing for 1 village, the skilled wage for 3 villages, and the proportion of households with significant off-farm income is missing for 4 villages. If we were to use only complete observations, we would have to exclude 7 villages. In order to salvage the incomplete observations we substituted the district level averages of the variables for the missing values. (a) Dependent variables We analyze the factors influencing the number of public primary and middle schools in a village, as well as some characteristics of primary schools, 3 such as the number of teachers, the presence of usable blackboards, and the physical structure of the school. The determinants of the number of public primary and middle schools and the number of teachers within the primary schools are analyzed by estimating Tobit models. On the other hand, since the schooling infrastructure variables (the floor material, physical structure, and the presence of usable blackboards) are not continuous, we analyze the factors influencing them by estimating ordered Probit models. For these models, our dependent variables take on values between 0 and 2. For floor material we assign a value of 2 for schools with cement, stone, or tile floors, 1 for schools with mud or brick floors, and 0 if there is no primary school present in the village. For physical structure we assign a value of 2 for Pucka schools, 1 for other schools, and 0 if there is no school in the village. For blackboards we assign a value of 2 for schools with usable blackboards, 1 for schools without, and, again, 0 if there is no school present. The definitions of the dependent variables are summarized in Table 4. (b) Factors influencing school infrastructure In order to keep the empirical model close to the theoretical one, we include variables that capture its relevant terms: x (the political welfare weight of the village), h (the productivity of P 1q human capital in the village as a whole), and ca bca Aca (the part of the productivity of human capital influenced by the demographic composition of the village). To achieve this, we consider demographic, economic, and political variables, measured at the village and district levels and estimate their effect on the provision of education-related public goods (which we refer to as schooling infrastructure). It is clear from the discussions in the theoretical section that some of the variables may affect schooling infrastructure through more than one mechanism. This makes the interpretation of their coefficients somewhat ambiguous and we discuss possible mechanisms in this section. (c) Demographic variables We begin by including the number of households in the village to capture the size of each village. We expect that larger villages are likely to have better schooling infrastructure. This could be both because large villages are likely to have greater demand for such infrastructure and because they are likely to have greater power to lobby for resources and therefore have a higher welfare weight (x) in the objective function of policy Table 4. Definition of dependent variables Variable Definition Number of public primary schools Self-explanatory Number of public middle schools Self-explanatory Number of teachers in public Takes the value 0 if there is no public primary school, otherwise primary school the number of teachers in the school Blackboards present in pps Takes the value 0 if there is no public primary school in the village, 1 if there is one without usable blackboards, and 2 if there is one with usable blackboards Physical structure of pps Takes the value 0 if there is no public primary school in the village, 1 if its physical structure is Katcha or Semi-Pucka, and 2 if it is Pucka Floor material in pps Takes the value 0 if there is no public primary school in the village, 1 if its floor material is mud or brick, and 2 if it is cement, stone, or tile Estimation model Tobit Tobit Tobit Ordered probit Ordered probit Ordered probit 200 WORLD DEVELOPMENT makers. To capture possible non-linearities in the effect of village size, we also include a quadratic term in this variable. In addition to this broad size variable, we include a number of variables that capture the characteristics of the village population and its ability to lobby for resources. In this context, we include the fraction of households in the village, who belong to the upper castes, which we expect will have a positive effect on schooling infrastructure in the village. We also include the proportion of Scheduled Caste households in the village, which we expect will have a negative effect, possibly because these households tend not to have significant power to lobby officials at the district level. To the extent that these variables capture the lobbying capacity of the village, they can also be seen as influencing x. It is also possible that Scheduled Caste households prefer direct monetary transfers to public goods (Pande, 2003). In both Bihar and UP, the Scheduled Castes and Tribes have begun to dominate the political scene through the influence of certain explicitly caste-based political parties like the Bahujan Samaj Party and the Samajwadi Party. In this context, we may expect that the Scheduled Castes use political power to direct public resources toward their members. To allow for this, we include an interaction term between the number of SC households in the village and a variable indicating whether the constituency is reserved for SC candidates. While we cannot be sure whether the village level Pfraction of SC affects schooling infrastructure through x or ca bca A1q ca , it is likely that the interaction with reservation works only through its effect on the welfare weight, x. 4 Since we also expect ethnic groups to have more political power in areas where they are present in larger numbers, we include an interaction between the village and district level fractions of SC. We expect this to have a positive effect if the Scheduled Castes can translate greater numbers at the district level into political power and use this power to direct resources toward their members. In a similar vein, we include the proportion of Muslim households in the village and an interaction between village and district level Muslim populations. We hypothesize that school infrastructure will be poorer in majority Muslim villages, but that villages with a large Muslim population will fare better in districts with a large Muslim population. To capture the possible effect of ethnic diversity, we include a caste fractionalization index, which is constructed using upper caste, middle caste, SC, backward agricultural and other castes, and Muslims as its basis. It is calculated as P i 2 1 ni¼1 NNtotal ; where Ni is the number of households belonging to caste group i and Ntotal is the number of households in the village. Since previous studies have found that fractionalized communities are less able to lobby effectively for public goods (Alesina, Baqir, & Easterly, 1999), we hypothesize that caste fractionalization has a negative effect on the political welfare weight (x) and will therefore lead to lower allocations of schooling infrastructure. In addition to these variables, we also include the population density of the district as a proxy for the degree of urbanization of the district. (d) Economic variables In addition to the demographic variables discussed above, we include a number of economic variables, reflecting the potential prosperity of the households in these villages. We include the average skilled and unskilled wages in the villages (the skilled wage is reported in the dataset; we define the unskilled wage as the average of wages in agriculture and construction). These variables are likely to influence schooling infrastructure through their effect on the returns to education. A high unskilled wage (and a low skilled wage) means that the returns to human capital are low in the village and the gains from keeping children out of school and letting them work are high. Of course, it is also possible that high wages positively affect the welfare weight of the village, since wealthier villages may be better connected politically. As additional proxies for village prosperity, we include the proportion of households with a significant source of off-farm income, as well as the proportion of households, which have members working as rural laborers. Households with a high fraction of non-farm income may be wealthier and politically better connected, or they may have a higher return to human capital (if the return is higher outside agriculture), while rural laborers are likely to be less wealthy. We also include the proportion of households that are completely landless. We expect that schooling infrastructure will improve as households have more access to off-farm income, and decrease with the proportion of rural laborers and landless households. Finally, we include two variables that help to control for other specific village level factors and may influence both the welfare weight (x) and the return to education (h). The first is an indicator of whether the majority of land in the village is irrigated. Irrigated land is likely to be more productive and therefore to sustain more prosperous households. We also include the ratio of irrigated land to total agricultural land in the district to indicate whether the district in which the village is located is prosperous or not. Second, we include an indicator for villages that have access to electricity, which may also proxy for the quality of infrastructure in general. In addition, both irrigation and electrification may reflect an organizational ability within the village, which enables it to obtain such facilities from the landlords and/or the state. (e) Political variables The political variables we include relate to the extent of political competition in the district (as reflected in the margin of victory of the incumbent party, the extent of party fractionalization, and the probability that an incumbent party loses an election); political awareness of inhabitants of the village (as reflected in turnout at elections) and empowerment of minority communities (as reflected in reservation of seats for such communities in assembly elections). The political variables are based on the results of state elections. For Bihar, we use the elections of 1977, 1980, 1985, 1990, and 1995; for UP the ones of 1977, 1980, 1985, 1989, 1991, 1993, and 1996. While the variables are available at the level of the assembly constituency (AC), the rest of our variables are only available at administrative district level. We therefore matched assembly constituencies to parliamentary constituencies (PC) using a matching scheme available from the Election Commission and manually mapped the parliamentary constituencies on to the districts. For each district, we take the weighted averages over all parliamentary constituencies and all elections in the relevant period. We include five political variables, reflecting political reservation, electoral turnout, the margin of victory, party turnover, and party fractionalization. For political reservation, we include a dummy variable that equals 1 if the constituency is reserved for a candidate from the Scheduled Castes. As indicated earlier, this variable is likely to have a negative impact on a village within it unless the village happens to be largely of Scheduled Caste people. Two effects are possible. Pande (2003) finds that in districts where seats are reserved for SC candidates, less is spent on education. If this holds in our case, we would expect schooling infrastructure to be poorer in POLITICAL MARKET CHARACTERISTICS AND THE PROVISION OF EDUCATIONAL INFRASTRUCTURE IN NORTH INDIA 201 districts where political reservation exists. However, as pointed out before, it is possible that a predominantly SC village will do better in a district in which seats are reserved for SC candidates. We attempt to capture this effect by including an interaction between political reservation and the proportion of Scheduled Caste households in the village. We also include electoral turnout, which is the number of voters as a fraction of the electorate. This indicates the political activism of the electorate and may therefore indicate its ability and willingness to lobby for public goods. It is likely that regions with higher levels of political activism are given greater weight by policy makers as they attempt to keep voters content. We include three variables reflecting political competition in the district. First, the margin of victory is the difference between the winner and the runner-up as a proportion of the voters. This reflects the extent of political competition in the district and therefore also the amount of pressure on the existing candidates to keep voters happy. There is more pressure on incumbents with smaller margins of victory and they are therefore more likely to provide public services liberally. This is particularly true for services that can be easily traced back to the candidate like the number of schools in a village. It is less likely to be true for school infrastructure variables, for instance, which are less easy to quantify and harder to trace back to the politician (see Keefer & Khemani, 2005). Second, we include party turnover, that is, the probability that an incumbent party loses an election. Again, as in the case of the margin of victory, this measures the pressure on the incumbent party to keep voters in the constituency content. Third, we include a party fractionalization index, which is calculated in the same way as the caste fractionalization index but here N is the number of votes cast for each political party. This index measures fractionalization among parties in the constituency with respect to the fraction of votes cast for each party. Again, it is a measure of competition amongst the parties. A look at the summary statistics in Table 5 shows that 21.3% of constituencies in our sample are reserved for the Scheduled Castes, which is only marginally larger than the proportion of SCs in the population as a whole (20.3%). The average turnout in the sample constituencies is 56.8%, which is roughly average for India and the average margin of victory is 12.0%. With 52.5%, the probability that an incumbent party loses the election appears quite high, but is not unusual for India. In fact, incumbents tend to be at a disadvantage in India and if an incumbent runs for re-election his/her average probability of winning is only 50% (Uppal, 2009). Overall therefore, the level of political competition in the districts in our sample is relatively strong. 6. RESULTS Table 6 presents the results relating to the number of primary and middle schools, as well as the number of teachers in a village. Table 7 presents the results relating to the provision of schooling infrastructure in a village. In what follows we will discuss these results in more detail. Beginning with the village level demographic variables, we find that the size of the village is one of the most consistently significant variables in our models. Thus, larger villages tend to have a larger number of schools and teachers and better school infrastructure. This is a heartening result and indicates that the allocation mechanism seems to be working as it is meant to, directing schooling resources to regions where there is likely to be greater demand for these resources. Table 5. Summary statistics of the explanatory variables Village level variables Public primary schools Private primary schools Public middle schools Private middle schools Village size (households) Caste fractionalization index Upper caste (fraction of households in village) Scheduled Caste (fraction of households in village) Muslim (fraction of households in village) Skilled wage Unskilled wage Access to off-farm income (fraction of households in village) Rural laborers (fraction of households in village) Landless (fraction of households in village) Village irrigated Village access to electricity District level variables Irrigated land (district level fraction of agricultural land) Population density Scheduled Caste population (district level fraction) Muslim population (district level fraction) Political variables Political reservation (for SC) Margin of victory Party fractionalization Party turnover Turnout Mean Standard deviation 1.033 0.283 0.233 0.142 260.0 0.688 0.137 0.819 0.611 0.425 0.350 162.9 0.215 0.200 0.274 0.208 0.119 0.227 64.5 29.0 0.392 13.5 6.4 0.256 0.399 0.221 0.236 0.209 0.683 0.525 0.467 0.501 0.418 0.181 6.10 0.203 2.44 0.055 0.123 0.086 0.213 0.120 0.715 0.525 0.568 0.126 0.042 0.034 0.096 0.066 We also find that the presence of primary schools, teachers, and the quality of infrastructure is significantly positively related to the extent of irrigation (IRRIG) in a village. This result might reflect two effects. First, irrigation may reflect the prosperity of a region, and the regions that are more prosperous are likely to demand more schooling services than poorer regions. Second, irrigation may well reflect the organizational capabilities and expectations of a village. Villages that are able to organize to procure irrigation infrastructure (either via grassroots democracy or through a benevolent landlord) are likely to be able to make use of the same organizational infrastructure to procure good quality schools. Our results also indicate that predominantly upper caste villages are likely to have more primary schools and also to be better equipped with teachers and blackboards. The finding that caste and ethnicity matter for the allocation of educational infrastructure is reinforced by the result that predominantly Muslim villages are less likely to have schools with blackboards. Does economic prosperity have an impact on school allocation? Our results indicate that while skilled wages do not significantly influence school infrastructure in a village, unskilled wages do. Surprisingly, the effect is contrary to expectations. We find that higher unskilled wages in a village 202 WORLD DEVELOPMENT Table 6. Factors Influencing the number of schools and teachers in a village Number of public primary schools Number of public middle schools Number of teachers Marginal effect Standard error Marginal effect Standard error Marginal effect Standard error 8.08* 4.28 7.47 9.74 10.2 12.3 0.27 0.0087*** 6.8 * 106*** 0.059 1.00** 1.01 0.13 6.7 * 104 0.046** 0.66** 0.66 0.58 0.72*** 0.31* 0.42*** – 0.51 0.0019 2.3 * 106 0.43 0.42 1.40 0.72 0.0075 0.017 0.32 0.40 0.48 0.20 0.18 0.13 – 1.57 0.013*** 1.1 * 105** 1.48 0.76 1.70 2.69 0.0099 0.027 0.64 1.16 0.58 0.30 0.43 0.36 1.01 0.0049 5.2 * 106 1.02 0.78 3.14 2.27 0.017 0.035 0.65 0.93 0.98 0.42 0.38 0.42 0.50 0.023*** 2.1 * 105*** 0.092 3.19*** 1.23 2.18 0.0055 0.090* 1.11 1.31 1.03 1.05* 0.38 0.076 – 1.47 0.0055 6.6 * 106 1.24 1.20 4.07 2.05 0.022 0.050 0.90 1.15 1.41 0.55 0.52 0.37 – District variables Irrigated land Population density Scheduled Caste Muslim 2.02** 0.068 8.70** 0.89 0.94 0.050 3.88 1.54 4.36** 0.012 9.55 5.19 2.10 0.10 8.60 3.58 4.44 0.11 15.5 0.60 2.70 0.14 11.2 4.45 Political variables Political reservation Margin of victory Party fractionalization Party turnover Turnout 5.64** 1.72 8.89** 3.22** 2.80 2.27 2.74 4.10 1.36 3.55 2.99 14.2* 0.25 1.59 11.6 4.72 7.80 8.85 3.00 7.84 3.17 11.9 8.71 4.49 7.01 6.50 7.93 11.8 3.88 10.2 2.18 5.78 27.0** 2.90 7.29 12.0 21.3 13.9 2.14 16.9 16.1 23.9 5.57 9.51 4.73 8.72 21.1 34.2 Constant Village variables Bihar dummy Village size Village size squared Caste fractionalization Upper caste Scheduled Caste Muslim Skilled wage Unskilled wage Off-farm income Rural laborers Landless households Irrigated Access to electricity Private primary schools Private middle schools Interaction variables Muslim pop. (district village) Scheduled Caste pop. (district village) Political reservation Scheduled Caste pop. (village) * Denotes statistically significant at the 10% level. Denotes statistically significant at the 5% level. *** Denotes statistically significant at the 1% level. ** are correlated with worse schooling infrastructure. It is possible high unskilled wages signal that education is not crucial for earning a living, that is, h is smaller than in villages with low unskilled wages. Parents in this situation are less concerned about educating their children and therefore less concerned about the availability of schooling. Other village level factors—the availability of off-farm employment and having a large proportion of landless laborers—also have a significant impact on the different dependant variables, which is in the expected direction. Turning to consider whether political competition has any impact on schooling infrastructure in a village, we look at the results to the five political variables. We find that one variable—party turnover—has a significant positive effect on the number of primary schools and the quality of their infrastructure. It therefore seems that political competition is important in determining schooling provision in rural India. This reinforces Bardhan and Mookherjee’s (2000) argument that “with limited political contestability of local elections, leaders may be subjected to capture by special interest groups.” Such elite capture is less likely when party turnover is high, making the party more accountable to the electorate. This finding is further reinforced by the fact that a higher margin of victory results in fewer middle schools in the village, which suggests that politicians’ greater confidence regarding their political position leads to a lower allocation of resources to schooling. Party fractionalization and party turnover are positively correlated with the presence of primary schools indicating a significant beneficial effect of political competition at the party level. While party turnover has a significant positive impact on schooling infrastructure (presence of both blackboard and pucka structure), party fractionalization has no significant impact on these variables at all. Our results for the political reservation variable seem to confirm Pande’s (2003) argument that reservation of seats actually worsens, if anything, the number of public primary schools and their infrastructure: where significant, political reservation has a negative effect on both schooling availability and infrastructure. The positive and highly significant coefficient of the interaction term between political reservation and the proportion of POLITICAL MARKET CHARACTERISTICS AND THE PROVISION OF EDUCATIONAL INFRASTRUCTURE IN NORTH INDIA 203 Table 7. Factors influencing schooling infrastructure in a village Blackboard Village variables Bihar dummy Village size Village size squared Caste fractionalization Upper caste Scheduled Caste Muslim Skilled wage Unskilled wage Off-farm income Rural laborers Landless households Irrigated Access to electricity Private primary schools * Pucka structure Cement floor Coefficient Standard error Coefficient Standard error Coefficient Standard error 0.44 0.0088*** 6.3 106 0.59 1.63** 0.62 2.53** 0.0030 0.069** 0.63 1.06 2.49** 0.38 0.26 0.28 0.91 0.0033 4.0 106 0.76 0.75 2.49 1.20 0.013 0.031 0.56 0.72 0.89 0.33 0.32 0.23 0.062 0.011*** 9.5 106** 1.06 0.68 0.84 0.072 0.017 0.056* 0.91* 0.23 1.88** 0.57* 0.50 0.37* 0.88 0.0034 4.0 106 0.78 0.68 2.54 1.23 0.014 0.031 0.54 0.68 0.86 0.33 0.31 0.22 0.42 0.0079*** 5.5 106 0.46 0.39 2.66 1.05 0.011 0.059** 0.65 0.016 1.39* 0.50 0.20 0.23 0.82 0.0031 3.8 106 0.67 0.65 2.24 1.15 0.012 0.028 0.50 0.64 0.78 0.31 0.29 0.21 District variables Irrigated land Population density Scheduled Caste Muslim Political variables Political reservation Margin of victory Party fractionalization Party turnover Turnout 4.03** 0.098 16.4** 3.72 1.72 0.088 7.15 2.82 2.20 0.12 6.51 0.46 1.58 0.083 6.93 2.71 1.99 0.15* 13.6** 1.04 1.50 0.082 6.49 2.53 8.19** 4.84 10.1 6.02** 6.43 4.09 5.01 7.27 2.43** 6.30 5.74 3.54 7.39 3.87* 1.69 3.92 4.60 7.09 2.29 5.98 6.64* 3.64 6.91 2.03 4.21 3.75 4.38 6.60 2.16 5.65 Interaction variables Muslim pop. (district village) Scheduled Caste pop. (district village) Political reservation Scheduled Caste pop. (village) Threshold 1 Threshold 2 14.4*** 3.51 40.4* 9.17 11.9 5.03 12.7 21.7 7.45 7.49 1.03 3.72 21.5 5.21 6.05 5.27 12.9 21.1 7.22 7.22 2.10 12.5 31.6 4.96 6.10 4.86 11.6 19.6 6.70 6.70 Denotes statistically significant at the 10% level. Denotes statistically significant at the 5% level. Denotes statistically significant at the 1% level. ** *** SC households in the village indicates that, whereas reservation for SC does not improve schooling infrastructure in general, it leads to improvements in villages which are of the same caste identity as the reserved candidate. Thus, reservation works to the benefit of the group for whom the reservation is being made. Since the result indicates that Scheduled Caste politicians try to direct educational resources toward SC villages, the lower allocation of schools to villages with a large fraction of SC households cannot be explained by their lower preference for education alone (though that may still play a role). This group seems to use political power to increase its allocation of educational infrastructure, when it has the opportunity. Once again, we find that the probability of having a school with a blackboard is higher for villages that have a high proportion of Muslims, which are also located in districts with a large Muslim population. Thus, while a Muslim village on its own does not seem to be significantly advantaged or disadvantaged, Muslim villages which are in districts with a large Muslim population have an advantage at least in terms of some schooling infrastructure like blackboards. Again, this is related to the impact of minorities when they are in a strong position politically. Thus, Muslims do seem to value educational infrastructure and use the political power they receive from larger numbers to direct educational resources toward themselves. 7. CONCLUSION In this paper, we considered the provision of schools in rural India, particularly the influence of demographic and economic characteristics as well as of local democracy. Is school allocation an outcome of the size of the village and its prosperity? Does ethnic composition also play a role? What role does the local democracy play in this allocation? To consider these issues, we analyzed the impact of these variables on a range of school availability and infrastructure variables. Political competition is proxied by including the margin of victory of the governing party in a constituency, the turnover amongst the governing parties in the last few elections, party fractionalization, and the reservation of a seat for a minority community candidate. Our results indicate that one political variable—turnover amongst the governing parties—has a significant positive effect on the number of primary schools and the quality of their infrastructure. This result is reinforced by the fact that a higher margin of victory results in fewer middle schools in the village. Thus, political competition, as reflected in the ease with which the incumbent party can be thrown out of power, has a significant effect on the provision of educational infrastructure. Our results also indicate that the reservation of political seats for the Scheduled Castes decreases the number 204 WORLD DEVELOPMENT of primary schools in the village. This may indicate that, as Pande (2003) suggests, SC households tend to prefer monetary transfers to the provision of public goods. This might be at least partly because of uncertainty regarding whether schools and other such public services will be accessible to Scheduled Castes. However, we also find that when a village with a high proportion of Scheduled Caste households is in a constituency with political reservation, then it improves the number of pri- mary schools and some quality indicators in that village. In the latter case, of course, politicians are aware that in a majority SC village, fair access to public goods is more likely and therefore they use their political power to procure public goods rather than merely monetary resources. Expanding political reservation is therefore likely to increase the Scheduled Castes’ access to educational infrastructure, though it may do so at the cost of reducing access for other groups. NOTES 1. This could be seen as differential returns to schooling for different classes of the population. 2. Since it’s a Cobb–Douglas specification the model does not allow for zero-levels of the goods, since then the income would drop to zero, but one could argue that theoretically every village has access to some non-zero level of education, even if the closest school is located quite far away. 3. The public primary school whose characteristics are included in the data set is chosen at random, if the village had more than one public primary school. 4. Though it is true that political reservation may increase the return to education for SC/ST households by giving them better access to well-paid public-sector jobs, this would in itself indicate an increase in their political power. REFERENCES Alesina, A., Baqir, R., & Easterly, W. (1999). Public goods and ethnic divisions. Quarterly Journal of Economics, 114(4), 1243–1284. Banerjee, A., & Somanathan, R. (2007). The political economy of public goods: Some evidence from India. Journal of Development Economics, 82(2), 287–314. Bardhan, P., & Mookherjee, D. (2000). Capture and governance at local and national levels. American Economic Review, 90(2), 135–139. Bardhan, P., & Mookherjee, D. (2006). Pro-poor targeting and accountability of local governments in West Bengal. Journal of Development Economics, 79, 303–327. Besley, T., & Burgess, R. (2002). The political economy of government responsiveness: Theory and evidence from India. Quarterly Journal of Economics, 117(4), 1415–1451. Besley, T., Pande, R., & Rao, V. (2007). Just rewards? Local politics and public resource allocation in South India. Development economics discussion paper Series, 49, STICERD. London School of Economics. Betancourt, R., & Gleason, S. (2000). The allocation of publicly-provided goods to rural households in India: On some consequences of caste, religion and democracy. World Development, 28(12), 2169–2182. Borooah, V. K., & Iyer, S. (2005). Vidya, Veda and Varna: The influence of religion and caste on education in rural India. Journal of Development Studies, 41(8), 1369–1404. Chattopadhyay, E., & Duflo, R. (2004). Women as policy makers: Evidence from a randomized policy experiment in India. Econometrica, 72(5), 1409–1443. Chaudhury, N., Hammer, J., Kremer, M., Muralidharan, K., & Halsey Rogers, F. (2006). Missing in action: Teacher and health worker absence in developing countries. Journal of Economic Perspectives, 20(1), 91–116. Das, M. B., & Dutta, P. V. (2008). Does caste matter for wages in the Indian labor market? In Paper presented at the third IZA/World Bank conference on employment and development, Rabat Morocco May 2008. Dreze, J., & Gazdar, H. (1996). Uttar Pradesh: The burden of inertia. In J. Dreze, & A. Sen (Eds.), Indian development (pp. 33–128). Delhi: Oxford University Press. Dreze, J., & Sen, A. (1996). India: Economic development and social opportunity. New Delhi: Oxford University Press. Galasso, E., & Ravallion, M. (2005). Decentralized targeting of an antipoverty program. Journal of Public Economics, 89, 705–722. Government of India (2001). Census of India. Government of India (2002a). Economic survey. 2002/3. Government of India: Central Statistical Organisation. Government of India (2002b). Selected socio economic statistics, India 2002. Government of India: Ministry of Statistics and Programme Implementation, Central Statistical Organisation. Government of India (2002c). Statistical abstract, India. Government of India: Ministry of Statistics and Programme Implementation, Central Statistical Organisation. Grossman, G., & Helpman, E. (1996). Electoral competition and special interest politics. Review of Economic Studies, 63, 265–285. Keefer, P., & Khemani, S. (2005). Democracy, public expenditures, and the poor: Understanding political incentives for providing public services. World Bank Research Observer, 20(1), 1–27. Kingdon, G. G. (1998). Does the labour market explain lower female schooling in India?. Journal of Development Studies, 35(1), 39–65. Lieten, G. K. (1996). Development, devolution and democracy: Village discourse in West Bengal. Delhi: Sage Publications. Madeshwaran, S., & Attewell, P. (2007). Caste discrimination in the Indian urban labour market: Evidence from the National Sample Survey. Economic and Political Weekly, 42(41), 4154– 4156. Matthew, G., & Nayak, R. (1996). Panchayats at work: What it means for the oppressed?. Economic and Political Weekly, 28, 1765–1771. Pande, R. (2003). Can mandated political representation increase policy influence for disadvantaged minorities? Theory and evidence from India. American Economic Review, 93(4), 1132–1151. Reserve Bank of India (2001). Credit deposit ratio: Report on trends and progress of banking in India, 2000–1. Reserve Bank of India. Schmid, J. P. (2007). Was the district primary education programme in India effective? Draft Ph.D. thesis chapter. August. Zurich: Nadel ETH (Department for Humanities and Social Sciences, Swiss Federal Institute of Technology). Thakur, R. (1995). The Government and Politics of India. New York: St. Martin’s Press. UNDP (2001). National Human Development Report. India. Unni, J. (2001). Earnings and education among ethnic groups in rural India. NCAER working paper series 79. National Council of Applied Economic Research. Uppal, Y. (2009). The disadvantaged incumbents: Estimating incumbency effects in Indian state legislatures. Public Choice, 138(1–2), 9–27. Available online at www.sciencedirect.com