UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Caste, religious conflict and economic development : the Indian experience Roy Chaudhuri, Arka 2015

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2015_september_roychaudhuri_arka.pdf [ 642.59kB ]
Metadata
JSON: 24-1.0166458.json
JSON-LD: 24-1.0166458-ld.json
RDF/XML (Pretty): 24-1.0166458-rdf.xml
RDF/JSON: 24-1.0166458-rdf.json
Turtle: 24-1.0166458-turtle.txt
N-Triples: 24-1.0166458-rdf-ntriples.txt
Original Record: 24-1.0166458-source.json
Full Text
24-1.0166458-fulltext.txt
Citation
24-1.0166458.ris

Full Text

Caste, Religious Conflict and EconomicDevelopment: The Indian ExperiencebyArka Roy ChaudhuriB. Sc (Hons.), University of Calcutta, 2006M.S (Q.E), Indian Statistical Institute, 2008A THESIS SUBMITTED IN PARTIAL FULFILLMENTOF THE REQUIREMENTS FOR THE DEGREE OFDoctor of PhilosophyinTHE FACULTY OF GRADUATE AND POSTDOCTORALSTUDIES(Economics)The University of British Columbia(Vancouver)July 2015c© Arka Roy Chaudhuri, 2015AbstractThis thesis aims to understand the economic and political changes in India andhow it affects different marginalized groups in India. It looks at the effects ofmandated political representation of Scheduled Castes and Scheduled Tribes, theeffect of British colonization on Hindu-Muslim conflict in post-Independent Indiaand the evolution of economic conditions of Muslims in India in the past threedecades.The first research chapter looks at the effect of political quotas for ScheduledCastes and Scheduled Tribes on households belonging to these groups. ScheduledCastes and Scheduled Tribes form some of the most disadvantaged groups in In-dia. I exploit the policy rule mandating reservations for these groups to identifythe effect of political representation of these groups. I find that for ScheduledCaste politicians effectively target narrow based public goods such as participa-tion in workfare program to members of their own ethnic groups but do not do sofor broad based public goods such as health, education and access to subsidizedfood grains.The second research chapter looks at the effect of British colonization oniipost-Independence religious conflict in India. British colonialism has often beenblamed for the worsening of Hind-Muslim relations.Comparing districts ruled bynative kings with districts which were ruled directly by the British, I find no ad-verse effect of British colonialism.The third research chapter looks at the evolution of the economic conditions ofMuslims in the last three decades-a period which has been characterized by rapideconomic growth in India. I compare Muslims with non-Muslims in education,occupation choice, wages and consumption expenditure. I find that Muslims areworse off compared to non-Muslims and this relative deprivation gets more acuteover time.iiiPrefaceThis thesis is original, unpublished, independent work by the author, Arka RoyChaudhuri.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Mandated Political Representation and Development Outcomes: Ev-idence from India . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2 Institutional Background . . . . . . . . . . . . . . . . . . . . . . 132.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18v2.3.1 Population and Political Data . . . . . . . . . . . . . . . . 182.3.2 Education Data . . . . . . . . . . . . . . . . . . . . . . . 192.3.3 Infant Mortality Data . . . . . . . . . . . . . . . . . . . . 202.3.4 Public Distribution System Data . . . . . . . . . . . . . . 212.3.5 Data on Employment under National Rural EmploymentGuarantee Act . . . . . . . . . . . . . . . . . . . . . . . . 222.4 Scheduled Caste Reservations . . . . . . . . . . . . . . . . . . . . 232.4.1 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . 232.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.4.3 Robustness Checks . . . . . . . . . . . . . . . . . . . . . 312.4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 342.5 Scheduled Tribe Reservations . . . . . . . . . . . . . . . . . . . . 372.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413 Colonization and Religious Violence: Evidence from India . . . . . 593.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 603.2 British Colonization and Rise in Religious Violence . . . . . . . . 643.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 723.4 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . 743.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 763.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 804 The Economic Lives of Muslims in India, 1983-2012 . . . . . . . . . 934.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93vi4.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 984.3 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 1004.3.1 Education Attainment . . . . . . . . . . . . . . . . . . . . 1004.3.2 Occupational Choice . . . . . . . . . . . . . . . . . . . . 1044.3.3 Wages . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1064.3.4 Consumption . . . . . . . . . . . . . . . . . . . . . . . . 1104.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1125 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145viiList of TablesTable 2.1 Summary Statistics: Education . . . . . . . . . . . . . . . . . 45Table 2.2 Summary Statistics: Infant Mortality . . . . . . . . . . . . . . 46Table 2.3 Summary Statistics: PDS Participation and Food Grains . . . . 47Table 2.4 Summary Statistics: NREGA Participation . . . . . . . . . . . 48Table 2.5 SC Reservation:Primary Education . . . . . . . . . . . . . . . 49Table 2.6 SC Reservation:Infant Mortality . . . . . . . . . . . . . . . . . 50Table 2.7 SC Reservation:Public Distribution System . . . . . . . . . . . 51Table 2.8 SC Reservation:NREGA Participation . . . . . . . . . . . . . . 52Table 2.9 SC Reservation: Robustness Checks: Primary Education . . . . 53Table 2.10 SC Reservation:Robustness Checks: Infant Mortality . . . . . . 54Table 2.11 SC Reservation:Robustness Checks: PDS Food Grains . . . . . 55Table 2.12 SC Reservation:Robustness Checks: PDS Participation . . . . . 56Table 2.13 SC Reservation:Robustness Checks: NREGA Participation . . 57Table 2.14 Effect of ST Reservation . . . . . . . . . . . . . . . . . . . . . 58Table 3.1a Summary Statistics: Dependent Variables . . . . . . . . . . . 81viiiTable 3.1b Differences in Geographical Controls . . . . . . . . . . . . . . 82Table 3.1c Differences in Population Controls . . . . . . . . . . . . . . . 83Table 3.1d Differences in Political Controls . . . . . . . . . . . . . . . . . 84Table 3.2 OLS: Probability of Riot . . . . . . . . . . . . . . . . . . . . . 85Table 3.3 OLS: Total Cases . . . . . . . . . . . . . . . . . . . . . . . . . 86Table 3.4 OLS: Total casualties . . . . . . . . . . . . . . . . . . . . . . 87Table 3.5 IV: First Stage . . . . . . . . . . . . . . . . . . . . . . . . . . 88Table 3.6 IV: Probability of Riot . . . . . . . . . . . . . . . . . . . . . . 89Table 3.7 IV: Total Cases . . . . . . . . . . . . . . . . . . . . . . . . . . 90Table 3.8 IV: Total casualties . . . . . . . . . . . . . . . . . . . . . . . . 91Table 3.9 Robustness Check: Effect of Death of Ruler Without NaturalHeir . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92Table 4.1 Oaxaca-Blinder decomposition of log wage gaps . . . . . . . . 131Table 4.2 Oaxaca-Blinder decomposition of log wage gaps: Rural . . . . 132Table 4.3 Oaxaca-Blinder decomposition of log wage gaps: Urban . . . . 133Table 4.4 Oaxaca-Blinder decomposition of log wage gaps: Muslim vsUpper Caste . . . . . . . . . . . . . . . . . . . . . . . . . . . 134Table 4.5 Oaxaca-Blinder decomposition of log wage gaps: Muslim vsSC/ST . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135Table 4.6 Oaxaca-Blinder decomposition of log wage gaps: Male . . . . 136Table 4.7 Oaxaca-Blinder decomposition of log wage gaps: Female . . . 137Table 4.8 Oaxaca-Blinder decomposition of log consumption gaps . . . . 138ixTable 4.9 Oaxaca-Blinder decomposition of log consumption gaps: Rural 139Table 4.10 Oaxaca-Blinder decomposition of log consumption gaps: Urban 140Table 4.11 Oaxaca-Blinder decomposition of log consumption gaps: Mus-lim vs Upper Caste . . . . . . . . . . . . . . . . . . . . . . . . 141Table 4.12 Oaxaca-Blinder decomposition of log consumption gaps: Mus-lim vs SC/ST . . . . . . . . . . . . . . . . . . . . . . . . . . . 142xList of FiguresFigure 2.1 Reservations for Gujarat . . . . . . . . . . . . . . . . . . . . 43Figure 2.2 Time lag in Reservations . . . . . . . . . . . . . . . . . . . . 44Figure 4.1 Distribution of education attainment . . . . . . . . . . . . . . 115Figure 4.2 Distribution of education attainment-Rural . . . . . . . . . . . 116Figure 4.3 Distribution of education attainment-Urban . . . . . . . . . . 117Figure 4.4 Distribution of education attainment-Muslims vs Upper Caste 118Figure 4.5 Distribution of education attainment-Muslims vs SC/ST . . . . 119Figure 4.6 Distribution of education attainment-Male . . . . . . . . . . . 120Figure 4.7 Distribution of education attainment-Female . . . . . . . . . . 121Figure 4.8 Distribution of occupational choices . . . . . . . . . . . . . . 122Figure 4.9 Distribution of occupational choices-Rural . . . . . . . . . . . 123Figure 4.10 Distribution of occupational choices-Urban . . . . . . . . . . 124Figure 4.11 Distribution of occupational choices-Muslims vs Upper Caste 125Figure 4.12 Distribution of occupational choices-Muslims vs SC/ST . . . . 126Figure 4.13 Distribution of occupational choices-Male . . . . . . . . . . . 127xiFigure 4.14 Distribution of occupational choices-Female . . . . . . . . . . 128Figure 4.15 The Log Wage Distributions for 1983 and 2004-2005 . . . . . 129Figure 4.16 The Log Consumption Distributions for 1983 and 2004-2005 . 130xiiAcknowledgmentsI am deeply indebted to the members of my thesis supervisory committee, Dr.Siwan Anderson, Dr. Ashok Kotwal and Dr. Kevin Milligan for their guidancethroughout the length of my PhD. It would have been impossible to complete thedissertation without their advice on this difficult journey. Their advice has been in-strumental in me being able to understand what research actually entails. I wouldalso like to thank my thesis committee members for generous financial assistancein the form of research assistantships and travel grants at different points of myPhD.I would also like to thank Dr. Mukesh Eswaran, Dr. Patrick Francois, Dr.Amartya Lahiri, Dr. Marit Rehavi and Dr. Shinichi Sakata for many helpfulcomments and suggestions on my research. I am grateful to seminar participantsat the Canadian Economics Association conferences, Econ 640 workshop at UBCand the UBC economics empirical lunch for many helpful suggestions.I am grateful to my teachers at Indian Statistical Institute for inculcating in mea spirit of enquiry and for making me realize that one needs to shed all dogmain a life as a researcher. I would like to express my gratitude to my friends andxiiicolleagues for their help and support.I would especially like to thank Anirbanand Sourabh for their help and guidance throughout my PhD. I would also liketo thank Ashokankur, Rahul and Sayan at Indian Statistical Institute and Nishantand Subrata at UBC for their help.I would especially like to thank Maureen Chinfor taking care of all the administrative work and her guidance throughout theprogram.This thesis would not have been possible without the constant support of myfamily especially my mother. Finally I would like to express my gratitude to mypartner, Shampa for her unwavering support and advice regarding different aspectsof my thesis.xivChapter 1IntroductionFor a researcher in development economics, India is one of the most interestingplaces to study. Not only is it the second most populous countries in the world, itis also one of the fastest growing economies in the world. The last three decadeshas been one of the most interesting phases in Indian history. Various politicalchanges have occurred such as the weakening of the Congress party which wasat the forefront of the independence movement in India, the rise of the Hindu na-tionalist party, the Bharatiya Janata Party (BJP) and the rise of identity politicsof marginalized groups spearheaded by parties like the Samajwadi Party (SP) andthe Bahujan Samaj Party (BSP). This was also a time of rapid economic growth.Although India was officially a mixed economy, in practise it suffered from a highdegree of protectionism which led to the very low average growth rates of 3.5 %,pejoratively referred to as the Hindu rate of growth. The Indian economy startedopening up in the mid-1980s and this process of liberalization received a mas-1sive boost when facing a Balance of Payments crisis it was forced to open up itseconomy in return for IMF support (Topalova (2007)). Since then India has beenone of the fastest growing economies in the world. In the light of these politicaland economic changes it becomes imperative to look how different marginalizedgroups have fared so far.Post-colonial India was founded on democratic principles. India is a devel-oping country with high degree of democratic participation with voter turnout ofmore than 60% (Khemani (2004)). However, it is often not clear whether the over-all democratic set-up and the improvement in economic circumstances have trans-lated into better socio-economic outcomes of marginalized communities. In Indiathe issues faced by disadvantaged minorities are multifaceted as they simultane-ously face problems relating to security and economic equity (Wilkinson (2006),Iyer et al. (2011)). This study aims to evaluate the Indian experience throughthe prism of disadvantaged minorities specifically Scheduled Castes, ScheduledTribes and Muslims.Scheduled Castes form the lowest tier in the Hindu caste hierarchy. Sched-uled Tribes, also known as adivasis (original inhabitants, are) composed of tribalgroups who were distinguished by their physical isolation and were removed fromthe mainstream Hindu society (Chin and Prakash (2011)). Various forms of dis-crimination including untouchability has been practiced against both ScheduledCastes and Scheduled Tribes for centuries by upper castes. This has led to acuteimpoverishment among these groups. The 1950 Indian Constitution had guaran-teed mandated political representation for Scheduled Castes and Tribes. Chapter22 looks at the effect of such political reservation on primary education attainment,infant mortality, access to the subsidized food grain distribution system known asthe Public Distribution System (PDS) and participation in the workfare programthe National Rural Employment Guarantee Act (NREGA). The aim of this chap-ter is to analyze whether mandated political representation leads to improved out-comes for hitherto marginalized groups. My empirical analysis which exploits thepolicy rule that determines political quotas shows that in case of Scheduled Castepoliticians, for broad based public goods like education, health and PDS there isno such effect whereas for narrow based program like NREGA there is positiveeffect on outcomes of members of own ethnic group. I find no comparable effectsfor Scheduled Tribe politicians.Religious violence in India, more specifically Hindu-Muslim religious vio-lence has imposed a great strain on the social fabric of India. More than 7000deaths have occurred in India over the 1950-1995 period. These riots have alsoled to large scale displacement. The cost of these communal riots should not beseen only in terms of the immediate loss of life and property but also in terms ofthe long-run cost it imposes on a country’s institutions due to the increased ethnictensions. It has long been argued that British colonialism had a role in institution-alizing Hindu-Muslim communal discord (Kabir (1969), Das (1990). In Chapter3 I examine this claim by comparing the incidence of Hindu-Muslim evidencebetween British ruled areas and areas ruled by native kings during the colonialperiod. I find that contrary to conventional wisdom British ruled areas actuallysee lower incidence of Hindu-Muslim violence.3Muslims form one of the most disadvantaged groups in modern India. Thisrelative deprivation of Muslims is more stark considering the fact that Muslimshad ruled large arts of India for about 600 years preceding British colonialism. InChapter 4 I document the evolution of economic condition of Muslims vis-a-visnon-Muslims in the last three decades in terms of education, occupation choice,wages and consumption expenditure. I document how the relative deprivation ofMuslims has been getting worse over time. I find that this worsening of condi-tions for Muslims over time happens both in respect to upper caste Hindus andScheduled Castes/Tribes and is more stark in respect of Muslim men. I performa quantile analysis to shed light on which points of the wage and consumptiondistribution are these changes most evident.4Chapter 2Mandated Political Representationand Development Outcomes:Evidence from India2.1 IntroductionAffirmative action policies have often been used in various countries as a meansof targeting historically disadvantaged minority groups. India has a history of var-ious affirmative action policies which have been established with a view to helphistorically disadvantaged minorities such as the Scheduled Castes, ScheduledTribes and women. One of those policies aimed at giving political representationto Scheduled Castes and Scheduled Tribes. The Constitutional Amendment Actguaranteed mandated political representation for Scheduled Castes and Tribes in5accordance with their population shares. This chapter analyses the effect of reser-vation for Scheduled Castes (SC) and Scheduled Tribes (ST) on the developmentoutcomes of Scheduled Caste and Scheduled Tribe households. Specifically I lookat how reservations for Scheduled Castes and Scheduled Tribes in India affectschooling outcomes, health outcomes, access to the food security system, PublicDistribution System (PDS) and participation in the workfare program, NationalRural Employment Guarantee Act (NREGA).Scheduled Castes form the lowest tier in the Hindu caste hierarchy. Sched-uled Tribes, also known as adivasis (original inhabitants) are composed of tribalgroups who were distinguished by their physical isolation and were removed fromthe mainstream Hindu society (Chin and Prakash (2011)). Various forms of dis-crimination including untouchability has been practiced against both ScheduledCastes and Scheduled Tribes for centuries by upper castes. Political representa-tion for Scheduled Castes and Scheduled Tribes was mandated in the 1950 IndianConstitution. The political reservations were originally introduced for a periodof ten years and have been continuously renewed thereafter. Reservations werebrought in order to provide political voice to hitherto marginalized communities.It was also thought that Scheduled Caste and Scheduled Tribe politicians wouldbe better able to serve the interests of their own ethnic community. In light ofthese objectives, the importance of developmental outcomes is important. Doesreservation lead to better development outcomes for disadvantaged groups?There has been a growing literature on the poor record of India in educationand health outcomes with different authors having analysed various factors such as6teacher and doctor absenteeism (Kremer et al. (2005),Banerjee and Duflo (2006)).What these papers point to is a pervasive lack of state capacity even to maintainthe existing infrastructure. Similarly Jha and Ramaswami (2010) have writtenabout the large extent of leakage in PDS.1. Others similarly have written aboutthe poor implementation of NREGA (Johnson (2009)). It is in this backdrop ofa broken down public good/service delivery mechanism that the question I askin this chapter becomes important-whether reservation has led to better or worseoutcomes for Scheduled Castes and Scheduled Tribes.To identify the effect of reservations for Scheduled Castes, I use the policyrule governing political reservations in India in a manner similar to that used inPande (2003) but I extend it to the district level.2 The 1950 Constitution man-dates that reservation for Schedule Castes (SC) in proportion to their populationshares. This rule is implemented by allotting to each state a number of reservedconstituencies based on the state’s population share of SC’s. Given that share, adistrict’s share of reserved constituencies is based on its share of SC’s among thestate SC population . Since changes in reservation only happen during a sittingof the Delimitation Commission3, the exact share of reserved constituencies in a1. Leakage refers to the differences in the supply of subsidized food grains by the governmentand the actual receipt of subsidized food grains by households2. Each state in India is divided into a number of administrative divisions which are districts.Elections are fought at the electoral constituency level with each district comprising of severalelectoral constituencies.Since I can identify the district of residence of a household in my surveydata and not the electoral constituency of residence, I aggregate up the political variables from theconstituency level to the district level. The fact that legislative assembly constituency boundariesfall entirely within a single district allows me to do so3. Delimitation Commission determines the boundaries of assembly constituencies and theirreservation status7state and district depends on the population shares of the last census before thesitting of the Delimitation Commission. This makes it possible for us to controlfor the present population shares which is crucial since minority population sharesmight be correlated with development outcomes.4For Scheduled Tribes, the policy rule is that each state gets a number of re-served constituencies for Scheduled Tribe legislators in proportion to the popu-lation share of Scheduled Tribes in the state. However in contrast to the case ofScheduled Castes, the share of each district is not separately determined. Withina state all electoral constituencies are ranked according to the population shareof Scheduled Tribes in the constituency and then the first n constituencies are re-served for Scheduled Tribes where n is the number of constituencies to be reservedfor Scheduled Tribes in that state. The case of Scheduled Tribe reservations areanalyzed separately in Section 2.5.I find that for broad based goods such as primary education, infant mortalityand PDS access, Scheduled Caste households fare relatively worse in districts witha higher proportion of reserved constituencies. However for a narrow based publicgood such as NREGA access, Scheduled Caste legislators do target the membersof their own ethnic group. I find no such pattern for Scheduled Tribe legislators.This chapter contributes to the broad literature that looks at the effect of politi-4. In the US context, Gordon (2004) uses a similar strategy to estimate the effects of the fed-eral government’s education program, Title I on state and local education revenue and spendingdecisions by recipient school districts. Assessing the effects of Title I is difficult because a dis-tricts poverty determines its Title I allocation, but poverty also affects a district’s outcome throughother channels. The author uses the fact that federal grants are based on decennial census povertyestimates with a time ag of three years which allows her to control for current poverty in herregression.8cian identity (Levitt (1996), Milligan and Smart (2005), Washington (2008), Hodlerand Raschky (2014)). Rehavi (2007), Clots-Figueras (2011) and Clots-Figuerasa(2012) look at the effects of electing women legislators to office. Many papersin this literature look at whether legislators target members of their own ethnicgroup-however the evidence remains mixed. Franck and Rainer (2012) finds thatfor primary schooling and infant mortality there is significant evidence of ethnicfavouritism in sub-Saharan Africa and Burgess et al. (2014) find that Kenyan dis-tricts that share the ethnicity of the President receive twice as much expenditureon road building and has four times the length of paved roads compared to whatwould be predicted by their population share. However Kudamatsu (2009) findthat there is no effect of the President’s ethnicity on infant mortality in Guineaand Kasara (2007) shows that for farmers sharing the ethnicity of the head of stateleads to the imposition of higher tax rates on cash crops. In the US context a largeliterature looks at the effect of electing Black politicians to local political officeon Black employment outcomes (Eisinger (1982), Sass and Mehay (2003), Nye,Rainer, and Stratmann (2014))5.In the Indian context a large literature has emerged in the last decade whichlooks at the effect of reservations for minorities at a more local level. Most ofthis literature has looked at the effect of political representation on public goodsprovision. The 73rd and 74th Constitutional Amendment Act mandated setting upof regularly elected Gram Panchayats (village councils) which were supposed tolook after the administrative needs of the local population including public good5. See Hajnal (2001) for a brief review of the literature9provision such as public buildings, water and roads (Chattopadhyay and Duflo(2004)).The Constitutional amendments also mandated political reservation forwomen, Scheduled Castes and Tribes at the village level. Most of the papers inthis literature look at the effects of this policy in a selected few states. Chat-topadhyay and Duflo (2004) was one of the first papers to study the effects ofreservations at the gram panchayat level in the Birbhum district of West Bengaland Udaipur district of Rajasthan. Exploiting the fact that reservations for womenfor the Gram Panchayat Pradhan (village council chief) were assigned randomly,they found that women Panchayat Pradhans tend to invest more in public goodsthat are valued more by their female constituents: female Pradhans invested morein drinking water and roads in West Bengal while they invested more in drink-ing water in Rajasthan. Using survey data from three southern states of India,Besley et al. (2004) found that sharing the Pradhan’s caste identity leads to in-creased access to public goods but only for low spillover public goods. Bardhan,Mookherjee, and Parra Torrado (2010) find that reservation of Pradhan posts forminority members in West Bengal was associated with a significant increase inbenefits received by the members of the group of the pradhan.This chapter is most closely tied to the literature that looks at the effects ofreservation for different disadvantaged groups in state legislative assemblies.Pande(2003) looks at the effect of reservations for Scheduled Caste and Scheduled Tribelegislators on state level legislation and spending. She finds that for broad basedgoods such as education spending there is no favourable effect of minority repre-sentation but for narrowly targeted goods such as job quota or welfare spending10there is a positive effect. Chin and Prakash (2011) using state level data andthe methodology used in Pande (2003) find that reservation for Scheduled Casteshave no effect on poverty reduction while reservations for Scheduled Tribes havea positive effect on poverty reduction. Looking at rural areas in a sample of 65districts, Krishnan (2007) finds a positive effect of Schedule Caste reservation onprimary schools and no effects for public health facilities, roads and drinking wa-ter and no effects or negative effects of Schedule Tribe legislators on provision ofpublic goods. She finds no evidence of targeting by legislators belonging to ei-ther Scheduled Castes or Scheduled Tribes towards members of their own ethnicgroup.This chapter contributes to the above literature by looking at the effect of reser-vations for Scheduled Castes in India on actual outcomes using district level reser-vation data. Specifically, it contributes to the existing literature in three differentways. Firstly, previous literature has emphasized on the role of legislator iden-tity on legislation (Pande (2003)) or public good provision (Krishnan (2007)). Incontrast my research focusses on actual individual level outcomes (primary edu-cation, infant mortality) and household level participation in public service pro-grams (PDS, NREGA).6Pande (2003) showed that policies enacted by legislatorsin reserved seats for SC/STs differ from the policies enacted by legislators in nonreserved seats. However what remains unanswered is whether the differences inpolicy adopted by SC/ST legislators translates into differential outcomes at anindividual or household level. Policy changes might not always truly reflect the6. Chin and Prakash (2011) focusses on the effect of reservations on state-level poverty.11changes in outcomes. Politicians often target a number of policies together andthus focussing on one policy or a subset of policies can underestimate the trueimpact (Alesina (1997)). This becomes more important in the context of less de-veloped countries like India where legislator effort manifests in different kinds ofinformal arrangements (Nayak, Saxena, and Farrington (2002))which are difficultto measure. Hence focussing on outcomes gives us an idea of the net effect of theeffort put in by legislators towards the welfare of her constituents. The analysisof the impact of reservation on individual and household level outcomes is alsoimportant since it gives an estimate of the magnitude of the problem for importantpredictors of welfare like education and health. My results show that a move fromthe 10th to the 90th percentile of the proportion of seats reserved for SCs reducesthe relative probability of primary education completion, infant mortality and percapita consumption of food grains under PDS for SCs by 6.1%, 17.2% and 15.7%of the sample mean respectively. Similarly a move between the two percentileincreases the relative probability of NREGA participation of SC households byabout 30% of the sample mean.Secondly, instead of looking at the cross-state variation in the proportion ofreserved seats which previous studies have emphasized (Pande (2003), Chin andPrakash (2011)), my study looks at the interstate cross-district variation usingnationally representative data. State is a mich higher level of aggregation andwe might face substantial loss of information using state-level variation in theproportion of reserved seats. Instead I use district level variation, district beingthe most important administrative unit after the state. Using district level variation12also allows me to include state-time fixed effects in addition to district fixed effectsin my regressions. Thus I can control for state-time level factors such as stateassembly election-specific shocks. This is not possible using state-level variation.Finally, my study shows that SC legislators do worse in terms of broad basedgoods and better in terms of narrow based goods. This pattern has been empha-sized in the literature (Keefer and Khemani (2005)) and theoretically analysed inKeefer (2002). In this chapter I show that this pattern holds true empirically.The chapter is organized according to the following sections. The next sectiondiscusses the institutional background. Section 2.3 describes the data used in thepaper. Section 2.4 discusses the effect of Scheduled Caste reservations- 2.4.1 out-lines the empirical strategy, 2.4.2 discusses the results, 2.4.3 contains robustnesschecks and 2.4.4 contains a short discussion of the results. Section 2.5 looks atthe effect of reservation for Scheduled Tribes. Finally section 2.6 concludes.2.2 Institutional BackgroundI use data from the State Legislative Assembly elections which are held after everyfive years. The elections are held in single member electoral constituencies withwinners being determined by plurality rule. The Assembly constituencies fallwithin a districts boundaries which enables me to aggregate up the electoral resultsof the assembly constituencies to the district level. This is essential since mysurvey data on outcomes allows me to identify only the district of residence of thehousehold and not the assembly constituency.Article 332 of the Indian Constitution provides for political reservation in state13elections. Only a person belonging to the Scheduled Castes and Scheduled Tribescan stand for election in constituencies reserved for Scheduled Castes and Sched-uled Tribes respectively. However the entire electorate votes to choose the repre-sentative of the constituency. Reserved constituencies are determined by a statu-tory commission known as the Delimitation Commission which was set up underthe Delimitation Commission Act,1952. Since Independence, the DelimitationCommission has been constituted four times -1952, 1963, 1972 and 2002. TheDelimitation Commission is entrusted with determining the electoral constituencyboundaries. It is also the body which determines which constituencies are to bereserved.India has a federal structure. The Constitution delineates responsibility be-tween the federal government and the state governments. There are three listswhich divides the executive and legislative powers between the central govern-ment and the state governments: the central list which lists out the subjects underthe jurisdiction of the centre, the state list contains the items under the jurisdictionof the state government and the concurrent list contains items on which the centreand state has joint jurisdiction. The central list contains items such as defence,atomic energy, foreign affairs and banking. The state governments control publicorder, police, public health and sanitation, agriculture and industries. The itemson the concurrent list include subjects like education, social security and socialinsurance, and labour over which state governments assume significant responsi-bility of administration7 . Even for schemes financed by the federal government,7. Education was under the state list till 1976 and subsequently transferred to the concurrent14state governments take the leading role in their implementation (Khemani (2004),Rao and Singh (2003)).For the items under the state list and many of the subjects under the concur-rent list, the State Assembly legislators are concerned with making laws, takingspending decisions, working with the local bureaucracy in the administration oftheir constituency and utilizing their constituency development funds8 (Khemani(2004), Rao and Singh (2003), Chin and Prakash (2011) ). Legislators can alsoput pressure on the government regarding issues in the constituency by askingquestions during the Question Hour 9. Legislators are often members of differentoversight committees of the Legislative Assembly and can affect the developmentoutcomes in their respective constituencies through participation in these commit-tees (Banerjee et al. (2010)) . The effect on development outcomes that I see inthis chapter can be due to any of the above roles performed by the state legislators.The basic point is that the MLA is a local political notable who has different in-struments at her disposal to influence development outcomes in her constituency.In this chapter I look at completion of primary schooling, infant mortality,participation in the Public Distribution System (PDS) and participation in Nationallist (Mukundan and Bray (2006))8. MLAs in India are entitled to constituency development funds which are discretionary fundsat their disposal that they can spend on public good provision in their constituencies. The exactamount at the disposal of the MLA varies across states and time. For example the current amountthat a MLA in Delhi can spend is 40 million rupees each year while MLAs in Uttar Pradeshcan spend 15 million rupees each year. Utilization of funds varies significantly across individuallegislators.9. The Question Hour is a time set apart during the session of the Legislative Assembly whenindividual legislators can raise questions which must be answered by the concerned Minister inthe government. Since the proceedings are usually broadcast on television or reported in the medialegislators use this tool to mobilize public opinion on different issues.15Rural Employment Guarantee Act (NREGA). In what follows I briefly talk aboutthese public goods and their administration.As mentioned earlier education is listed in the concurrent list. However statesbear most of the responsibility of the running of state government schools in addi-tion to building new schools and appointing teachers. State spending on educationis much higher than that of the Central Government. Surveys show that one of themain problems of the government schools is teacher absenteeism (Kremer et al.(2005)). This is one of the problems that could be addressed by the local bureau-cracy over which the Member of Legislative Assembly (MLA) have significantsupervisory powers.Public health is on the State list. State Governments mostly appoint healthcare workers (Singh (2008)) and they undertake most of the spending (Berman(1998)).Similar to government schools, health care services are also wracked byhealth worker absenteeism (Kremer et al. (2005))- a problem which is amenableto political pressure from the local MLA.10The Public Distribution System (PDS) is one of the largest safety net pro-grams in India under which primarily food grains are distributed to households.The quota available and the subsidy received by the household is determined bythe wealth of the household. There is considerable difference in the performance10. Often this kind of political pressure is manifested in the form of informal pressure put on thedistrict officials or direct public action taken by the MLA. For example The Tribune newspaper inan article in its December 13, 2005 Chandigarh edition reports an incident where a MLA in Punjab“raided” the local public hospital in response to a complaint from his constituents and found thedoctor on emergency duty absent. He subsequently gave media interviews and informed the localdistrict administration of the doctor’s absence who rushed a team of higher officials to the hospital(http://www.tribuneindia.com/2005/20051213/cth3.htm).16of PDS across states (Khera (2011)).The PDS has often been used by state govern-ments as a response to smooth out temporary income shocks(Besley and Burgess(2002)). One of the persistent problems in the functioning of PDS is the leakageof food grains along different points of the distribution chain from governmentgodowns to Fair Price Shops (FPS) from which PDS food grains are distributedto households.11 Using 2004-05 survey data Jha and Ramaswami (2010) find thatthe per capita consumption of food grains under PDS was 1.03 kg per month whilethe per capita supply of food grains based on official data on subsidized food grainsupply works out to be 2.27 kg per month. This points to an estimated diversionof about 55 percent of the food grains which are supplied by the government intothe PDS system. Legislators can work towards reducing this leakage which wouldinclude working with the local bureaucracy as well as ensuring the proper func-tioning of the Fair Price shops through which such food grains are distributed(Banerjee et al. (2010)).The National Rural Employment Guarantee Scheme (NREGS) was estab-lished in 2005 under which each rural household is guaranteed 100 days of em-ployment. The workfare scheme is supposed to create rural assets. The schemehas been criticized for numerous defects including not enough work available,11. Here leakage refers to the differences in the supply of subsidized food grains by the govern-ment based on administrative records and the actual receipt of subsidized food grains by house-holds. Although this might include transport losses and wastage along the distribution chain, thegeneral perception is that a significant part of these consists of illegal diversion of food grains (Jhaand Ramaswami (2010),Khera (2011) and Svedberg (2012)). Precise estimates of the leakage atdifferent points of the distribution chain are hard to come by. However based on media reportsof various cases of corruption scandals affecting PDS, Svedberg (2012)notes that “..cases includecorruption and theft at all levels, from fraudulent small FPS owners to chief ministers, in somecases of mind-boggling scales.”17non-payment of dues and poor quality of asset creation (Johnson (2009),Maiorano(2014))12. Although the scheme was supposed to be primarily administered bythe local governments, numerous case studies in different states have shown thatthe scheme is susceptible to political interference from MLAs (Maiorano (2014),Bhatia and Dreze (2006))2.3 Data2.3.1 Population and Political DataThe election data are taken from the election reports published by the ElectionCommission of India. My election data set spans from 1970-2012 and coversState Assembly elections in the major Indian states. Depending on the outcomevariable, I use a part of this dataset. The political data in the Election Commissionreports are at the constituency level. Since in both NSS and NFHS dataset, I canidentify the district of residence of the household and not the constituency of res-idence, the constituency level election data is aggregated up to the district level.The reservation status of the constituency is obtained from the Delimitation Com-mission reports. The Census population figures are taken from the India DistrictDatabase for 1961-1991 and from data CDs and website of the Census of Indiafor 2001 and 2011. The population figures for intercensal estimates are obtainedby linear interpolation.12. See Mookherjee (2014) for a brief review of the literature182.3.2 Education DataThe education data considered in this analysis come from the 68th round of Na-tional Sample Survey (NSS) conducted in 2011-2012. This survey collected ed-ucational and demographic information on all members of the surveyed house-holds. The main dependent variable, primary education completion is defined asa dummy variable equal to 1 if the individual has completed primary education.In order to ensure that individuals are old enough to complete primary education,we drop individuals aged less than 14 years from the sample.The NSS survey also provides information on other demographic characteris-tics such as the caste status, religion, area of residence (rural or urban), sex andland possessed by the household of the individual. These variables are includedas controls in my regression.I include the census and current figures of Scheduled Caste population in adistrict as a share of the total Scheduled Caste population in the state as controls inthe regressions estimating the impact of political reservation for Scheduled Castes.In addition, in the robustness check section, I also include the share of ScheduledCaste population in the total district population as control. In the regressionsestimating the impact of the reservation for Scheduled Tribes, I include censusand current share of Scheduled Tribe population in the total district population ascontrols. All these variables are calculated as averages over the period in whichthe individual was 4-11 years of age. I also include the proportion of Congress,Left and Hindu legislators in the assembly as controls. The summary statistics ofthe controls used in education regressions are given in Table 2.1.192.3.3 Infant Mortality DataThe data used in the infant mortality regressions study are derived from the sec-ond round of the National Family Health Survey of India (NFHS-2) conducted in1998. This data-set contains complete fertility histories for ever-married womenaged 15-49 in 1998-99, including the information on time and incidence of childdeaths. I have used this data-set to construct individual-level indicators of in-fant mortality. Infant mortality variable is defined as a dummy variable indicatingwhether a child died by the age of 12 months. The estimation sample containsmore than 200,000 children born over the period 1970-1998. The NFHS data alsoprovides information on a number of demographic characteristics like sex of thechild, month of birth, whether there was single or multiple birth, order of birth,parental education, mother’s work status and whether the child belongs to Sched-uled Caste, Scheduled Tribe or Muslim households. These variables are used ascontrols in the infant mortality regressions.I also include the proportion of Congress, Left and Hindu legislators in theassembly from the district as controls. The regressions estimating the impact ofScheduled Caste reservation additionally include the census and current figuresof Scheduled Caste population in a district as a share of the total state ScheduledCaste population. For estimating the impact of Scheduled Tribe reservation I in-clude census and current share of Scheduled Tribe population in the total districtpopulation as controls. Table 2.2 presents the summary statistics for the controlsused in infant mortality regressions.202.3.4 Public Distribution System DataThe regressions on PDS delivery are household level regressions and the variablesconsidered in these regressions come from the 61st (2004-05), 66th (2009-10) and68th (2011-12) rounds of National Sample Survey. The survey rounds providesinformation on the quantity of products available to a household under PDS in thelast 30 days preceding the date of survey. I have created two variables to indicatePDS access: one measuring the per capita quantity of food-grains available to ahousehold under PDS and a dummy indicating whether the household receivedany items under PDS. The survey rounds provide information on the caste statusof the household, religion of the household, area of residence (rural or urban),educational level of household members, age of household members, sex of thehousehold head, land possessed by the household and household size. These vari-ables are used as controls in the PDS regressions. I also include the proportionof Congress, Left and Hindu legislators in the assemblies from the district, thecensus and current figures of Scheduled Caste population in a district as a shareof the total state Scheduled Caste population for regressions estimating the impactof Scheduled Caste reservation and census and current share of Scheduled Tribepopulation in the total district population as controls in regressions estimating theimpact of Scheduled Tribe reservation. Table 2.3 presents the summary statisticsfor the controls used in PDS regressions.Items under PDS are provided at a subsidized rate for poor households. House-holds are first classified into general, below poverty line (BPL) and Antodaya(poorest households). The subsidies are highest for Antodaya households fol-21lowed by BPL households. However, the information on whether the householdbelongs to BPL or Antodaya categories is not available for 66th round of NSS. Inthe robustness check section I have included two additional dummies indicatingwhether the household belongs to BPL or Antodaya categories using data fromthe 61st and 66th rounds of NSS.2.3.5 Data on Employment under National RuralEmployment Guarantee ActEmployment under National Rural Employment Guarantee Act (NREGA) are alsohousehold level regressions. The variables come from the 66th (2009-10) and 68th(2011-12) rounds of National Sample Survey. The dependent variable is a dummyvariable equal to 1 if any member of the household got work under NREGA in the365 days preceding the date of survey. The demographic and political controlsused are similar to the ones used for the PDS regressions. These are dummiesindicating the caste status of the household, a Muslim dummy, a dummy for ru-ral area of residence, proportion of individuals belonging to different educationalcategories13, average age of household members and its square, a dummy indicat-ing the sex of the household head, land possessed by the household, householdsize, the proportion of Congress, Left and Hindu legislators in the assembly fromthe district. The regressions estimating the impact of SC reservation additionallyinclude census and current figures of Scheduled Caste population in a district as a13. Non literates, literates but less than primary education, primary but not completed middleschool, completed middle school but less than secondary education, completed secondary educa-tion but not graduates and graduates and above22share of the total state Scheduled Caste population and the regressions estimatingthe impact of ST reservation include census and current share of Scheduled Tribepopulation in the total district population as controls. The summary statistics ofthe variables used in the NREGA participation is presented in Table 2.4.2.4 Scheduled Caste ReservationsThis section looks at the effect of Scheduled Caste reservations.Section 2.4.1 dis-cusses the empirical strategy, 2.4.2 discusses the results, 2.4.3 contains robustnesschecks and 2.4.4 contains a short discussion of the results.2.4.1 Empirical StrategySince we are interested in estimating the effect of SC reservation on primarily SCoutcomes we would like to estimate an equation of the following natureYidst =αd + τt +ψst +β1SC Reservationdst +β2SC Reservationdst ×SCidst+β3SCidst +λXidst +uidst(2.1)where Yidst is the outcome variable for individual/household i in district d of states at time t, SCReservationdst is the proportion of assembly constituencies in adistrict which are reserved at time t and SCidst is a dummy variable taking thevalue of 1 if the household is a SC household. αd , τt and ψst denote district, timeand state-time fixed effects. Xidst includes a set of controls. Inclusion of districtfixed effects ensures that our identification comes from within-district variation23and state-time fixed effects controls for any policies which can vary at the stateand across time.The main coefficient of interest is β2 since it estimates the effect of SC leg-islator on SC household. However estimating equation Equation (1) would notprovide the causal effect of SC political representation. This is because theremight be omitted variables which determine reservation proportion of the districtand are correlated with the outcome variable. Although I include different fixedeffects (district, time and state-time) which partially address this concern yet theremight still be variables varying across district and time which might bias my re-sults.To address the above concern I use the policy rule that determines reservationfor SC’s in state legislative assemblies in India. The constitution of India man-dates that reservation for Scheduled Castes in each state to be proportional to thepopulation share of SC’s in that state. The Delimitation Commission followingthis constitutional principle sets apart a number of constituencies proportional tothe state’s SC population proportion for reservation.For reservation of constituen-cies within a state the Delimitation Act states the following:constituencies in which seats are reserved for the Scheduled Castesshall be distributed in different parts of the State and located, as far aspracticable, in those areas where the proportion of their population tothe total is comparatively largeTo implement the above provision, the Delimitation Commission allocates to24each district a number of reserved constituencies. This number is determined bythe share of the district’s SC population of the total SC population of the statewithin which the district is located.14To understand this rule let us take the example illustrated in Fig 2.1. The figureillustrates the case of the state of Gujarat and the determination of reservationbased on the 2001 census done by the last Delimitation Commission. The Gujaratstate legislative assembly has a total of 182 assembly seats and approximately 7percentage SC population. Hence the total number of seats reserved for SCs are13 seats. Within Gujarat the figure gives the example of three districts-Kachchh,Anand and Surat. Based on the district’s SC population share of the total stateSC population, Kachchh gets 1 reserved seat (rounded up from 0.67), Surat gets 0reserved seat (rounded down from 0.36) and Anand gets 1 seat (rounded up from0.61). As the last column shows that there was no change in existing number ofseats for Kachchh but the number of seats changed for Surat and Anand.Given the above policy rule we can directly control for the district SC popu-lation share of the state SC population which determines the number of reservedconstituencies in a district. We can control for this variable without this variablebecoming perfectly collinear with the reserved proportion in a district because asillustrated in figure 1 the number of reserved seats can only be a whole numberand not a fraction.We can exploit other features of the policy rule that lets us control for current14. Within a district the constituencies with the highest share of Scheduled Caste population getsreserved.25population share of SC. The population numbers that the Delimitation Commis-sion uses are that of the last census and hence would not be equal to the popu-lation figures for intercensal years which we can control directly.Moreover thereis usually a lag between the sitting of a Delimitation Commission and the imple-mentation of its directives by the Election Commission.Figure 2.2 illustrates illustrates the above feature of the policy rule. Considerthe hypothetical state which holds elections in 1972 and 1977. Since the Delimita-tion Commission which was set up in 1972 came up with its findings in 1974, the1972 election will have reservation of constituencies based on the 1961 censuspopulation figures. However the 1977 election will have reservation proportionbased on the population figures of 1971 census (Chin and Prakash (2011)).Given the above features of the policy rule we estimate an equation of thefollowing form:Yidst =αd + τt +ψst +β1SC Reservationdst +β2SC Reservationdst ×SCidst+β3SCidst +δ1Census Prop SCdst +δ2Current Prop SCdst +λXidst +uidst(2.2)whereCensus Prop SCdst =SC population in the district d in thecensus year used in last delimitationSC population in the state s in thecensus year used in last delimitation26Current Prop SCdst =Current SC population in the district dCurrent SC population in the stateThe set of controls Xidst includes various economic and demographic char-acteristics of the household such as rural dummy, religion, land possessed andhousehold size. I also include political controls which account for the proportionof constituencies held by the different political formations in India15.2.4.2 ResultsTables 2.5-2.8 present the baseline results. The tables show the impact of the pro-portion of assembly seats reserved for Scheduled Castes in a district on individualand household level outcomes. The outcome variables considered in this analy-sis are primary education completion of individuals, infant mortality, per capitahousehold food-grains consumption under Public Distribution System, a dummyindicating whether the household consumed any PDS item and whether any mem-ber of the household got employment under National Rural Employment Guaran-tee Act (NREGA) in the last 365 days from the survey date.The tables show thedifferential impact on members of Scheduled Castes from political reservations.This impact which is captured by the interaction between the Scheduled Caste15. I divide up the political parties in India as Congress which includes the Indian NationalCongress that has been in power for most of the time since India’s independence and its allies,Hindu which includes the right wing Hindu nationalist party BJP, its predecessor Jan Sangh andfraternal organizations like the Hindu Mahasabha, Left which includes the various Communistparties and its allies and Others which includes the old Janata Party (an alliance of anti-Congressparties set up in the wake of the Emergency rule imposed by Congress Prime Minister, IndiraGandhi in the 1970s), its breakaway groups and other smaller parties and independents.27dummy and the Scheduled Caste political representation is our main variable ofinterest.EducationTable 2.5 presents the estimates of equation (2) where the dependent variable is adummy indicating whether the individual completed primary education. I restrictthe sample to individuals aged 14 and older to ensure full exposure to primaryeducation completion16.The politicians in power during primary schooling years are likely to affectthe likelihood of completing primary education. It is also likely that the effectsof policies such as teacher training programs, building new schools can only befelt with a time lag. Given that a child generally attends primary school betweenthe age of 6 and 13, I have calculated the average proportion of assembly seatsreserved for Scheduled Castes in the district corresponding to the period when theindividual was 6 to 13 years old lagged by two years. Thus I have calculated theaverage proportion of Scheduled Caste reservation when the individual was 4 to11 years old17.The first column of Table 2.5 shows that the average proportion of seats re-served for Scheduled Castes do not have any significant impact on primary edu-cation completion. However it is seen from column 2 that Scheduled Caste reser-vation significantly reduces the probability of primary education completion for16. The sample is thus restricted to individuals who are old enough to have completed primaryeducation.17. The results are shown to be robust to the inclusion of the average proportion of ScheduledCaste reservation between the age 5 and 12 (lagged one period)28Scheduled Castes.Infant MortalityTable 2.6 shows the impact of Scheduled Caste reservation on infant mortality.The dependent variable is a dummy indicating whether the indicator child died bythe age of 12 months. Since infant mortality depend on policies during the yearbefore birth, following Bhalotra et al. (2014), I match the infant mortality data tothe share of Scheduled Caste reservation in the year before birth.Column 1 shows the average impact of Scheduled Caste reservation on infantmortality. As with education, the impact of Scheduled Caste reservation on in-fant mortality is insignificant. Again similar to the education results, column 2shows that higher proportion of Scheduled Caste reservation in district increasesthe mortality risk of children belonging to Scheduled Castes households.Public Distribution SystemThe impact of Scheduled Caste reservation on the delivery of public distribu-tion system is shown in Table 2.7. The dependent variables are household levelper capita amount of food-grains consumed under PDS (Column 1 and 2) anda dummy variable indicating whether the household consumed any PDS item.I match PDS variable with Scheduled Caste reservation variable lagged by twoyears.18Column 1 and 2 of Table 2.7 show the impact of Scheduled Caste reserva-18. In the robustness check section (section 2.4.3) I show that the results are robust to makingthe reservation variable lag by one year instead of two years.29tion on the per capita food-grains consumed under PDS. Column 1 shows thatScheduled Caste reservation has no average effect on the amount of food-grainsconsumed under PDS. Column 2 shows that higher proportion of assembly seatsreserved for Scheduled Castes in a district reduces per capita food-grains con-sumed under PDS for Scheduled Caste households.Column 3 and 4 of Table 2.7 demonstrate the impact of Scheduled Caste reser-vation on PDS participation. Column 3 shows that proportion of seats reserved forScheduled Castes in a district has no statistically significant impact on PDS par-ticipation of the average household. However, again, higher proportion of Sched-uled Caste legislators significantly reduces the probability of PDS participation ofScheduled Caste households.NREGATable 2.8 shows the impact of Scheduled Caste reservation on NREGA partici-pation. The dependent variable is a dummy variable indicating whether at leastone member of the household got work under NREGA in the last 365 days. Theproportion of Scheduled Caste legislators is lagged by two years to allow for thelagged effects of policies.Column 1 shows that Scheduled Caste reservation has no average impact onNREGA participation. However higher proportion of Scheduled Caste reservedseats in the districts significantly increases NREGA participation of ScheduledCaste households.302.4.3 Robustness ChecksI have modified my baseline specification in a number of ways to check the ro-bustness of my results. Tables 2.9-2.13 presents these results.Table 2.9 shows the robustness checks for primary education. The first col-umn of Table 2.9 presents the baseline result for primary education and is thesame as column 2 of Table 2.5. Column 2 includes the square of average districtScheduled Caste population as a proportion of the total Scheduled Caste popula-tion in the state19 in addition to the covariates mentioned in the empirical strategysection (section 2.4.1). In column 3, I have included the average district shareof Scheduled Caste population as a control. Column 4 allows the population,demographic and political controls20 to differ across Scheduled Castes and nonScheduled Castes. Thus I have included the interaction of population and de-mographic controls with the Scheduled Caste dummy as additional controls incolumn 4. Comparing columns 2, 3 and 4 with the baseline result given in column1, we can see that the results presented in columns 2 , 3 and 4 are similar to thebaseline result presented in column 1.In column 5 of Table 2.9, the reservation variable has been lagged by oneperiod instead of two periods. Thus, I have matched the primary education com-pletion dummy to the average proportion of Scheduled Caste reservation in thedistrict when the individual was 5-12 years old. The results are again similar to19. Average for the period when the individual was 4-11 years old20. These include rural area of residence, land possessed by the household, average census andcurrent district Scheduled Caste population as a proportion of the total state Scheduled Castepopulation, average proportion of Congress, Hindu and Left parties31the baseline results.In column 6 of Table 2.9, I have restricted my sample to individuals who wereexposed to Scheduled Caste reservation before 1993. In 1993, the 73rd and 74thconstitutional amendments required all states to form local governments whichwere to be elected every 5 years. However, the 1993 amendments left legislativedetails to the states since local government remained in the State List. In additionthe states in general have chosen to provide limited revenue autonomy to localgovernments, especially rural bodies. It can still be argued that the role of the stategovernment was particularly relevant before 1993 when local governments didnot have constitutional status and local affairs were entirely within state’s sphere.So I have estimated my results for the sub-sample consisting of observations onScheduled Caste reservation before 1993 and the result is presented in column 6.We can see that the results given in column 6 are similar to the baseline results.Table 2.10 shows the robustness checks for infant mortality. Column 1 presentsthe baseline result which is similar to column 2 of Table 2.6. In column 2, I haveincluded the square of district Scheduled Caste population as a share of the totalstate Scheduled Caste population as an additional covariate. Column 3 includesthe district share of Scheduled Caste population as a control. In column 4, I haveadded the interaction of the demographic and political controls with ScheduledCaste dummy as controls. Column 5 of Table 2.10 presents the results estimatedfor the children who are exposed to Scheduled Caste reservation before 1993, theyear when local government legislation was brought in the federal parliament. Itcan be seen that the results presented in columns 2, 3, 4 and 5 of Table 2.10 are32similar to the baseline result.Table 2.11 shows the robustness checks for the amount of food-grains con-sumed under PDS. Column 1 again shows the baseline results and is similar tocolumn 2 of Table 2.7. Column 2 and 3 include square of district ScheduledCaste population as a share of the total state Scheduled Caste population and dis-trict share of Scheduled Caste population respectively as controls. In column 4,I have added the interaction of Scheduled Caste dummy with the political anddemographic variables as controls. In column 5, the reservation variable and thepolitical controls are lagged by one period. The results presented in columns 2, 3,4 and 5 are similar to the result presented in column 1.PDS food grains are provided at a discounted rate for poor households. House-holds are first classified into three groups: general, poor (those who are belowpoverty line or BPL households) and the poorest households (households belong-ing to Antodaya category) and the subsidy varies across these categories. How-ever, the information on whether a household belongs to BPL or Antodaya cate-gories is not available for all NSS rounds. Only NSS 61 and 68 has informationon these variables21. Column 6 of Table 2.10 includes dummies for whether thehousehold belongs to BPL or Antodaya categories as additional controls. Thesample size however is smaller in this case. The results are again similar to thebaseline result.Table 2.12 shows the robustness checks for PDS participation. Column 1 is21. Round 66 which has been used in the estimation of the baseline results does not have thisinformation33similar to column 4 of Table 2.6 and presents the baseline estimates. Column 2includes the squared census Scheduled Caste population shares and column 3 in-cludes district share proportion of Scheduled Caste population as additional con-trols. I have included the interaction of the demographic and political controlswith the Scheduled Caste dummy as controls in column 4. In column 5, the reser-vation variable and the political controls are lagged by one period. Column 6includes dummies for BPL and Antodaya households and thus the regression isestimated for a smaller sample as noted above for Table 2.11. The results pre-sented in columns 2-6 are again similar to the the column 1.Table 2.13 shows the robustness checks for NREGA participation. Column1 again presents the baseline result. Column 2 and 3 includes square of censusScheduled Caste population shares and district population shares of ScheduledCastes respectively as additional controls. Column 4 includes the interaction be-tween the Scheduled Caste dummy and the demographic and political controls.The results in columns 2-4 are similar to the baseline result.2.4.4 DiscussionThe results show that compared to non-Scheduled Castes, Scheduled Castes haveworse educational and health outcomes in districts with high proportion of stateassembly seats reserved for Scheduled Castes. Scheduled Castes are also morelikely to get less food grains under the Public Distribution System in these districtscompared to the non-Scheduled Castes. However, Scheduled Castes are morelikely to get employment under NREGA as compared to non-Scheduled Castes.34The results show that a one unit increase in the proportion of SC candidatereduces the relative probability of completing primary education by 0.196 pointsfor SCs. One way of interpreting the estimates is by focusing on differences be-tween districts having least and most proportion of seats reserved for SCs. The90th percentile of proportion of SC reservation is 0.25 while the 10th percentileof proportion of SC reservation is 0.03. Therefore, a move from the 10th to the90th percentile induces an effect equal to 0.22 times the coefficient on the inter-action of SC with the SC reservation variable. Thus a move from 10th to 90thpercentile reduces the relative probability of primary education completion forSCs by about 0.043 which is 6.1% of the sample mean. Similarly a shift from10th to 90th percentile of SC reservation increases the relative infant mortalityrisk of SCs by about 0.016 which is 17.2% of the sample mean, reduces per capitafood grains consumption from PDS by 0.3 which is 15.7% of the sample meanand reduces the probability of PDS participation by 0.04 which is about 6% ofthe sample mean. On the other hand a movement from 10th to 90th percentile ofSC reservation would increase the relative probability of a SC household gainingemployment under NREGA by about 0.0735 which is 30% of the sample mean.While improvements in education, health and public distribution system re-quires broad based changes, employment under NREGA can be more preciselytargeted. Keefer and Khemani (2005) argues that politicians especially in de-veloping countries find it difficult to provide broad based public services as in-formational constraints makes it difficult for politicians to signal their effort byproviding broad based public goods. It becomes difficult for voters to attribute the35improvement in quality of the broad based public goods to politicians since theyinvolve input from a number of authorities.These problems get exacerbated in the presence of social divisons. In a polar-ized society like India voters tend to find credible the promises made by candidatesbelonging to their own ethnic group (Keefer and Khemani (2005)). These tensionswould be further heightened in a reserved constituency where the only candidatesare from lower castes. Upper caste voters living in a caste ridden society would becontemptuous of lower caste candidates and would vote based on party preferencerather than vote according to candidate ability. In such an environment lower castepoliticians would prefer providing public services which can be narrowly targetedto members of their own ethnic group.This kind of narrow targeting would harm lower caste voters more since giventheir initial disadvantaged status any improvements in public services would im-prove their conditions more than members of upper caste. In absence of effectivepublic services, upper caste households can substitute towards more private alter-natives such as private schools 22, private tuition, private healthcare and increasedparental inputs. For PDS delivery, Thorat and Lee (2005) mentions that lowercaste households are often residual claimants on food stocks that reach the PDSshops. Hence any worsening of the running of the PDS system such as increasedleakage across the distribution chain would lead to decreased consumption forlower caste households. This would explain why we see that in districts with a22. Goyal and Dre`ze (2003) notes that children enrolled in government schools in India mostlybelong to disadvantaged backgrounds.36higher proportion of seats reserved for Scheduled Castes, Scheduled Castes doworse than non-Scheduled Castes in terms of education, health and PDS access.However, if we consider benefits under targetable government programs like em-ployment under NREGA, it can be seen that the Scheduled Castes are more likelyto get employment compared to the non Scheduled Castes in districts with higherproportion of assembly seats reserved for the Scheduled Castes.2.5 Scheduled Tribe ReservationsThe Constitution of India mandates reservation for Scheduled Tribes in legislativeassemblies of the states in proportion to their population shares. However un-like the case of Scheduled Castes, constituencies that are to be reserved for eachdistrict are not separately determined. The Delimitation Act states that:constituencies in which seats are reserved for the Scheduled Tribesshall, as far as practicable, be located in areas where the proportionof their population to the total is the largest.Thus seats for Scheduled Tribes are to be reserved in the constituencies in whichthe percentage of their population to the total population is the largest. Thereforethe constituencies to be reserved for Scheduled Tribes will be those where thepercentage of the ST population to the total population of the constituency is thelargest, in descending order equal to the number of constituencies to be reservedfor Scheduled Tribes in that state. Given this criteria, the fraction of constituenciesin a district that end up getting reserved for Scheduled Tribes will be a function of37the proportion of Scheduled Tribes in that district. Hence I estimate an equationof the following form to analyze the effects of Scheduled Tribes:Yidst =αd + τt +ψst +β1ST Reservationdst +β2ST Reservationdst ×STidst+β3STidst +δ1Census Percent STdst +δ2Current Percent STdst +λXidst +uidst(2.3)whereCensus Percent STdst =ST population in the district in thecensus year used in last delimitationTotal population in the district in thecensus year used in last delimitationCurrent Percent STdst =Current ST population in the districtTotal current population in the districtwhere Yidst is the outcome variable for individual/household i in district d ofstate s at time t, ST Reservationdst is the proportion of assembly constituencies ina district which are reserved for Scheduled Tribe at time t and STidst is a dummyvariable taking the value of 1 if the household is a ST household. αd , τt and ψstdenote district, time and state-time fixed effects. Xidst includes a set of controlswhich as before include political controls which account for the proportion of con-stituencies held by the different political formations in India and various economicand demographic characteristics of the household such as rural dummy, religion,land possessed and household size.38However the strategy outlined above might give us biased estimates of β1 andβ2. This is because unlike the case of Scheduled Caste reservation there is nounique rule determining the fraction of constituencies that are to be reserved forScheduled Tribes in the district. As the above discussion shows the fraction ofreserved constituencies in a district depends on the census population share ofScheduled Tribes of that district (Census Percent ST)which we control for butit does not uniquely determine it. For example my regressions might run intotrouble if for example two districts having the same population share of ScheduledTribes have different fractions of reserved constituencies for Scheduled Tribes ifthe districts are very different in the way Scheduled Tribes are spatially distributedwithin the districts. For a given proportion of Scheduled Tribes in a district, thefraction of constituencies that gets reserved within that district depends on thespatial concentration of Scheduled Tribes.If Scheduled Tribes are concentratedwithin a small area in a district it gets higher fraction of reserved constituenciescompared to the case where Scheduled Tribe population is dispersed within adistrict. This would lead to an omitted variable bias if the spatial distributionof Scheduled Tribes within a district has an independent effect on the outcomevariables. This problem is taken care of to an extent by the inclusion of districtfixed effects but if the spatial distribution of Scheduled Tribes changes over time itmight lead to biased results. Hence the results in this section should be interpretedwith caution.Table 2.14 shows the impact of the proportion of assembly seats reserved forScheduled Tribes in a district on individual and household level outcomes. Col-39umn 1 presents the estimates of equation (3) where the dependent variable is adummy indicating whether the individual completed primary education. There isno significant effect of the average proportion of seats reserved for STs on pri-mary education completion. Column 2 shows the impact of ST reservation oninfant mortality. Again there is no significant effect of the proportion of seatsreserved for STs on infant mortality. The impact of ST reservation on the de-livery of public distribution system is shown in Columns 3 and 4. The resultsshow that ST reservation has no significant impact on the per-capita food-grainsconsumption. However, ST reservation variable increases PDS participation ofScheduled Tribes. Column 5 shows the impact of Scheduled Tribe reservation onNREGA participation. It can be seen that ST reservation has no significant effecton NREGA participation.The results for Scheduled Tribes show that except for the indicator variablesdenoting PDS participation, Scheduled Tribe legislators have no significant effecton members of their own ethnic group. This is in contrast to the results that Igot for Scheduled Castes. The differences in the nature of political mobilizationbetween Scheduled Castes and Scheduled Tribes might be a source for these dif-ference in results. Banerjee and Somanathan (2007) mentions that there has beena dearth of independent political mobilization among Scheduled Tribes in con-trast to the experience of Scheduled Castes. Specifically since the 1980s there hasbeen increased political mobilization of Scheduled Castes which has led to a risein an independent political leadership among Scheduled Castes which relies to alarge extent on the Scheduled Caste electorate for their electoral success. This40political mobilization has led to the Scheduled Castes having emerged as a strongpolitical bloc with the tendency to vote as a bloc and thus more effectively swaythe political fortunes of individual politicians. In the absence of similar politicalmobilization among Scheduled Tribes, party identity will matter more than legis-lator identity and thus we would not see any effect of individual legislator identitywhich my reservation variables are picking up.2.6 ConclusionElectoral incentives often determine whether a politician invests in broad basedpublic good or narrowly targeted government programs. Depending on the politi-cal preferences, policies adopted by the politicians can have long term impact onthe lives of individuals. This is particularly true for the disadvantaged sections indeveloping countries who are highly sensitive to policy fluctuations due to theirvulnerable economic position.This chapter shows that higher proportion of seats reserved for the ScheduledCastes negatively affects the educational, health and PDS consumption of food-grains for Scheduled Castes. However reservations also lead to Scheduled Castesbeing more likely to get employment under NREGA. It can be argued that theScheduled Caste politicians can signal their ability to the Scheduled Caste votersby credibly committing the delivery of targeted government programs like gov-ernment jobs. However education, health and distribution of items under publicdistribution system cannot be easily targeted. Thus Scheduled Caste politiciansare likely to invest less in these broad based public goods and Scheduled Caste41voters, who are most likely to benefit from the investment in these public goodsare likely to suffer more from the under-provision of these broad based publicservices. Finally the absence of any comparable effects for Scheduled Tribe leg-islators point to heterogenous effects of reservations for disadvantaged groups.Future work would be aimed at trying to provide more direct evidence of thisbroad vs narrow based distinction in the supply of public goods and services byminority politicians.42Figure 2.1: Reservations for GujaratNotes: The figure gives an example of the determination of reservation based on 2001 census done by the last Delimitation Commissionfor the Kachchh, Anand and Surat districts of Gujarat.43Figure 2.2: Time lag in Reservations1967census2001197719761971 1967-1976Seats based on 1961 censuscensus20071977-2007Seats based on 1971 census20122008-2012Seats based on 2001 census2008..........1961census....Notes: The figure illustrates the time lag in the determination of reservation of constituencies.44Table 2.1: Summary Statistics: EducationVariable Mean Std. Dev. NPrimary education completion 0.702 0.457 174632Average proportion of SC reservation 0.151 0.083 174632Average proportion of ST reservation 0.077 0.182 174632SC household 0.191 0.393 174632ST household 0.083 0.276 174632Muslim household 0.136 0.343 174632Female 0.494 0.5 174632Rural area of residence 0.704 0.456 174632Land possessed 595.9 1400.3 174632Average SC census population share 0.061 0.045 174632Average SC current population share 0.062 0.046 174632Average current district population share of SC 0.161 0.071 174632Average census district population share of ST 0.071 0.136 174632Average current district population share of ST 0.08 0.132 174632Average proportion Congress 0.38 0.241 174632Average proportion Hindu 0.161 0.209 174632Average proportion Left 0.102 0.216 174632Notes: The table reports the mean and standard deviation of variables used in the primaryeducation regressions. Primary education completion dummy, SC, ST, Muslim, femaleand rural dummies and land possessed by the household come from the 68th round ofNSS. The reservation variables, the population proportions and the proportion of politicalparties are averaged over the years an individual attained primary school, lagged by twoyears. Thus the average is taken for the years when the individual was between 4 and11 years old. Average proportion of SC and ST reservation and proportion of seatsobtained by Congress, Hindu and Left parties are obtained from the official website ofthe Election Commission of India. The population proportion variables come from theCensus of India.45Table 2.2: Summary Statistics: Infant MortalityVariable Mean Std. Dev. NInfant mortality 0.0918 0.2887 212501Proportion of SC reservation 0.1503 0.0805 212501Proportion of ST reservation 0.0754 0.1825 212501SC 0.1982 0.3986 212501ST 0.0897 0.2857 212501Muslim 0.8555 0.3516 212501Rural area of residence 0.7694 0.4212 212501Multiple birth 0.0135 0.1155 212501Female child 0.4794 0.4996 212501Mother’s years of schooling 2.5521 3.9466 212501Father’s years of schooling 5.3532 4.8998 212501Female headed household 0.0783 0.2687 212501Mother not working 0.5961 0.4907 212501Mother works at home 0.0666 0.2493 212501Mother works away 0.3374 0.4728 212501Mother’s age at birth 23.246 5.0672 212501SC census population share 0.0613 0.0492 212501SC current population share 0.0618 0.0499 212501Current district population share of SC 0.1598 0.0671 212501Census district population share of ST 0.0700 0.1374 212501Current district population share of ST 0.0780 0.1348 212501Proportion Congress 0.4340 0.3165 212501Proportion Hindu 0.1224 0.2136 212501Notes: The table reports the mean and standard deviation of variables used inthe infant mortality regressions. Infant Mortality, SC, ST, Muslim, female, ru-ral, multiple birth, female headed household, mother’s work status dummies,mother’s and father’s years of schooling and mother’s age at birth comes fromNFHS II. Proportion of SC and ST reservation and proportion of seats obtainedby Congress, Hindu and Left parties comes from the official website of the Elec-tion Commission of India. The population proportion variables come from theCensus of India.46Table 2.3: Summary Statistics: PDS Participation and Food GrainsVariable Mean Std. Dev. NPer capita food-grains from PDS 1.9182 3.1923 256763PDS participation 0.7180 0.4500 256763Proportion of SC reservation 0.1535 0.0821 256763Proportion of ST reservation 0.0794 0.1836 256763SC household 0.1966 0.3975 256763ST household 0.0858 0.2800 256763Muslim household 0.1153 0.3194 256763Proportion of educational groups in householdNon literates 0.3514 0.3234 256763Literates but below primary 0.1583 0.2095 256763Primary but below middle 0.1362 0.1985 256763Middle but below secondary 0.1396 0.2032 256763Secondary without graduate degree 0.1515 0.2341 256763Graduate and above 0.0630 0.1784 256763Female headed household 0.1135 0.3171 256763Average age of members 29.641 12.604 256763Household size 4.5661 2.2987 256763Land possessed 562.13 3498.8 256763Rural area of residence 0.7150 0.4514 256763SC census population share 0.0619 0.0454 256763SC current population share 0.0627 0.0459 256763Current district population share of SC 0.1651 0.0722 256763Census district population share of ST 0.0737 0.1324 256763Current district population share of ST 0.0814 0.1298 256763Proportion Congress 0.2950 0.2477 256763Proportion Hindu 0.2284 0.2575 256763Proportion Left 0.0973 0.2291 256763BPL households 0.3515 0.4774 140200Antodaya households 0.0445 0.2063 140200Notes: The table reports the mean and standard deviation of variables used in thePDS regressions. Per-capita food-grains consumption under PDS, PDS partici-pation dummy, SC, ST, Muslim, female headed household and rural dummies,proportion of educational groups in household, household size, average age ofhousehold members and land possessed by the household come from the 61st,66th and 68th rounds of NSS. Proportion of SC and ST reservation and pro-portion of seats obtained by Congress, Hindu and Left parties comes from theofficial website of the Election Commission of India. The population proportionvariables come from the Census of India. The dummies indicating whether thehousehold belongs to BPL or Antodaya categories is obtained from the 61st and68th rounds of NSS. This information is not available in the 66th rounds of NSS.47Table 2.4: Summary Statistics: NREGA ParticipationVariable Mean Std. Dev. NNREGA participation 0.236 (0.425) 89607Proportion of SC reservation 0.16 (0.082) 89607Proportion of ST reservation 0.085 (0.19) 89607SC household 0.216 (0.412) 89607ST household 0.106 (0.307) 89607Muslim household 0.108 (0.311) 89607Proportion of educational groups in householdNon literates 0.389 (0.319) 89607Literates but below primary 0.177 (0.219) 89607Primary but below middle 0.145 (0.2) 89607Middle but below secondary 0.136 (0.194) 89607Secondary without graduate degree 0.123 (0.201) 89607Graduate and above 0.03 (0.113) 89607Female headed household 0.118 (0.322) 89607Average age of members 30.06 (13.21) 89607Household size 4.524 (2.188) 89607Land possessed 687.7 (1453.3) 89607SC census population share 0.06 (0.045) 89607SC current population share 0.061 (0.046) 89607Current district population share of SC 0.171 (0.072) 89607Census district population share of ST 0.08 (0.136) 89607Current district population share of ST 0.087 (0.136) 89607Proportion Congress 0.264 (0.234) 89607Proportion Hindu 0.23 (0.264) 89607Proportion Left 0.109 (0.249) 89607Notes: The table reports the mean and standard deviation of variables used in the MN-REGA regressions. NREGA participation dummy, SC, ST, Muslim, female headedhousehold and rural dummies, proportion of educational groups in household, house-hold size, average age of household members and land possessed by the householdcome from the 66th and 68th rounds of NSS. Proportion of SC and ST reservation andproportion of seats obtained by Congress, Hindu and Left parties comes from the offi-cial website of the Election Commission of India. The population proportion variablescome from the Census of India.48Table 2.5: SC Reservation:Primary Education(1) (2)Average Proportion of SC Reservation * SC -0.196∗∗(0.0990)Average Proportion of SC Reservation -0.0904 -0.00795(0.185) (0.188)SC -0.0944∗∗∗ -0.0803∗∗∗(0.00841) (0.0165)Observations 174632 174632r2 0.247 0.257Notes: All regressions include district, year and state-year fixed effects, dum-mies for Scheduled Caste, Scheduled Tribe, Muslim, rural area of residence,land possessed by the household. Other controls include census and current dis-trict SC population as a proportion of the total state SC population, proportionof Congress, Hindu and Left parties. Standard errors are clustered at districtlevel and displayed in parentheses. * denotes significant at 10%; ** denotessignificant at 5% and *** denotes significant at 1%.49Table 2.6: SC Reservation:Infant Mortality(1) (2)Proportion of SC Reservation * SC 0.0719∗∗(0.0345)Proportion of SC Reservation -0.0304 -0.0437(0.0428) (0.0433)SC 0.00532∗∗ -0.00654(0.00256) (0.00625)Observations 212501 212501r2 0.0459 0.0460Notes: All regressions include district, year of birth, state-year ofbirth fixed effects, dummies for month of birth, order of birth, sex ofthe child, multiple birth, Scheduled Caste, Scheduled Tribe, Muslim,rural area of residence, sex of the household head, mother’s workstatus, mother’s and father’s education and mother’s age at birth.Other controls include census and current district SC population as aproportion of the total state SC population, proportion of Congress,Hindu and Left parties. Standard errors are clustered at district leveland displayed in parentheses. * denotes significant at 10%; ** de-notes significant at 5% and *** denotes significant at 1%.50Table 2.7: SC Reservation:Public Distribution System(1) (2) (3) (4)PDS Food Grains PDS ParticipationProportion of SC Reservation * SC -1.279∗∗∗ -0.199∗∗∗(0.428) (0.0577)Proportion of SC Reservation 0.163 0.410 0.0282 0.0666(0.753) (0.755) (0.139) (0.138)sc 0.578∗∗∗ 0.801∗∗∗ 0.0373∗∗∗ 0.0719∗∗∗(0.0348) (0.0809) (0.00508) (0.0104)Observations 256763 256763 256763 256763r2 0.283 0.284 0.301 0.302Notes: All regressions include district fixed effect, dummies for round, state-round fixed ef-fect, dummies for Scheduled Caste, Scheduled Tribe, Muslim, rural area of residence, sex ofthe household head, average education of household members, average age and its square,land possessed by the household. Other controls include census and current district SC popu-lation as a proportion of the total state SC population, proportion of Congress, Hindu and Leftparties. Standard errors are clustered at district level and displayed in parentheses. * denotessignificant at 10%; ** denotes significant at 5% and *** denotes significant at 1%..51Table 2.8: SC Reservation:NREGA Participation(1) (2)Proportion of SC Reservation * SC 0.369∗∗∗(0.111)Proportion of SC Reservation -0.0514 -0.112(0.261) (0.261)sc 0.117∗∗∗ 0.0493∗∗(0.00965) (0.0210)Observations 89607 89607r2 0.225 0.226Notes: All regressions include district fixed effect, dummies forround, state-round fixed effect, dummies for Scheduled Caste,Scheduled Tribe, Muslim, rural area of residence, sex of the house-hold head, average education of household members, average ageand its square, land possessed by the household. Other controls in-clude census and current district SC population as a proportion ofthe total state SC population, proportion of Congress, Hindu andLeft parties. Standard errors are clustered at district level and dis-played in parentheses. * denotes significant at 10%; ** denotessignificant at 5% and *** denotes significant at 1%.52Table 2.9: SC Reservation: Robustness Checks: Primary Education(1) (2) (3) (4) (5) (6)Non Linear District Interaction of PreCensus share covariates with 1993Baseline population of SC SC Dummy Lag 1 SampleAverage Proportion of SC Reservation * SC -0.196∗∗ -0.196∗∗ -0.196∗∗ -0.221∗∗ -0.197∗∗ -0.299∗∗(0.0990) (0.0990) (0.0991) (0.105) (0.0993) (0.133)Average Proportion of SC Reservation -0.00795 -0.00786 -0.0165 -0.00556 -0.0796 0.0881(0.188) (0.188) (0.191) (0.188) (0.164) (0.272)SC -0.0803∗∗∗ -0.0803∗∗∗ -0.0802∗∗∗ -0.0567∗ -0.0828∗∗∗ -0.0993∗∗∗(0.0165) (0.0165) (0.0165) (0.0295) (0.0166) (0.0233)Observations 174632 174632 174632 174632 179930 70290r2 0.257 0.257 0.257 0.258 0.260 0.258Notes: All regressions include district, year and state-year fixed effects, dummies for Scheduled Caste, Scheduled Tribe, Muslim, rural areaof residence, land possessed by the household. Other controls include census and current district SC population as a proportion of the totalstate SC population, proportion of Congress, Hindu and Left parties. Column 1 gives the baseline results. Column 2 includes the square of SCcensus population share as control. Column 3 includes district share of SC as control. In Column 4 I have included the interaction between theSC dummy and the political and demographic covariates. In column 5 the reservation variable and the political controls are lagged one period.Column 6 shows the baseline results for the restricted sample for pre 1993 period. Standard errors are clustered at district level and displayed inparentheses. * denotes significant at 10%; ** denotes significant at 5% and *** denotes significant at 1%..53Table 2.10: SC Reservation:Robustness Checks: Infant Mortality(1) (2) (3) (4) (5)Non Linear District Interaction of PreCensus share covariates with 1993Baseline population of SC SC Dummy SampleProportion of SC Reservation * SC 0.0719** 0.0719** 0.0719** 0.0745** 0.0879**(0.0345) (0.0345) (0.0345) (0.0347) (0.0407)Proportion of SC Reservation -0.0437 -0.0430 -0.0439 -0.0445 -0.0508(0.0433) (0.0434) (0.0432) (0.0433) (0.0457)SC -0.00654 -0.00653 -0.00654 -0.00535 -0.00812(0.00625) (0.00625) (0.00625) (0.0134) (0.00760)Observations 212501 212501 212501 212501 164214r2 0.0460 0.0460 0.0460 0.0463 0.0478Notes: All regressions include district, year of birth, state-year of birth fixed effects, dummies for month of birth,order of birth, sex of the child, multiple birth, Scheduled Caste, Scheduled Tribe, Muslim, rural area of residence,sex of the household head, mother’s work status, mother’s and father’s education and mother’s age at birth. Othercontrols include census and current district SC population as a proportion of the total state SC population, proportionof Congress, Hindu and Left parties. Column 1 gives the baseline results. Column 2 includes the square of SCcensus population share as control. Column 3 includes district share of SC as control. In Column 4 I have includedthe interaction between the SC dummy and the political and demographic covariates. Column 5 shows the baselineresults for the restricted sample for pre 1993 period. Standard errors are clustered at district level and displayed inparentheses. * denotes significant at 10%; ** denotes significant at 5% and *** denotes significant at 1%.54Table 2.11: SC Reservation:Robustness Checks: PDS Food Grains(1) (2) (3) (4) (5) (6)Non Linear District Interaction of BPL,Census share covariates with AntodayaBaseline population of SC SC Dummy Lag1 DummiesProportion of SC Reservation * SC -1.279∗∗∗ -1.281∗∗∗ -1.279∗∗∗ -1.033∗∗ -1.192∗∗∗ -1.316∗∗∗(0.428) (0.427) (0.428) (0.436) (0.423) (0.393)Proportion of SC Reservation 0.410 0.526 0.365 0.348 -0.337 0.398(0.755) (0.764) (0.772) (0.763) (1.099) (0.943)sc 0.801∗∗∗ 0.801∗∗∗ 0.801∗∗∗ -0.440 0.786∗∗∗ 0.457∗∗∗(0.0809) (0.0808) (0.0809) (0.279) (0.079) (0.0766)Observations 256763 256763 256763 256763 256763 140200r2 0.284 0.284 0.284 0.286 0.283 0.511Notes: All regressions include district fixed effect, dummies for round, state-round fixed effect, dummies for Scheduled Caste,Scheduled Tribe, Muslim, rural area of residence, sex of the household head, average education of household members, averageage and its square, land possessed by the household. Other controls include census and current district SC population as aproportion of the total state SC population, proportion of Congress, Hindu and Left parties. Column 1 gives the baseline results.Column 2 includes the square of SC census population share as control. Column 3 includes district share of SC as control.In Column 4 I have included the interaction between the SC dummy and the political and demographic covariates. Column5 includes dummies for whether the household belongs to bpl or antodaya categories. Since this information is not availablefor 66th round of NSS, the sample size is smaller in this case. Standard errors are clustered at district level and displayed inparentheses. * denotes significant at 10%; ** denotes significant at 5% and *** denotes significant at 1%.55Table 2.12: SC Reservation:Robustness Checks: PDS Participation(1) (2) (3) (4) (5) (6)Non Linear District Interaction of BPL,Census share covariates with AntodayaBaseline population of SC SC Dummy Lag1 DummiesProportion of SC Reservation * SC -0.199∗∗∗ -0.199∗∗∗ -0.199∗∗∗ -0.113∗ -0.202∗∗∗ -0.177∗∗∗(0.0577) (0.0577) (0.0576) (0.0605) (0.0568) (0.0550)Proportion of SC Reservation 0.0666 0.0486 0.0591 0.0537 0.154 0.109(0.138) (0.140) (0.140) (0.138) (0.161) (0.220)SC 0.0719∗∗∗ 0.0719∗∗∗ 0.0719∗∗∗ 0.0956∗∗ 0.072 0.0442∗∗∗(0.0104) (0.0104) (0.0104) (0.0400) (0.0104) (0.0109)Observations 256763 256763 256763 256763 256763 140200r2 0.302 0.302 0.302 0.303 0.3019 0.367Notes: All regressions include district fixed effect, dummies for round, state-round fixed effect, dummies for Scheduled Caste,Scheduled Tribe, Muslim, rural area of residence, sex of the household head, average education of household members, averageage and its square, land possessed by the household. Other controls include census and current district SC population as aproportion of the total state SC population, proportion of Congress, Hindu and Left parties. Column 1 gives the baseline results.Column 2 includes the square of SC census population share as control. Column 3 includes district share of SC as control. InColumn 4 I have included the interaction between the SC dummy and the political and demographic covariates. Column 5 includesdummies for whether the household belongs to bpl or antodaya categories. Since this information is not avaiable for 66th round ofNSS, the sample size is smaller in this case. Standard errors are clustered at district level and displayed in parentheses. * denotessignificant at 10%; ** denotes significant at 5% and *** denotes significant at 1%.56Table 2.13: SC Reservation:Robustness Checks: NREGA Participation(1) (2) (3) (4)Non Linear District Interaction ofCensus share covariates withBaseline population of SC SC DummyProportion of SC Reservation * SC 0.369∗∗∗ 0.368∗∗∗ 0.369∗∗∗ 0.260∗∗(0.111) (0.111) (0.111) (0.123)Proportion of SC Reservation -0.112 -0.0786 -0.146 -0.0901(0.261) (0.265) (0.260) (0.261)sc 0.0493∗∗ 0.0496∗∗ 0.0495∗∗ -0.0735(0.0210) (0.0210) (0.0210) (0.0881)Observations 89607 89607 89607 89607r2 0.226 0.226 0.226 0.228Notes: All regressions include include district fixed effect, dummies for round, state-roundfixed effect, dummies for Scheduled Caste, Scheduled Tribe, Muslim, rural area of resi-dence, sex of the household head, average education of household members, average ageand its square, land possessed by the household. Other controls include census and cur-rent district SC population as a proportion of the total state SC population, proportion ofCongress, Hindu and Left parties. Column 1 gives the baseline results. Column 2 includesthe square of SC census population share as control. Column 3 includes district share of SCas control. In Column 4 I have included the interaction between the SC dummy and the polit-ical and demographic covariates. Standard errors are clustered at district level and displayedin parentheses. * denotes significant at 10%; ** denotes significant at 5% and *** denotessignificant at 1%.57Table 2.14: Effect of ST Reservation(1) (2) (3) (4) (5)Primary Infant PDS PDSEducation Mortality Food Grains Participation NREGAProportion of ST Reservation * ST -0.0143 0.0127 0.2428 0.1365∗∗ 0.0332(0.0493) (0.0130) (0.2326) (0.0537) (0.0435)Proportion of ST Reservation 0.375 0.0715 -0.4553 -0.0405 -0.2862(0.352) (0.0556) (1.2423) (0.1000) (0.1896)ST -0.195∗∗∗ 0.000450 0.5823∗∗ 0.0418∗∗∗ 0.1021∗∗∗(0.0167) (0.00436) (0.0781) (0.0103) (0.0186)Observations 174632 212501 256763 256763 89607r2 0.257 0.0460 0.2834 0.3018 0.2261Notes: Column 1 shows the impact of ST reservation on primary education completion. The regression includedistrict, year and state-year fixed effects, dummies for Scheduled Caste, Scheduled Tribe, Muslim, rural area of res-idence, land possessed by the household, census and current proportion of ST population in a district as a proportionof the total district population and proportion of Congress, Hindu and Left parties. Column 2 shows the impact ofST reservation on infant mortality. The regression include district, year of birth, state-year of birth fixed effects,dummies for month of birth, order of birth, sex of the child, multiple birth, Scheduled Caste, Scheduled Tribe,Muslim, rural area of residence, sex of the household head, mother’s work status, mother’s and father’s educationand mother’s age at birth, census and current proportion of ST population in a district as a proportion of the totaldistrict population and proportion of Congress, Hindu and Left parties. Column 3 and 4 shows the impact of STreservation on PDS food-grains consumption and a dummy indicating whether the household purchased any itemunder PDS. The regressions include district fixed effect, dummies for round, state-round fixed effect, dummies forScheduled Caste, Scheduled Tribe, Muslim, rural area of residence, sex of the household head, average education ofhousehold members, average age and its square, land possessed by the household, census and current proportion ofST population in a district as a proportion of the total district population and proportion of Congress, Hindu and Leftparties. Column 5 shows the impact of ST reservation on NREGA participation. The regression include includedistrict fixed effect, dummies for round, state-round fixed effect, dummies for Scheduled Caste, Scheduled Tribe,Muslim, rural area of residence, sex of the household head, average education of household members, average ageand its square, land possessed by the household, census and current proportion of ST population in a district as aproportion of the total district population and proportion of congress, Hindu and left parties. Standard errors areclustered at district level and displayed in parentheses. * denotes significant at 10%; ** denotes significant at 5%and *** denotes significant at 1%.58Chapter 3Colonization and ReligiousViolence: Evidence from IndiaTypically, imperial powers depend on the inability of oppressed localpopulations to muster a unified resistance, and the most successfuloccupiers are skilled at exploiting the differences among the occu-pied. Certainly that was the story of the British Empire’s success,and its legacy of nurtured local hatreds can be seen wherever theUnion Flag flew, from Muslim-Hindu hatred in Pakistan and India,to Catholic-Protestant hatred in Ireland, to, yes, Jew-Arab, hatred inmodern Israel. — James Carroll, Constantine’s Sword (2001)593.1 IntroductionAs the above quote suggests colonizers have often been blamed for creating riftsbetween different indigenous communities in the lands that they colonized so asto prevent any consolidation of indigenous forces against them. In the context ofIndia, this is particularly true. British colonization has often been blamed for ush-ering in an era of Hindu-Muslim communal discord after centuries of communalharmony under the Mughals. This chapter is an attempt at evaluating whether thisassertion is true. I empirically test whether British annexation has any long-termeffect on religious violence in post-Independent India controlling for selective an-nexation by the British.Hindu-Muslim religious violence has been one of the most pressing issuesin post-Independent India. According to the Varshney and Wilkinson (2006)dataset on Hindu-Muslim conflict in India there have been more than 1100 casesof Hindu-Muslim violence in India causing around 7000 deaths over the period1950-1995. In addition, riots result in substantial property damage, loss of liveli-hood and residential segregation (Field et al. (2008), Baber (2004), Mitra and Ray(2014)).I compare districts ruled directly by the British with districts ruled by the na-tive Indian rulers and see if British colonization has any long run effects on post-Independence religious violence in India. To account for potential selective an-nexation by the British I use the instrumental variable strategy used in Iyer (2010).Iyer (2010) compares public good provision across directly ruled and indirectly60ruled areas using the Doctrine of Lapse policy used by the British in annexingnative states. According to the Doctrine of Lapse policy instituted by Lord Dal-housie in 1848, the British reserved the right to annex native states whose kingsdied without leaving a natural heir. Thus one can use the death of a native kingwithout an heir in the period from 1848-1856 as an instrument for annexationby the British. Using this instrumental variable helps me to control for selectiveannexation and thus get rid of any endogenity in the variable indicating British an-nexation. Using the Doctrine of Lapse policy as an instrument, I find that contraryto the popularly held view, British ruled districts experienced lesser instances ofreligious violence compared to those ruled by native states. This result is robustto controlling for different geographic features, population and economic charac-teristics and political variables.My research contributes to the literature analyzing the causes behind reli-gious violence in India. The leading explanations for religious violence focus oneconomic and political factors. Studies have shown that greater economic com-petition between Hindus and Muslims leads to more religious violence (Kumar(2005)) and religious violence is used as a tool to usurp resources belonging tomembers of the rival religion (Mitra and Ray (2014)). On the other hand polit-ical scientists have tended to focus on political reasons behind riots. Wilkinson(2006)) shows that even after controlling for a town’s socio-economic attributesand its level of previous Hindu-Muslim violence, “electoral cycles and the level ofelectoral completion exert an independent effect on the likelihood of communalriots.” By comparing directly ruled areas with native states, this chapter adds to61the above literature by looking at the effect of the identity of the historical ruleron Hindu-Muslim violence.My research is part of the expanding literature on the role of historical insti-tutions in explaining contemporary outcomes (Acemoglu, Johnson, and Robinson(2001), Engerman and Sokoloff (1997), and La Porta et al.). In the Indian context,Iyer and Banerjee (2005) analyze how different land tenure systems established bythe British have affected long-term economic outcomes and Iyer (2010) comparespublic good provision across directly ruled and indirectly ruled areas. Howeverresearch on the role of historical institutions in explaining ethnic violence is lim-ited in the economics literature (Alesina, Easterly, and Matuszeski (2011)). Jha(2013) is one of those few papers which does so in the Indian context. The paperanalyzes the role of medieval trade in explaining Hindu-Muslim riots during theperiod 1850-1950. It argues that religious violence is reduced if Hindus and Mus-lims could share the gains of trade in the medieval period and found that medievaltrading ports were less likely to experience a religious riot between 1850-1950.My research complements the literature on the role of historical institutions onethnic violence by focusing on the role of colonial rule in explaining the postindependence Hindu-Muslim riots.This chapter is most closely tied to the significant literature in history whichanalyzes the role of the British colonizers in fomenting Hindu-Muslim conflict.Indian nationalist historians have often claimed that the British followed a “divideand rule” strategy which created rifts between communities and laid the foun-dations for later day religious violence (Mehta and Patwardhan (1942), Kabir62(1969), Das (1990)). This claim has been contested by other historians who ar-gue either that communal tensions had already been simmering before the Britishcame (Bayly (1985)) or that factors independent of British rule like pan-Islamismand the rise of Hindu and Muslim revivalist movements (Hardy (1972)) led to arise in communal discord in the colonial era. There might be other channels toothrough which British annexation might affect religious violence. The British laidthe foundations for a modern law and order machinery which, due to institutionalpersistence, might affect present religious violence in India. Moreover directlyruled British areas have a longer experience of democratic systems of governancethrough a system of directly elected government councils. This too might affectreligious violence in independent India. Olsson (2009) showed that there is astrong positive effect of colonial duration on democracy, particularly for formerBritish colonies. This too might affect religious violence in independent India.Thus given arguments on both sides, whether British rule lead to a deterioration inHindu-Muslim relations becomes an empirical question which has not been testedso far in a rigorous manner. This chapter attempts to address this gap in literature.As mentioned this chapter complements the aforementioned literature in anumber of ways. Firstly by looking at a historical institution namely colonizationit brings in a new dimension to the empirical literature on Hindu-Muslim politicalviolence which has largely focussed on economic or political causes. Secondlythis chapter adds to the growing literature on the effect of colonial institutionsby looking at one of the relatively unexplored areas in economics which is therole of historical institutions in ethnic conflict. Most importantly this chapter tries63to resolve the question that has been debated among historians whether Britishcolonization has led to increased Hindu-Muslim conflict. By controlling for se-lective annexation by the British it is able to address endogenity concerns.1 Theresults challenge the popular narrative that British colonization led to increasedHindu-Muslim conflict.The rest of the chapter is organized as follows: Section 3.2 discusses the his-torical background. It describes in detail the “divide and rule” strategy that isalleged to have been followed by the British and also some alternative channelsthrough which British rule might affect Hindu-Muslim communal tension in thelong-run, Section 3.3 describes the data used in this chapter, Section 3.4 discussesthe empirical strategy, Section 3.5 discusses the results and Section 3.6 concludes.3.2 British Colonization and Rise in ReligiousViolenceIn this section I first discuss the various measures taken by the British which areattributed by historians to the strategy of divide and rule and have been suggestedas playing a significant role in the rise of Hindu-Muslim communal discord. I thenbriefly discuss some alternative channels through which British annexation mighthave affected Hindu-Muslim religious violence in a different manner.Various accounts suggest that the British followed a divide and rule strategywhich incited religious violence and helped the British to maintain their hold over1. Lange and Dawson (2009) in a sample of 160 countries find evidence that “inter-communalviolence is a common legacy of colonialism.” However his results might be subject to endogenityconcerns common in cross-country studies. Most importantly he does not control for selectiveannexation by the British.64their Indian subjects. In the ensuing account I describe the narrative that blamesthe British for the worsening of Hindu-Muslim relations. The narrative essentiallycontends that that British policy essentially consisted of two phases-an initial pe-riod of Hindu appeasement and suppression of Muslim aspiration followed by aperiod of inciting Muslim communalism to serve as a counterweight to emergingIndian nationalism. In the ensuing account I describe the narrative that blames theBritish for the worsening of Hindu-Muslim relations.The first phase of Hindu appeasement and suppression of Muslim aspirationis considered to have consisted mainly of three measures taken by the British: thePermanent Settlement Act of 1793, the Resumption Proceedings and the abolitionof Persian and adoption of English as the official language in 1835. Under thePermanent Settlement Act of 1793, zamindars (landlords) of the Bengal provincewere granted proprietary and hereditary rights over the land and their revenueobligation to the British government were fixed in perpetuity. Some commenta-tors like Kabir (1969) claim that this system was established via a massive landtransfer from the Muslim landed gentry to the Hindu landholding class. Otherslike Hardy (1972) claim that the Permanent Settlement Act affected Muslims ad-versely by “virtual closing the door of landlordism to Muslims”. Hardy (1972)states that Hindu cultivators suffered as much as that of the Muslims cultivatorsunder the Permanent Settlement but the number of Muslim cultivators in Ben-gal at that time were greater. Moreover the moneylenders who were the lendersof the last resort for the individual cultivators to pay their rent to the landlordswere mostly Hindu-this too led to communal antagonism. In fact some of the65major Hindu-Muslim communal disturbances during the colonial era such as therebellion of Titu Mir in 1830, the Faraizi movement in the 1830s and 1840s andthe Malabar Rebellion in 1921 were essentially class struggles waged by Mus-lim cultivators against Hindu landlords and the British. The next measure that isconsidered to have affected Muslims adversely was the Resumption Regulationof 1820, under which the East India Company appropriated lakhiraj, revenue-freeland granted mainly to Muslims. These land rights had been granted by bothHindu and Muslim rulers to support learning and education (Hardy (1972)). Inorder to maximize their tax collections from land revenue, the East India Com-pany embarked on a policy which called for investigation and resumption of thoseholdings which did not possess proper title deeds. Some commentators contendthat though some Hindus were also affected by the resumption proceedings, Mus-lims were the worst hit since Muslim grantees were much larger in number thanHindu grantees and also because as the erstwhile ruling elite they did not preservetheir title deeds properly(Hardy (1972)). This gave a further blow to the Muslimmiddle and upper classes as it adversely affected their traditional educational sys-tem, which was based mostly on revenues from these grant lands and thus mighthave led to deepening of Muslim communal feelings (Kabir (1969)).The third major step of the British which is said to have resulted in the im-poverishment of the Muslims vis--vis Hindus in colonial India was the aboli-tion of Persian and adoption of English as the official language in India by LordBentinck in 1835. This measure was also followed by the introduction of Englishin schools supported by the East India Company replacing Persian and Sanskrit.66Both these steps benefitted Hindus and disadvantaged Muslims primarily becauseof two reasons-firstly because Hindus had already been learning English and therewas already a significant section of the Hindu elite who were well-versed in En-glish and secondly because Muslims thought it to be against their religion to learnEnglish (Khalidi (2006)). The replacement of Persian by English as the officiallanguage resulted in a huge loss for the Muslims and resulted in a significant lossof employment for Muslims in government service and also diminished the prob-ability of Muslims finding government employment in the future.The Indian Sepoy Mutiny of 1857 worsened British-Muslim relations. Al-though both Hindus and Muslims participated in the rebellion, a significant major-ity of British officials considered it to be Muslim-led in character (Kabir (1969)).The Mughal crown was abolished and the last Mughal emperor was sent to Ran-goon on exile. Along with the annexation of Awadh from the Muslim nawab(king) a year earlier in 1856, the British suppression of the Sepoy Mutiny and thechanges it brought thereafter completed the destruction and disintegration of theMuslim elite in much of North India, thus “further curtailing the prospects of sol-diery, intelligentsia and artisans dependent on feudal patronage” (Khalidi (2006)).However these events also led to a change in Muslim attitudes. The survivingelite realized in order to prevent further economic loss they should shake off theirhitherto insular attitude towards the British. The Muslims under the leadership ofSir Syed Ahmed Khan, founder of the Aligarh movement, embraced English edu-cation and co-operated more closely with the British (Hardy (1972)). On the otherside the rising Hindu middle class, a class which had been established due to the67favored treatment of the British, started expressing themselves politically againstthe British by demanding more political autonomy. This led to the formation ofthe Indian National Congress in 1885. With rising Hindu antipathy towards theBritish manifested in the actions of not only the Congress but also many militantorganizations who were advocating violence against the British colonizers, theBritish started raising Muslim communalism as a counter-weight to the emerg-ing Hindu nationalism (Sahoo (2008)). According to many historians this Britishpolicy manifested itself in three key measuresthe partition of Bengal in 1905, theMinto-Morley Reform of 1909 and the Montagu-Chelmsford Reforms of 1919(Mehta and Patwardhan (1942),Sahoo (2008)).The British had set up base first in Bengal. In fact colonial rule in India is gen-erally considered to have started with the victory of the British over the Nawab(king) of Bengal in the Battle of Plassey in 1757. Under British patronage Ben-gal soon became one of the leading provinces in India. Bengali Hindus partic-ularly took to English education and soon established themselves in the colonialbureaucracy. However the Bengali Muslims lagged behind their Hindu counter-parts. For example in 1901, only 22 out of every 10,000 Muslims knew Englishwhile the corresponding number for the Hindus was a much higher at 114 (Ray(1977)). The cultural, economic and political capital of Bengal was in Calcutta.The British proposal to carve out a Muslim majority province of East Bengal fromthe Bengal province thus received support from Muslims as they saw a chance toimprove their fortunes through this proposal (McLane (1965)). On the other handthe upper caste Hindu Bengali elite, with most of their roots in the western part of68Bengal, saw a British conspiracy to undermine their ascendancy and staunchly op-posed this move. Thus the partition led to a further deterioration in Hindu-Muslimrelations in Bengal (McLane (1965)).The Minto-Morley Reform of 1909 is considered to have further deepenedcommunal discord between the two communities. The reforms were undertakenwith a view to tame the nationalist fervor, especially militant activity in Bengal,following the partition of Bengal. The Reforms sought to give native Indians agreater role in governance. However one of the proposals in these reforms wasthe provision of separate electorates for Muslims. The provision of separate elec-torates meant that candidates of either religion could pander to the narrow interestsof their own community and not have to serve members of the other community inorder to win votes. This move of separate electorates has also been held responsi-ble in encouraging Muslim communalism in India (Hasan (1980)).The Montagu-Chelmsford reforms of 1919 were aimed to introduce autonomousinstitutions of self-governance gradually to India. A system of dyarchy was es-tablished under which law and order subjects and subjects responsible for main-taining the supremacy of British Empire like the railways were kept under thecontrol of the British appointed bureaucracy who reported to the Governor of theprovince while subjects like education, public health, agriculture were transferredto the provincial governments which were run by Indians. Both the Central andprovincial legislative assemblies were enlarged and franchise was extended to newgroups of citizens. However with these measures the provision of separate elec-torates were not only maintained but the principle of Muslim over-representation69i.e. representation more than their share in population were introduced in thenewly enlarged central and provincial legislative assemblies. Moreover the natureof the reforms gave power to the newly appointed Muslim legislators to distributepatronage to members of their own brethren at the cost of Hindus (Hardy (1972)).The reforms of 1919 instead of ushering in an era of Hindu-Muslim cooperation inself-governance is said to have increased communal antagonism (Hasan (1980)).From the above analysis we see that there exists a narrative in which the Britishare held responsible for sowing the seeds of communal discord between Hindusand Muslims. However this is not an unchallenged interpretation of history. His-torians like Peter Hardy, emphasize the gradual rise of more aggressive, revivaliststreams of Hinduism and Islam, which although originated in the late eighteenthor early nineteenth centuries but received a fillip by the spread of modern trans-port and communications after 1860. Hardy (1972) also argues that the Britishfollowed a strategy of “balance and rule” rather than a strategy of “divide andrule”. Others have argued the rise of new arenas of local power (Robinson (2007))and the spread of pan-Islamism in the late nineteenth century led to deepeningof the communal fissures in Indian society. Still others like Bayly (1985) andVan der Veer (1994) have argued that there is a “pre-history of communalism”and communalism is not just a product of the colonial era. They argue that itwas “community-based state policies” practiced by the various Hindu and Mus-lim rulers who succeeded the Mughlas and “increasing competition between adeclining Muslim service gentry and rising Hindu merchant classes” which cre-ated communal conflict in India in the pre-colonial period (Talbot (2007)). Hence70it is a matter of debate whether there was any policy of divide and rule activelyfollowed by the British and whether this policy had any long-term impact on reli-gious violence in India.Apart from the channels mentioned above there might be alternative channelsthrough which British rule might have a very different long run impact on Hindu-Muslim religious violence in India. The British instituted a system of modern lawand order in the provinces that they controlled. This system not only consisted ofan efficient police force which was required to keep the native population in linebut also a network of judicial courts. Various accounts suggest that the British po-lice force was more efficient in curbing law and order problems than their counter-parts in the native states (Freitag (1991)). Lange (2004) in his sample of 33 Britishcolonies shows that indirect rule had a negative effect on the institutional measure“Rule of Law” in the post-colonial period. Hence due to institutional persistenceareas those were under direct British rule might have a more able police force,better equipped to deal with communal disturbances than areas that were underthe native princes.British rule might have a long run effect on religious violence is throughthe functioning of democratic institutions. Areas under direct British rule havea longer experience of democratic institutions since the Minto-Morley reformsof 1909. While the Minto-Morley reforms brought in limited self-governmentin British India, the subsequent Montagu-Chelmsford reforms of 1919 and Gov-ernment of India Act, 1935 led to regular elections in the provinces. Moreoverthe fight for independence against British rule exhibited a large degree of Hindu-71Muslim cooperation. If greater experience with democratic institutions and a his-tory of Hindu-Muslim cooperation lead to better functioning of local administra-tion or development of higher social capital, British ruled areas might see lowerincidence of religious violence compared to princely states in independent India.3.3 DataI construct a district-level panel dataset ranging from 1950-1989. The data forthis district-level dataset comes primarily from three sources-the Varshney andWilkinson (2006) dataset on religious violence in India, the replication dataset forthe paper, Iyer (2010) and the India District Database which has data from theIndian Census. The Varshney- Wilkinson dataset contains information on occur-rence of religious riots over the period 1950-1995. I concentrate on the period1950-1989 since from 1990 onwards there was massive Hindu political mobiliza-tion which heralded in a new era of Hindu-Muslim antagonism.The Varshney-Wilkinson dataset collects information about Hindu-Muslimreligious violence from reports appearing in The Times of India newspaper onHindu-Muslim conicts in India over the period 1950-1995. The dataset also recordsfor each incident of communal violence the name of the city/town/village, the dis-trict and state, its duration, the number of people killed, injured and arrested andthe reported proximate cause of the riot. Although there might be some under-reporting on the incidence of riots in small towns the authors take great care tocross check the validity of the dataset with other sources. The replication datasetfor Iyer (2010) available on the The Review of Economics and Statistics data72archive contains all the data used in Iyer (2010). The dataset contains districtlevel information on the ruler status of each district (colonial vs. native ruled),date of annexation by the British, mode of annexation, deaths of native rulers,heirs left by the native rulers, length of British rule and colonial era land revenueinformation.District level demographic and economic data come from the 1951-1991 In-dian Censuses which is available on the Indian District Database.. The IndianCensus is a decennial Census. I use district level data on total population, propor-tion of rural population, population of Muslims, proportion of literates, proportionof employed and proportion of SC/ST population. Since the Census data is de-cennial, I fill the data in the inter Census years through linear interpolation.I also collected district level geographical information from the India Agri-culture and Climate data set assembled by the World Bank. This dataset has dis-trict level information on altitude, latitude, mean annual rainfall, soil type and acoastal dummy. To control for state level political representation, I collected dataon state-level political variables which include the number of effective parties in astate legislature and proportion of seats occupied by different political groupings.2The political variables were taken from the EOPP Indian States database which ismaintained by Timothy Besley and Robin Burgess.2. Effective parties is a widely used measure of party competition which weighs parties with ahigher vote/seat share more heavily than parties with lower vote/seat share. The formula used is1/∑ v2i where vi is the vote/seat share of the ith party.733.4 Empirical StrategyI compare incidence of religious violence between directly ruled British districtand districts belonging to native states by running regressions of the followingform:Ydst =αs + τt +βBritd +λXdst +udst (3.1)where d indexes districts and t time periods. Ydst is our dependent variable whichis either a dummy variable taking a value of 1 if a district experiences a riot in agiven year and 0 if not or it is a count variable which takes the value of the numberof riots or casualties in a given district i in year t. Britd is a dummy variablewhich takes the value of 1 if the district was part of the British empire and zerootherwise. αs and τt are state and time fixed effects and Xdst are district levelcontrols. There are both time-varying and time-invariant (mainly geographicalcharacteristics) variables in the set of controls Xdst .β is the main coefficient of interest-it measures the differential effect of Britishannexation on religious violence in post-Independent India compared to the effectof being ruled by a native king/queen. However β might not represent the causaleffect of British annexation if the variable Briti is potentially endogenous. For ex-ample the British might have been more successful in conquering areas which ex-hibited high levels of initial Hindu-Muslim conflict by exploiting Hindu-Muslimdisunity. In that case β would not represent the true effect of British annexation74and would be biased upward. To overcome this problem of endogenity I use theinstrumental variables used in Iyer (2010). Specifically I exploit the fact that be-tween 1848-1856, under the command of Governor-General Lord Dalhousie, theBritish instituted a policy known as the Doctrine of Lapse under which nativestates whose rulers died without a male heir were to be taken over by the British.The policy was withdrawn when the British Crown took over the reins of govern-ment after the Indian Sepoy Mutiny of 1857. The event of death of a native stateruler without leaving a natural heir is exogenous to our dependent variable, reli-gious violence in post-Independent India. Hence using the Doctrine of Lapse asan instrument for British annexation will help me in recovering the causal impactof British annexation on post-Independent religious violence in India.3 Similarto Iyer (2010) I construct the instrument Lapse as follows: Lapse equals 1 if thenative state was not annexed before 1848 and the ruler died without a male heirin the period 1848-1856; Lapse equals zero if the native state was not annexedbefore 1848 and there was no such death in the period 1848-1856. Since I cannotassign Lapse to districts that were annexed before 1848, my IV sample essentiallyrestricts the sample to only those districts that were not annexed before 1848. Theinstrument Lapse would help us recover the causal effect of British annexationon post-Independent religious violence in India as long as Lapse does not havea direct effect on post-Independent religious violence in India even if the Britishwere selective in their use of the Doctrine of Lapse policy.3. There are 17 districts out of a total of 160 districts in my IV sample where such death ofruler without a natural heir occurred.753.5 ResultsI start my empirical analysis with investigating the descriptive statistics of mykey independent variables. Table 3.1a presents the mean of my dependent vari-ables. Table 3.1b-3.1d present the means of my independent variables for Britishruled and native districts separately and the differences in the means. Table 3.1bpresents the means and the difference in means for geographical controls.Britisruled areas have higher rainfall and more red soil. The means and the differencein means for population controls are summarized in Table 3.1c. There is no sig-nificant difference in means except for log of population. Table 3.1d shows thesummary statistics for political controls. The means of none of the political con-trols are significantly different across British ruled and native ruled districts.I now move on to OLS estimates of the effect of British annexation on themeasures of the intensity of riots in districts. Tables 3.2-3.4 present the results.Table 3.2 shows the effect of British dummy on the probability of occurrenceof any riot in a district d in time t. I introduce different controls sequentially.Column 1 presents the results from estimating equation (1) without any controls.Only state and time dummies are included. In column 2, I introduce geographicalcontrols latitude and altitude, soil dummies, mean annual rainfall and a coastaldummy. The coefficient is positive and significant in both these columns whichseem to support the traditional divide and rule theory. However once I enter pop-ulation controls in column 3 the positive effect goes away-the effect is now neg-ative although insignificant. In my regressions population controls include log of76population, proportion of urban, proportion of Muslim and squared proportion ofMuslim, proportion of literate and proportion of SC/ST. Since post-Independentreligious violence might differ across native ruled districts according to the reli-gion of the ruler I include in the last column a Muslim ruler dummy and a Sikhruler dummy (Hindu ruler being the omitted category). To account for the factthat most religious riots are motivated by political concerns in India, I also in-clude the number of effective parties and proportion of seats won by various po-litical groupings in the state Legislative Assemblies in India. The coefficient onthe British dummy continues to be negative and insignificant after including theseset of additional controls.Table 3.3 shows the OLS estimates of the British dummy on the total numberof riots in a district d at time t. Again controls are introduced sequentially. Similarto Table 3.2, the coefficients are positive in the first two columns (corresponding toincluding no controls and only geographical controls) but turns negative with theintroduction of population (column 3) and religion of ruler and political controls(Column 4). However, none of the coefficients are statistically significant.Table3.4 shows the OLS estimates of the British rule on the total number of riot casu-alties in a district d at time t. Again none of the coefficients are significant.Table 3.5 presents the estimates for the first stage of my instrumental variableestimation. As can be seen, the instrument (the Lapse dummy) is positive andsignificant for all specifications including the one with the full set of controlsgiven in column 4. Thus, as expected, the instrument or the Lapse dummy is astatistically significant predictor of the dummy indicating British Rule.77Tables 3.6-3.8 present my IV estimates. In all the regressions I exclude dis-tricts which were annexed before 1848 since there was no Doctrine of Lapse pol-icy in force then. Table 3.6 presents the IV estimates where the dependent vari-able is the probability of occurrence of any riot. Columns 1 and 2 include onlystate and year fixed effects, columns 3 and 4 include only geographical controls,columns 5 and 6 include geographical and political controls and columns 7 and8 include the full set of controls including political controls and dummies for thereligion of the native ruler. Columns 1, 3, 5 and 7 show the OLS estimates usingthis reduced sample. The OLS estimates are all insignificant in this reduced sam-ple. However in column 2 when I use my instrumental variable Lapse, the Britishdummy becomes negative and significant. This result is the main result of this re-search. This result signifies, at least for the restricted sample considered here, thatcontrary to the popularly held idea that British rule led to deterioration in Hindu-Muslim relations, British rule actually has a negative effect on the probability ofoccurrence of riots in post-Independent India. The effect is robust with the inclu-sion of additional controls as shown in columns 4 (only geographical controls), 6(geographical controls and population controls) and 8 (full set of controls). ThusIV estimates show that British rule reduces probability the occurrence of riots byabout 5 percentage points.In tables 3.7 and 3.8, I estimate the effect of British annexation on total num-ber of riots and total casualties. The controls are again included sequentially.Columns 1 and 2 do not include any controls, columns 3 and 4 include only geo-graphical controls, columns 5 and 6 include geographical and population controls78and columns 7 and 8 includes all controls. The OLS estimates are presented incolumns 1, 3, 5 and 7 and are all insignificant for both the total number of riots(Table 3.7) and total casualties (Table 3.8). Columns 2, 4, 6 and 8 show the IVestimates. It can be seen from table 3.7 that again British dummy significantlyreduces the total number of riots and the effect is robust across all specifications.Columns 2, 4, 6 and 8 of Table 3.8 shows again that British rule reduces the totalriots casualties. The coefficient of the British dummy is significant in the columns2 (no controls) and column 4 (only geographical controls) of Table 3.8. Howeverit loses its significance with the introduction of population controls (column 3 ofTable 3.8) and religion of ruler and political controls (column 4 of Table 3.8).Finally, I have done a falsification exercise in order to test the validity of myinstrument, the Lapse dummy. It can be argued that the Lapse dummy is not avalid instrument if the death of a ruler without natural heir is somehow directlycorrelated with the occurrence of riots and the IV estimates obtained in this chap-ter are capturing that effect. In order to test if this is indeed true I checked whetherthe death of a ruler without a natural heir in years when the Doctrine of Lapse wasnot in place has any impact on the occurrence of riots (Iyer (2010)). Thus I regressthe riots variables on a dummy that equals 1 if the ruler died without a natural heirin the period 1858 to 1884 during which such a death would not result in Britishannexation.4 The estimates are presented in Table 3.9. It can be seen that theresults are all statistically insignificant and small compared to the IV estimates.4. The Doctrine of Lapse policy was withdrawn when the British Crown took direct control ofadministration of British India in 1858 following the First War of Independence/Sepoy Mutiny in1857.793.6 ConclusionIn this chapter I exploit the exogenous nature of the Doctrine of Lapse policyto estimate the causal effect of British annexation. Using instrumental variablestrategy I show that British annexation does not lead to greater Hindu-Muslim vi-olence in post-Independent India. Since the British have often been blamed forincreased tensions between Hindus and Muslims, this result assumes significanceas it challenges the established popular narrative that British colonization led toincreased Hindu-Muslim conflict. Future work would be directed at trying to shedlight on the precise channels through which British annexation might affect post-Independence religious conflict. One possible area of future research would be tolook at the role of land relations in religious violence. Many instances of com-munal violence in the colonial period such as the Malabar rebellion in 1921 andthe rebellion by TituMir in the late 1820s were primarily class based in nature.Since the British brought in many innovations in land relations (Iyer and Baner-jee (2005)), it would be interesting to see the role of these changes in religiousviolence.80Table 3.1a: Summary Statistics: Dependent Vari-ablesVariable Mean Std. Dev. NProbability of Riot 0.0454 0.2082 12240Total Cases 0.0712 0.4859 12240Total casualties 1.656 22.46 12240Notes: The table reports the mean and standard devi-ations of the dependent variables used in this analysis.Probability of Riot is a dummy variable equal to 1 if adistrict experiences a riot in a given year. Total cases andtotal casualties indicate the number of riots and the to-tal number of riot casualties in a district in a given year.The data comes from the Varshney and Wilkinson (2006)dataset on religious violence in India81Table 3.1b: Differences in Geographical ControlsBritish State Native state DifferenceAltitude 393.0 407.0 -13.97(45.26)Latitude 22.84 22.92 -0.083(1.602)Black soil 0.184 0.296 -0.112(0.098)Red soil 0.195 0.096 0.100*(0.060)Alluvial soil 0.534 0.478 0.056(0.108)Coastal dummy 0.139 0.086 0.053(0.070)Mean Annual Rainfall 1419.3 1075.4 343.9**(135.5)Notes: The table reports the summary statistics of geographical controlsused in this paper. Column 1 reports the mean of the variables for districtsunder British rule and the column 2 reports the means for the districtsunder native rule. Column 3 presents the differences in the means. Thegeographical data comes from the India Agriculture and Climate data setassembled by the World Bank. * denotes significant at 10%; ** denotessignificant at 5% and *** denotes significant at 1%.82Table 3.1c: Differences in Population ControlsBritish State Native state DifferenceLog Population 14.42 13.77 0.659***(0.117)Proportion Urban 0.190 0.174 0.016(0.019)Proportion Muslim 0.113 0.111 0.002(0.036)Proportion workers 0.368 0.375 -0.007(0.016)Proportion literate 0.304 0.260 0.045(0.029)Proportion SC/ST 0.239 0.253 -0.014(0.026)Notes: The table reports the summary statistics of population controlsused in this paper. Column 1 reports the mean of the variables fordistricts under British rule and the column 2 reports the means for thedistricts under native rule. Column 3 presents the differences in themeans. The data population controls come from the 1951-1991 IndianCensuses. * denotes significant at 10%; ** denotes significant at 5%and *** denotes significant at 1%.83Table 3.1d: Differences in Political ControlsBritish State Native state DifferenceProportion Congress 0.518 0.561 -0.043(0.028)Proportion Hard-Left 0.078 0.040 0.038(0.034)Proportion Soft-Left 0.039 0.028 0.011(0.008)Proportion Janata 0.127 0.115 0.012(0.029)Proportion Hindu 0.022 0.022 -0.0003(0.006)Notes: The table reports the summary statistics of political con-trols used in this paper. Column 1 reports the mean of the vari-ables for districts under British rule and the column 2 reports themeans for the districts under native rule. Column 3 presents thedifferences in the means. The political variables are taken fromthe EOPP Indian States database. * denotes significant at 10%; **denotes significant at 5% and *** denotes significant at 1%.84Table 3.2: OLS: Probability of Riot(1) (2) (3) (4)Geographical Population Ruler Religion andNo Controls Controls Controls Political ControlsBritish dummy 0.0158* 0.0215* -0.00326 -0.00195(0.00901) (0.0120) (0.00799) (0.00938)Geography Controls NO Yes Yes YesPopulation Controls NO NO Yes YesRuler Religion and Political Controls NO NO NO YesYear Dummies Yes Yes Yes YesState Dummies Yes Yes Yes YesObservations 12240 11080 11080 8858Notes: All regressions include state fixed effect and time fixed effects. Column 1 shows the baseline results whichincludes only state and year fixed effects. In column 2, I have included geographical controls (altitude, latitude,dummies for soil type, coastal dummy and mean annual rainfall of district). Column 3 includes population controls(log population, proportion of urban population, proportion of Muslim population, proportion of workers, proportionof literates and proportion of SC/ST in district) in addition to geographical controls. In Column 4, I have includedcontrols for the religion of the ruler and state level political representation variables (number of effective parties ina state legislature and proportion of seats occupied by Congress, hard Left, soft Left, Janata and Hindu). Standarderrors are clustered at native state level and displayed in parentheses. * denotes significant at 10%; ** denotessignificant at 5% and *** denotes significant at 1%.85Table 3.3: OLS: Total Cases(1) (2) (3) (4)Geographical Population Ruler Religion andNo Controls Controls Controls Political ControlsBritish dummy 0.0323 0.0337 -0.0217 -0.0124(0.0211) (0.0226) (0.0207) (0.0324)Geography Controls NO Yes Yes YesPopulation Controls NO NO Yes YesRuler Religion and Political Controls NO NO NO YesYear Dummies Yes Yes Yes YesState Dummies Yes Yes Yes YesObservations 12240 11080 11080 8858Notes: All regressions include state fixed effect and time fixed effects. Column 1 shows the baseline results whichincludes only state and year fixed effects. In column 2, I have included geographical controls (altitude, latitude,dummies for soil type, coastal dummy and mean annual rainfall of district). Column 3 includes population controls(log population, proportion of urban population, proportion of Muslim population, proportion of workers, proportionof literates and proportion of SC/ST in district) in addition to geographical controls. In Column 4, I have includedcontrols for the religion of the ruler and state level political representation variables (number of effective parties ina state legislature and proportion of seats occupied by Congress, hard Left, soft Left, Janata and Hindu). Standarderrors are clustered at native state level and displayed in parentheses. * denotes significant at 10%; ** denotessignificant at 5% and *** denotes significant at 1%.86Table 3.4: OLS: Total casualties(1) (2) (3) (4)Geographical Population Ruler Religion andNo Controls Controls Controls Political ControlsBritish dummy 1.312 0.894 -0.476 0.654(0.828) (0.717) (0.704) (1.251)Geography Controls NO Yes Yes YesPopulation Controls NO NO Yes YesRuler Religion and Political Controls NO NO NO YesYear Dummies Yes Yes Yes YesState Dummies Yes Yes Yes YesObservations 12240 11080 11080 8858Notes: All regressions include state fixed effect and time fixed effects. Column 1 shows the baseline results whichincludes only state and year fixed effects. In column 2, I have included geographical controls (altitude, latitude,dummies for soil type, coastal dummy and mean annual rainfall of district). Column 3 includes population controls(log population, proportion of urban population, proportion of Muslim population, proportion of workers, proportionof literates and proportion of SC/ST in district) in addition to geographical controls. In Column 4, I have includedcontrols for the religion of the ruler and state level political representation variables (number of effective parties ina state legislature and proportion of seats occupied by Congress, hard Left, soft Left, Janata and Hindu). Standarderrors are clustered at native state level and displayed in parentheses. * denotes significant at 10%; ** denotessignificant at 5% and *** denotes significant at 1%.87Table 3.5: IV: First Stage(1) (2) (3) (4)Geographical Population Ruler Religion andNo Controls Controls Controls Political ControlsInstrument 0.596*** 0.560*** 0.486*** 0.438***(0.172) (0.161) (0.133) (0.123)Geography Controls NO Yes Yes YesPopulation Controls NO NO Yes YesRuler Religion and Political Controls NO NO NO YesYear Dummies Yes Yes Yes YesState Dummies Yes Yes Yes YesObservations 6200 5560 5560 4428F-stat 12.02 12.12 13.40 12.50Notes: Notes: All regressions include state fixed effect and time fixed effects. Column 1 shows the baseline resultswhich includes only state and year fixed effects. In column 2, I have included geographical controls (altitude,latitude, dummies for soil type, coastal dummy and mean annual rainfall of district). Column 3 includes populationcontrols (log population, proportion of urban population, proportion of Muslim population, proportion of workers,proportion of literates and proportion of SC/ST in district) in addition to geographical controls. In Column 4, I haveincluded controls for the religion of the ruler and state level political representation variables (number of effectiveparties in a state legislature and proportion of seats occupied by Congress, hard Left, soft Left, Janata and Hindu).Standard errors are clustered at native state level and displayed in parentheses. * denotes significant at 10%; **denotes significant at 5% and *** denotes significant at 1%.88Table 3.6: IV: Probability of Riot(1) (2) (3) (4) (5) (6) (7) (8)No Controls Geographical Controls Population Controls Ruler Religion andPolitical ControlsOLS IV OLS IV OLS IV OLS IVBritish dummy 0.0106 -0.0458** 0.00541 -0.0513*** -0.00292 -0.0438** -0.0101 -0.0571**(0.0149) (0.0193) (0.0152) (0.0195) (0.0159) (0.0218) (0.0224) (0.0280)Geography Controls NO NO Yes Yes Yes Yes Yes YesPopulation Controls NO NO NO NO Yes Yes Yes YesRuler Religion and Political Controls NO NO NO NO NO NO Yes YesYear Dummies Yes Yes Yes Yes Yes Yes Yes YesState Dummies Yes Yes Yes Yes Yes Yes Yes YesObservations 6200 6200 5560 5560 5560 5560 4428 4428Notes: All regressions include state fixed effect and time fixed effects. Column 1 shows the baseline results which includes only state and yearfixed effects. In column 2, I have included geographical controls (altitude, latitude, dummies for soil type, coastal dummy and mean annual rainfallof district). Column 3 includes population controls (log population, proportion of urban population, proportion of Muslim population, proportionof workers, proportion of literates and proportion of SC/ST in district) in addition to geographical controls. In Column 4, I have included controlsfor the religion of the ruler and state level political representation variables (number of effective parties in a state legislature and proportion of seatsoccupied by Congress, hard Left, soft Left, Janata and Hindu). Standard errors are clustered at native state level and displayed in parentheses. *denotes significant at 10%; ** denotes significant at 5% and *** denotes significant at 1%.89Table 3.7: IV: Total Cases(1) (2) (3) (4) (5) (6) (7) (8)No Controls Geographical Controls Population Controls Ruler Religion andPolitical ControlsOLS IV OLS IV OLS IV OLS IVBritish dummy -0.00402 -0.0716*** -0.0242 -0.0988** -0.0486 -0.0847** -0.0837 -0.120**(0.0165) (0.0251) (0.0262) (0.0385) (0.0329) (0.0390) (0.0515) (0.0523)Geography Controls NO NO Yes Yes Yes Yes Yes YesPopulation Controls NO NO NO NO Yes Yes Yes YesRuler Religion and Political Controls NO NO NO NO NO NO Yes YesYear Dummies Yes Yes Yes Yes Yes Yes Yes YesState Dummies Yes Yes Yes Yes Yes Yes Yes YesObservations 6200 6200 5560 5560 5560 5560 4428 4428Notes: All regressions include state fixed effect and time fixed effects. Column 1 shows the baseline results which includes only state and yearfixed effects. In column 2, I have included geographical controls (altitude, latitude, dummies for soil type, coastal dummy and mean annual rainfallof district). Column 3 includes population controls (log population, proportion of urban population, proportion of Muslim population, proportionof workers, proportion of literates and proportion of SC/ST in district) in addition to geographical controls. In Column 4, I have included controlsfor the religion of the ruler and state level political representation variables (number of effective parties in a state legislature and proportion of seatsoccupied by Congress, hard Left, soft Left, Janata and Hindu). Standard errors are clustered at native state level and displayed in parentheses. *denotes significant at 10%; ** denotes significant at 5% and *** denotes significant at 1%.90Table 3.8: IV: Total casualties(1) (2) (3) (4) (5) (6) (7) (8)No Controls Geographical Controls Population Controls Ruler Religion andPolitical ControlsOLS IV OLS IV OLS IV OLS IVBritish dummy -0.218 -1.509** -0.608 -2.189** -0.777 -1.434 -1.211 -1.622(0.315) (0.603) (0.543) (0.860) (0.628) (0.960) (0.995) (1.243)Geography Controls NO NO Yes Yes Yes Yes Yes YesPopulation Controls NO NO NO NO Yes Yes Yes YesRuler Religion and Political Controls NO NO NO NO NO NO Yes YesYear Dummies Yes Yes Yes Yes Yes Yes Yes YesState Dummies Yes Yes Yes Yes Yes Yes Yes YesObservations 6200 6200 5560 5560 5560 5560 4428 4428Notes: All regressions include state fixed effect and time fixed effects. Column 1 shows the baseline results which includes only state and yearfixed effects. In column 2, I have included geographical controls (altitude, latitude, dummies for soil type, coastal dummy and mean annualrainfall of district). Column 3 includes population controls (log population, proportion of urban population, proportion of Muslim population,proportion of workers, proportion of literates and proportion of SC/ST in district) in addition to geographical controls. In Column 4, I haveincluded controls for the religion of the ruler and state level political representation variables (number of effective parties in a state legislatureand proportion of seats occupied by Congress, hard Left, soft Left, Janata and Hindu). Standard errors are clustered at native state level anddisplayed in parentheses. * denotes significant at 10%; ** denotes significant at 5% and *** denotes significant at 1%.91Table 3.9: Robustness Check: Effect of Death of Ruler Without Natural Heir(1) (2) (3)Probability of Riots Total Cases Total CasualtiesRuler Died Without a Natural Heir Dummy 0.00136 -0.0261 -0.606(0.00991) (0.0296) (0.560)Geography Controls Yes Yes YesPopulation Controls Yes Yes YesRuler Religion and Political Controls Yes Yes YesYear Dummies Yes Yes YesState Dummies Yes Yes YesObservations 3377 3377 3377Notes: All regressions include state fixed effect time fixed effects, geographical controls (altitude, latitude,dummies for soil type, coastal dummy and mean annual rainfall of district), population controls (log popula-tion, proportion of urban population, proportion of Muslim population, proportion of workers, proportion ofliterates and proportion of SC/ST in district), dummies for the religion of the ruler and state level politicalrepresentation variables (number of effective parties in a state legislature and proportion of seats occupied byCongress, hard Left, soft Left, Janata and Hindu). Standard errors are clustered at native state level and dis-played in parentheses. * denotes significant at 10%; ** denotes significant at 5% and *** denotes significant at1%.92Chapter 4The Economic Lives of Muslims inIndia, 1983-20124.1 IntroductionMuslims are one of the most significant minority groups in India. Although Mus-lims constitute around 14 percentage of India’s population and is one of the mostunderprivileged sections of society very little systematic study has been done toexamine the evolution of economic lives of Muslims in India. Quite a substantialliterature now exists studying various facets of the economic gap between upperand lower castes and between males and females. However very little study hasbeen done on the Hindu-Muslim gap. This is precisely the gap in the literaturethat I want to address. This chapter is a an analysis of the evolution of economiccondition of Muslims over the period of 1983 to 2011-12.93The study of economic conditions of Muslims in India is an interesting subjectfor many reasons. One, as mentioned above, Muslims constitute one of the mostsignificant minority groups in India. Not only are they significant in terms of rawnumbers, they are also significant as a political symbol of a secular India. In spiteof being a Hindu majority country, India has the third largest Muslim populationin the world after Indonesia and Pakistan. One of the reasons for the painful andviolent partition of the Indian sub-continent after the British left in 1947 was thecontention put forward by a large section of the Muslim leadership at that time thatMuslims would be treated as second class citizens in independent Hindu majorityIndia. Hence since Independence the socio-economic conditions of Muslims hasalways been a politically sensitive topic in India. Muslims are also seen as apolitically powerful minority with popular accounts suggesting that they vote asa bloc and hence control the fate of almost 100 seats in the Union Parliament inIndia.Another reason why the study of economic conditions of Muslims is an inter-esting topic is that unlike other marginalized groups in India, the lower castes andwomen, Muslims are not a historically disadvantaged community. In fact most ofthe Indian sub-continent had been under Islamic rule for about six centuries (12thto 18th century) prior to the arrival of the British and the establishment of Britishcolonial rule in India. In those six centuries Muslims were at the forefront in so-cial, economic and cultural life in Indian society. Hence the relative deprivationfaced by Muslims is a relatively modern phenomena and not a medieval or ancientlegacy.94The study of the evolution of the economic fortunes of the Muslims over thelast three decades is particularly important because of the tremendous economicgrowth that India achieved in these three decades compared to the sluggish growthit had suffered earlier. Hnatkovska, Lahiri, and Paul (2012) shows that othermarginalized groups like Schedule Castes and Tribes (SC/ST’s) have fared quitewell in this period and has closed down the gap with their upper caste counter-parts. It remains to be seen whether Muslims have also shared in the economicsuccess of this time period. The Sachar Committee (2006) report tried to shedlight on the deprivation faced by Muslims in India. This chapter is an extensionof that effort. It shows the relative deprivation that Muslim face with respect toother social groups and undertakes a quantile analysis to understand the extent ofdeprivation at various parts of the distribution. I also undertake a decompositionanalysis to understand the relative contribution of different factors to the relativegap that Muslims face.In this chapter I try to analyze the economic conditions of Muslims in the pe-riod 1983-2012 following a similar structure as in Hnatkovska, Lahiri, and Paul(2012). First I compare all Muslims with Hindus. Then I break up the sample ac-cording to the sector of residence (rural or urban). I also carry out the comparisonsacross gender. I also break up the Hindu sample into two caste groups-upper castesand SC/STs and then compare Muslims with these two different caste groups ofHindus. I look at educational attainment, occupational choices, wages and con-sumption expenditure1 and analyze how Muslims have fared in relation to other1. I do not include wage and consumption data after 2005 since I only have consistent deflators95groups over the sample period. Using seven rounds of National Sample Survey(NSS) data spanning the period 1983 to 2011-12, I find that Muslims are indeeddeprived with respect to Hindus in education, wages and consumption. More sig-nificantly this relative deprivation is increasing over time. An interesting fact isthat the deprivation faced by Muslims is particularly worse at the top end of thewage and consumption distribution. The other interesting fact that comes out ofthis analysis is that while disadvantaged Hindu castes (SC/STs) have been able toclose their gap with Muslims, the gap between Upper Castes and Muslims havebeen increasing over time. These results stand in stark contrast to the results ob-tained by Hnatkovska, Lahiri, and Paul (2012) who showed that there has been astriking convergence between SC/ST’s and non-SC/ST in India.There is a very thin economics literature studying explicitly the economic con-ditions of Muslims in India. Borooah and Iyer (2005) focus on differences inschool enrolment rates across religious groups. They find that the difference inthe enrolment rates between Hindu and Muslim children was disproportionatelygreater than the difference in their economic position. However they find that thedifference in enrolment rates of Hindus and Muslims at schools has been narrow-ing down over time. Bhaumik and Chakrabarty (2009) examine earning differ-entials between Hindus and Muslims. Their results suggest that lower educationattainment is the key factor behind Hindu-Muslim wage differentials. An inter-esting recent paper Kuran and Singh (2013) shows that in the late colonial periodMuslims were relatively less likely than Hindus to use large-scale and long-livingtill 2004-200596economic organizations, and less likely to serve on corporate boards. The authorsprovide evidence that these differences lay in the differences in the inheritanceslaws of the two communities. While Hindu inheritance laws favored capital ac-cumulation within families and the preservation of family fortunes across genera-tions, the Islamic inheritance system tended to fragment family wealth.The analysis undertaken in this chapter adds to the existing literature in anumber of ways. Firstly by considering education, occupation, wages and con-sumption all together I can present a holistic picture of the evolution of economicconditions of Muslims over the entire sample period. Secondly by undertakinga quantile analysis I can study changes at key points in the distribution of wagesand consumption and not just the mean. Thirdly by focusing on both static andinter-temporal aspects of the education, wage and consumption distributions I canprovide a more nuanced perspective of the evolution of the economic conditionsof Muslims over the entire sample period. Fourthly by comparing Muslims withboth upper and lower caste Hindus I am able to point out that the relative depri-vation of Muslims is not only in comparison with upper caste Hindus, they havebeen faring worse economically than lower caste Hindus over the period understudy.The rest of the chapter is organized as follows. Section 4.2 describes the data,Section 4.3 contains the empirical analysis with sub-sections 4.3.1, 4.3.2, 4.3.3and 4.3.4 looking at educational attainment, occupational choice, wages and con-sumption expenditure respectively. Section 4 concludes.974.2 DataThe data for this project comes from various rounds of the National Sample Sur-vey (NSS) of India. I use household survey data data from the 38th (1983), 43rd(1987-88), 50th (1993-94), 55th (1999-2000), 61st (2004-05), 66th (2009-10) and68 (2011-12) rounds of the NSS. These are thick quinquennial rounds which con-tain information on socio-economic attributes of over hundred thousand house-holds. Since the main focus of this chapter is on tracing differentials in labormarket outcomes across Muslims and Hindus, my primary sample consists of allworking age individuals between the ages of 16 and 65.In Section 4.3.3 where I specifically look at wage differentials across dif-ferent religions, my sample is restricted to only those who report working fora wage/salaried job. This leaves out self-employed individuals from my wage-analysis sample. I restrict the wage analysis till 2004-05. I deflate wage datausing state-level poverty lines using 1983 rural Maharashtra prices as the base. Ido not include the wage data from the 66th and 68th NSS rounds because I do nothave comparable poverty line data for these rounds. 2 Wages are reported in theNSS survey as the daily wage/salaried income received by respondents during thereference week, which is one week previous to that of the survey week. Both cashand in-kind income received is recorded where the in-kind income is convertedinto monetary terms using retail prices.3 I derive average daily wage from thedeflated weekly wage.2. This is because the way poverty lines were defined were changed in 2005 and the two seriesare not comparable.3. This conversion is done by NSS.98I also analyze differences in consumption expenditure which are recorded ona monthly basis at the household level. In Section 4.3.4 where the focus is onconsumption expenditure, the analysis is done at the household level since con-sumption expenditure is the same for all members of the household. Similar to thewage analysis I restrict my analysis till 2004-05 due to the absence of consistentdeflator as described above. I deflate consumption using state-level poverty lineswith 1983 rural Maharashtra prices as the base and then calculate per capita dailyconsumption expenditure.The other important variables included in my analysis are educational vari-ables and occupation variables. In the NSS education is coded as a categeoricalvariable. For expositional simplicity I group the education categories into fivebroad categories: not-literate (Edu1), literate but below primary (Edu2), primary(Edu3), middle (Edu4), and secondary and above (Edu5) in the analysis. For oc-cupational choice, I have aggregated the 3-digit occupation codes that individualsreport in NSS survey into one-digit code. These codes are then broadly groupedinto three categories: white collar jobs, blue collar jobs and agriculture. I havefurther divided blue collar jobs and agriculture into the self employed and wageemployed.994.3 Empirical Analysis4.3.1 Education AttainmentI this section I examine the evolution of the distribution and gap in educationalattainment between Muslims and Hindus. In addition to comparing all Muslimswith all Hindus, I divide the sample according to the sector of residence (rural orurban) and gender and compare Muslims with Hindus over these subsamples. Ialso compare Muslims separately with SC/ST and upper caste Hindus. The sampleconsists of all working age Hindus and Muslims between the ages of 16 and 65who are currently not enrolled in any educational institution.Figure 4.1 examines the pattern of difference between Muslims and Hindusacross different education categories. Panel (a) of Figure 4.1 shows the distribu-tion of Hindus (left) and Muslims (right) across the five education categories foreach of the seven NSS rounds. The first thing to note about the figure is that bothgroups seem to be getting more educated over time. The lowermost bars whichrepresent the percentage of illiterates among the two groups get shorter over timewhereas the uppermost bars representing higher education categories are gettinglonger over time. The reduction of illiterates among either group is quite substan-tial with the percentage of illiterates falling from about 59 percent to about 39 forHindus and 63 to 45 percent for Muslims.The difference in the percentage of people completing secondary educationand above (the topmost bar) is more pronounced. This gap is also increasing overtime. While the percentage of Hindus in this highest education category increased100from 9.3 percent to 26 percent from 1983 to 2011-2012, for Muslims it increasedfrom 5.8 percent to 15.4 percent over the same period.To investigate the relative gap in educational attainment more closely we nowturn to Panel (b) of Figure 4.1. Each of the five bar for a given NSS round mea-sures represents the five educational categories and the height of the bar measuresthe relative gap between the share of Hindus and Muslims in that educational cat-egory. Specifically the height of the bar (H) for any given educational category iand NSS round t is given byHit =Share of HindusitShare of MuslimsitThe height of the bars going above 1 for any category i indicates a disproportionateshare of Hindus in that category while the height of the bar going below 1 denotesa disproportionate share of Muslims in that category. Looking at the figure wesee that Muslims are over-represented in the lower three education categories andunder-represented in the top two education categories, the under-representationbeing most sharp in the top category. Again we note that the relative gap betweenMuslims and Hindus has slightly increased in the top category over time. The overrepresentation of the Muslims in the illiterate category has also increased over thesample period. However for the middle school category (Edu4) the relative gaphas been decreasing.We now turn to different sample comparisons. I start by breaking up the sam-ple into rural and urban sectors. First looking at Figures 4.2 and 4.3 we see that101in rural areas the relative gap in the topmost education category (secondary andabove) between Muslims and Hindus is increasing over time whereas in the urbansector there has been a decrease in the relative gap for this category. The over-representation of Muslims in the bottom category has increased slightly over timefor both rural and urban sectors. Again for the middle school category (Edu4) therelative gap has been decreasing in both rural and urban areas.Next I compare Muslims with upper caste Hindus. Figure 4.4 presents theseresults. There has not been much change in the relative gap in education for thetopmost category between the upper castes and the Muslims and upper castescontinue to be about two times more represented compared to Muslims in thetop category. The over-representation of muslims in the illiterate category haveslightly increased. Similar to above, the relative gap is declining for the middleschool category.Figure 4.5 compares Muslims with lower caste Hindus and shows that lowercastes are improving vis-s vis Muslims.4 Panel (b) shows that Muslims are morerepresented than lower castes in the top educational categories and are less repre-sented in the lowest educational category. However the SC/STs are catching upovertime. The height of the bars for the top two education categories (middle, sec-ondary and above) has been increasing over time. This signifies that compared toMuslims, the relative proportions of SC/STs in the higher education categories has4. I drop Muslims who self-identify as SC/ST from the sample. This is because Muslims donot get official SC/ST status from the government even if they had historically belonged to thesegroups. However the number of Muslims who claim to be SC/ST are negligible in the sample. Inthe 68th round they are less than 0.02 % of the entire sample and around 1% of Muslims.102been increasing over time. The height of the bar representing the lowest educationcategory (illiterates) has slightly decreased over time.Next I compare Muslims with Hindus separately for males and females. Figure4.6 and 4.7 present these results. We can see that for the topmost category, whilethe Hindu-Muslim gap has increased for males, it has reduced for females. Theover-representation of Muslims in the lowest educational category has increasedfor both males and females, the increase being higher for males.Having looked at all the graphs from Figure 4.1 to 4.7 an overall pictureemerges. Muslims seem to lag behind Hindus in education and this gap has beenincreasing over time. When we subdivide the Hindu sample into upper and lowercastes we see that Muslims lag behind upper castes and there is no convergenceover time. Muslims seem to enjoy a narrow advantage over lower caste Hindus.However this relative gap has been decreasing over time with the lower castes fastcatching up with the Muslims. So as far as education attainment is consideredMuslims would rank slightly higher than lower caste Hindus but much lower thanupper caste Hindus. However Muslims seem to be faring worse than both lowerand upper caste Hindus over time-they are not being able to reduce their gap withthe upper caste Hindus and lower caste Hindus are catching up with them. Be-tween males and females Muslim men are doing much worse than Hindu menovertime compared to Muslim females vis-a-vis Hindu females.1034.3.2 Occupational ChoiceIn this section I analyze the differences in occupational choice across Muslims andHindus. I group the occupational categories into the following broad categories-the first category (Occ1) comprises white collar workers-administrators, execu-tives, managers, professionals technical and clerical workers; the second category(Occ 2) consists of blue collar workers such as sales workers, service workers andproduction workers; while Occ 3 comprises farmers, fishermen, loggers, huntersetc. To get a clearer understanding of occupational choice within these broad cat-egories I subdivide both the Occ2 and Occ3 categories into a self-employed partand a wage-employee part. For example a retail dealer would be classified as Occ2(self-employed) whereas a retail salesman would be classified as Occ2 (wage em-ployee). Cultivators would be classified as Occ3 (self-employed) whereas agri-cultural labourers would be classified as Occ3 (wage employee).Figure 4.8 depicts the occupational distribution of the two religious groupings.From panel (a) we see that there has been a consistent decline in the percentageof people belonging to the Occ3 categories for both groups and an increase in thepercentage of people working in occupations belonging to categories Occ1 andOcc2. Muslims are more concentrated in Occ2 occupations than Hindus.In Panel (b) of Figure 4.8 the height of the bar for any occupation category fora given NSS round given by the share of Hindus in that category divided by theshare of Muslims in that category similar to what I did for educational categories.Looking at the graph there does not seem to be much convergence across occupa-tion categories. While Muslims dominate occupations belonging to Occ2, Hindus104dominate occupation belonging Occ3. Within the Occ3 category although Hindushave greater relative shares of both wage employment and self-employment in thecategory, the discrepancy seems to be higher for self-employment which mainlydenotes cultivators. Again the gap in agricultural employment, particularly selfemployment has increased overtime. However the gap in white collar jobs hasdecreased slightly overtime.Figures 4.9 and 4.10 compares Hindus and Muslims in Rural and Urban sec-tors respectively. We see that compared to upper caste Hindus, Muslims are mov-ing out of the agricultural self employment occupation in both the sectors. Mus-lims are also moving out of agricultural wage employment in the rural sector.Muslims also seem to be moving into Occ1 occupations over time (the height ofthese bars are decreasing) in both rural and urban sectors.As a next step I compare Muslims with both upper and lower caste Hindus.Figures 4.11 and 4.12 present these results. We see that compared to upper casteHindus, Muslims are moving out of the agricultural occupations (Occ3), partic-ularly for agricultural self employment (Figure 4.11). Muslims also seem to bemoving into Occ1 occupations over time as compared to upper castes. From Fig-ure 4.12, we can see that Muslims are moving out of agricultural self employmentand blue-collar jobs (wage employed) compared to SC/STs. In Figures 4.11 and4.13, I compare Muslims with Hindus separately for males and females. Againboth Muslim men and women are moving out of agricultural self employment.1054.3.3 WagesI start my wage analysis by plotting the kernel densities of the wage distributionsfor Muslims and Hindus for 1983 and 2004-05 in panel (a) of Figure 4.15 . Thegraph shows that the wage distributions for both groups have shifted to the rightover time indicating that both groups have seen their real wages rise during thesample period. The other interesting fact seen from the graph is that the Mus-lim wage density has a heavier middle and a thinner upper tail than Hindu wagedensity in 1983 indicating that there was a greater share of middle-income wageearners among Muslims but a lower share of high income earners. The heaviermiddle disappears to a large extent in 2004-05 with the Muslim wage density al-most tracking the Hindu wage density in the middle but the thinner upper tail stillremains.Panel (b) of Figure 4.15 plots the differences in log wages between Hindus andMuslims at different percentiles of the wage distribution for the two time periods.In both time periods the graph slopes up indicating that the wage distribution ofHindus is more unequal than that of Muslims and the Hindu-Muslim gap is largerat higher percentiles. However the key fact that one learns from the graph is thatthere is non-convergence in the wage distribution across time periods as the 2004-05 curve lies entirely above the 1983 curve. Hindus seems to be doing better thantheir Muslim counterparts at almost all percentiles of the wage distribution overtime. Infact the initial advantage that Muslims enjoyed at lower quantiles of thewage distribution seems to have been almost wiped out over time.To evaluate the role of various factors in the wage gap, I employ an Oaxaca-106Blinder decomposition technique to decompose the observed mean and quantilewage gaps into explained and unexplained parts and quantify the contribution ofeducation to the gap. I employ Recentered Influence Function(RIF) regressions(Firpo, Fortin, and Lemieux (2009)-henceforth FFL) for decompositions at the10th, 50th and 90th quantiles and OLS regression for the decomposition of mean.In these regressions I use the following covariates apart from the education dum-mies: age, age squared, a rural dummy and state dummies. I first carry out de-compositions for 1983 and 2004-05 separately (Panel A and B of Table 4.1) andthen decompose the changes over the entire sample period (Panel C of Table 4.1).Panel A and B of Table 4.1 report the decomposition the wage gap in 1983 and2004-05 into the explained and unexplained components. Column 1 reports the to-tal gap, column 2 the gap explained due to the covariates in my regression, column3 reports the contribution of education to the explained gap and column 4 reportsthe unexplained gap. At the very top of the distribution i.e. at the 90th quantile thetotal gap is positive for both time periods. At the lower end of the distribution i.e.at the 10th quantile the total gap is negative and statistically significant in 1983,i.e. the Muslims enjoyed an advantage in 1983. While the total gap remains neg-ative in 2004-05 for the 10th percentile, the magnitude falls sharply and no longerremains statistically significant. Again for the mean and the median, the gap wassmall and statistically insignificant (negative in case of median) in 1983. How-ever, by 2004-05 both the mean and median gap became positive and statisticallysignificant with sharp increase in magnitude.Now looking at the components of the total gap we see that the explained gap107is positive for all points of the wage distribution in 2004-05 and for the mean and90th quantile in 1983. As can be seen from column 3 differences in educationalattainment is an important contributor to the explained gap and in some cases itexplains more than 100 percent of the explained gap. This signifies that educationis one of the key hurdles in Muslims achieving greater economic prosperity.In Panel C of Table 4.1, I turn to decomposition of the inter-temporal changesin the gaps reported in the previous table. The table shows the changes in thewage gap between 1983 and 2004-05 and its various components. As is clearform column (1) the wage gap has increased at all points of the wage distribution.The unexplained part of the gap seems to be quite high for the top percentile whichmight point towards increased discrimination in the top quantile. However morerigorous analysis is needed for a more concrete conclusion on this phenomenon.To understand the different facets of the change in the wage gap between Mus-lims and Hindus, I repeat the exercise in Table 4.1 with the sample being dividedinto rural and urban sample. Table 4.2 compares Muslims with Hindus in ruralsector. We can see from Panel A and B of Table 4.2 that Muslims enjoy a rela-tive advantage at the lower end of the wage distribution (10th quantile and mean)in both the periods. Column 3 shows that the coefficient of education is almostalways positive and significant (except for the bottom quantile in 1983) and thuseducation remains an important obstacle for Muslims in the rural sector. Panel Cshows that there has not been much change over the period 1983 to 2004-05.Table 4.3 presents the same analysis for the urban sector and we can see thatin comparison to the rural sector, Muslims have done much poorly in the urban108sector. Panels A and B show that the wage gap between Hindus and Muslims ispositive for almost all the quantiles in both the periods, except the bottom quantilein 1983. Again education remains a key contributor to the explained gap in boththe periods and unexplained gap is positive and significant for the top quantile in2004-05. Panel C shows that the Hindu-Muslim wage gap at mean and the 10thand 90th quantiles have increased significantly from 1983 to 2004-05.Next I compare Muslims separately with upper and lower castes Hindus intables 4.4 and 4.5. Comparing Muslims with upper castes we see that the wagegap is positive for the 50th and 90th quantiles in both the periods. Muslims seemsto have a advantage in the bottom quantile in 1983 which have disappeared by2004-05 (Panel A and B of Table 4.4). Again education explains a significant partof the observed wage gap as shown in column 3. Panel C of Table 4.4 shows thatthe wage gap between upper castes and Muslims have increased significantly forthe mean and the 90th percentile with unexplained gap explaining a significantfraction of the increase.Table 4.5 compares Muslims with SC/STs. Panels A and B show that Muslimshave advantage over SC/STs in terms of wages. However looking at Panel C wesee that the Muslims are losing their initial advantage over SC/STs overtime, es-pecially for the bottom percentiles (10th and 50th) and mean, where the change isstatistically significant. Column 3 of Panel C show that convergence in educationexplains a significant part of the convergence in wage gaps between Muslims andSC/STs over the period 1983 to 2004-05.If we look at men and women (Tables 4.6 and 4.7) we see similar pattern for109men with the wage gaps increasing for all percentiles and mean over the period1983 to 2004-05 (Table 4.6). However for women although an initial gap existsbetween Muslims and Hindus in the bottom quantile, there does not seem to bemuch movement over time.4.3.4 ConsumptionNow I turn to studying consumption differentials between Muslims and Hindus.The variable that I consider is per capita daily household consumption expendi-ture. Since the household consumption is the same for all individuals in a givenhousehold my unit of analysis in this section is a household. Panel (a) of Fig-ure 4.16 plots the kernel densities of distribution of consumption expenditure forMuslims and Hindus for 1983 and 2004-05. The important feature of this graph isthat in both time periods the distribution for Hindus lie to the right of that of theMuslims signifying that Muslims spend much lower on consumption than Hindus.Panel (b) of Figure 4.16 plots the differences in log of daily consumption ex-penditure between Hindus and Muslims at different percentiles of the consump-tion distribution for the two time periods. In both time periods the graph slopesup indicating that the consumption distribution of Hindus is more unequal thanthat of Muslims and the Hindu-Muslim gap is larger at higher percentiles. Similarto what we saw for the wage distribution, there is divergence in the consumptiondistribution across time periods as the 2004-05 line lies almost entirely above the1983 line. Hindus seems to be doing better than their Muslim counterparts atnearly all percentiles of the consumption distribution.110Similar to that of wages I now perform an Oaxaca-Blinder decomposition ofthe change in the consumption gap at the mean and various quantiles of the con-sumption distribution. I use the following covariates in the regressions for decom-position:household size, the number of earning members in the household, ruraldummy, state dummies, educational attainment of the household head and highestlevel of education attained by any household member.Panel A and B of Table 4.8 reports the decomposition the consumption gapin 1983 and 2004-05 into the explained and unexplained components. Column1 reports the total gap, column 2 the gap explained due to the covariates in myregression, column 3 reports the contribution of education to the explained gap andcolumn 4 reports the unexplained gap. It can be seen that the total and explainedgaps are positive and significant at all points of the distribution and in both timeperiods. Also the magnitude of the gap increases as we move up the distributionfrom the 10th quantile to the 90th quantile.Panel C of Table 4.8 reports the decomposition of the inter-temporal changesin the gaps reported in the Panels A and B. Column 1 reports the change in thetotal gap, column 2 the change in the explained gap, column 3 reports the changein the explained gap due to education and column 4 reports the change in the un-explained gap. From column (1) we see that the gap has increased significantlyat both the mean and median. It has also increased at the 10th and the 90th per-centile, however the change is not statistically significant. Most of the increase inthe gap can be explained and education explains a significant part of the explainedcomponent especially as one moves up the distribution.111As a next step I compare Muslim households with Hindu Households sepa-rately for rural and urban sectors. For the rural sector we can see that the Hindu-Muslim consumption gap is positive and significant for all percentiles in both theperiods, with education explaining a significant portion of the gap (Panel A and Bof Table 4.9). However there is not much movement overtime in the rural sectoras seen from Panel C of Table 4.9. On the other hand for the urban sector, theHindus Muslim consumption gap in addition to being positive and significant foreach of the time periods (Panel A and B of Table 4.10), is increasing overtimefor the mean, median and the 90th quantile. The increase is sharpest at the 90thquantile.Next I compare Muslim households separately with upper and lower casteHindu households. Comparing Muslim households to upper caste households inTable 4.11, we can see that there has been a significant increase in the gap at theupper end of the distribution (mean, median and 90th quantile). In contrast, incomparison to Muslim households, lower caste households have increased theirconsumption expenditure at the lower end of the distribution (10th and 50th quan-tile and the mean), as shown in Table 4.12.4.4 ConclusionIn this analysis I have analyzed the evolution of education attainment, occupa-tion choices, wages and consumption expenditure between 1983 and 2004-05 ofMuslims relative to that of Hindus. I found that over this period the deprivationfaced by Muslims has grown more acute, particularly in wages and consumption,112in spite of the tremendous economic growth enjoyed by India during this period.This deprivation is more true at upper ends of the wage and consumption distribu-tion.The analysis carried out in this chapter also presents some important ques-tions. Firstly the analysis shows that Muslims not only fared worse than uppercaste Hindus during the period under study they also lagged behind lower castes.I would like to further study as to why the SC/STs are gaining with respect toMuslims. In particular I would like to examine the if the political changes thathappened in India over this period are a reason for such differences. I would liketo examine whether the rise of a Hindu nationalist party, the Bharatiya Janta Party(BJP) on one hand and the increase in political mobilization of Dalits (ScheduledCastes) on the other hand (Banerjee and Somanathan (2007), both of which hap-pened during this period, has any role to play in these changes. The other keychange that happened during this period was the opening of the economy. Marketforces often result in different groups of winners and losers(Ravallion and Lok-shin (2004)). It would be interesting to see if the opening of the economy hasdisproportionately benefitted other groups in comparison to Muslims. Anotherinteresting question that seems to recur in my analysis is that although the gapbetween Hindu and Muslim men seem to be rising the same cannot be said aboutdifferences between Muslim and Hindu women. I would like to further study thispattern and see whether as Munshi and Rosenzweig (2006) suggest in a globaliz-ing world, new groups who enter the labour force such as women are in a betterposition to adapt to the fast changing economy than entrenched groups such as113men who rely on traditional networks for their labour market success.114Figure 4.1: Distribution of education attainment020406080100 Hindu Muslim19831987−881993−941999−002004−052009−102011−1219831987−881993−941999−002004−052009−102011−12Muslims vs HindusEdu1 Edu2 Edu3 Edu4 Edu5(a)0.511.51983 1987−88 1993−94 1999−00 2004−05 2009−10 2011−12Muslims vs HindusEdu1 Edu2 Edu3 Edu4 Edu5(b)Notes: The figure shows the pattern of educational distribution and educa-tional gap between Hindus and Muslims over five broad educational cat-egories. The time points correspond to seven rounds of NSS covering theperiod 1983-2012. The five broad educational categories are not-literate(Edu1), literate but below primary (Edu2), primary (Edu3), middle (Edu4)and secondary and above (Edu5). Panel (a) shows the distribution acrossthe five broad education categories and the evolution of the distributionover the period 1983-2012. Panel (b) shows the the relative gap in educa-tional attainment between Hindus and Muslims over the same period.115Figure 4.2: Distribution of education attainment-Rural020406080100 Hindu Muslim19831987−881993−941999−002004−052009−102011−1219831987−881993−941999−002004−052009−102011−12Muslims vs Hindus (Rural)Edu1 Edu2 Edu3 Edu4 Edu5(a)0.511.51983 1987−88 1993−94 1999−00 2004−05 2009−10 2011−12Muslims vs Hindus (Rural)Edu1 Edu2 Edu3 Edu4 Edu5(b)Notes: The figure shows the pattern of educational distribution and educa-tional gap between Hindus and Muslims over five broad educational cat-egories for the rural sample. The time points correspond to seven roundsof NSS covering the period 1983-2012. The five broad educational cate-gories are not-literate (Edu1), literate but below primary (Edu2), primary(Edu3), middle (Edu4) and secondary and above (Edu5). Panel (a) showsthe distribution across the five broad education categories and the evolu-tion of the distribution over the period 1983-2012. Panel (b) shows thethe relative gap in educational attainment between Hindus and Muslimsover the same period.116Figure 4.3: Distribution of education attainment-Urban020406080100 Hindu Muslim19831987−881993−941999−002004−052009−102011−1219831987−881993−941999−002004−052009−102011−12Muslims vs Hindus (Urban)Edu1 Edu2 Edu3 Edu4 Edu5(a)0.511.522.51983 1987−88 1993−94 1999−00 2004−05 2009−10 2011−12Muslims vs Hindus (Urban)Edu1 Edu2 Edu3 Edu4 Edu5(b)Notes: The figure shows the pattern of educational distribution and educa-tional gap between Hindus and Muslims over five broad educational cat-egories for the urban sample. The time points correspond to seven roundsof NSS covering the period 1983-2012. The five broad educational cate-gories are not-literate (Edu1), literate but below primary (Edu2), primary(Edu3), middle (Edu4) and secondary and above (Edu5). Panel (a) showsthe distribution across the five broad education categories and the evolu-tion of the distribution over the period 1983-2012. Panel (b) shows thethe relative gap in educational attainment between Hindus and Muslimsover the same period.117Figure 4.4: Distribution of education attainment-Muslims vs Upper Caste020406080100 Upper Caste Muslim19831987−881993−941999−002004−052009−102011−1219831987−881993−941999−002004−052009−102011−12Muslims vs Upper CasteEdu1 Edu2 Edu3 Edu4 Edu5(a)0.511.521983 1987−88 1993−94 1999−00 2004−05 2009−10 2011−12Muslims vs Upper CasteEdu1 Edu2 Edu3 Edu4 Edu5(b)Notes: The figure shows the pattern of educational distribution and ed-ucational gap between Upper Castes and Muslims over five broad edu-cational categories. The time points correspond to seven rounds of NSScovering the period 1983-2012. The five broad educational categories arenot-literate (Edu1), literate but below primary (Edu2), primary (Edu3),middle (Edu4) and secondary and above (Edu5). Panel (a) shows the dis-tribution across the five broad education categories and the evolution ofthe distribution over the period 1983-2012. Panel (b) shows the the rel-ative gap in educational attainment between Upper Castes and Muslimsover the same period.118Figure 4.5: Distribution of education attainment-Muslims vs SC/ST020406080100 SC/ST Muslim19831987−881993−941999−002004−052009−102011−1219831987−881993−941999−002004−052009−102011−12Muslims vs SC/STEdu1 Edu2 Edu3 Edu4 Edu5(a)0.511.51983 1987−88 1993−94 1999−00 2004−05 2009−10 2011−12Muslims vs SC/STEdu1 Edu2 Edu3 Edu4 Edu5(b)Notes: The figure shows the pattern of educational distribution and educa-tional gap between SC/STs and Muslims over five broad educational cat-egories. The time points correspond to seven rounds of NSS covering theperiod 1983-2012. The five broad educational categories are not-literate(Edu1), literate but below primary (Edu2), primary (Edu3), middle (Edu4)and secondary and above (Edu5). Panel (a) shows the distribution acrossthe five broad education categories and the evolution of the distributionover the period 1983-2012. Panel (b) shows the the relative gap in educa-tional attainment between SC/STs and Muslims over the same period.119Figure 4.6: Distribution of education attainment-Male020406080100 Hindu Muslim19831987−881993−941999−002004−052009−102011−1219831987−881993−941999−002004−052009−102011−12Muslims vs Hindus−MaleEdu1 Edu2 Edu3 Edu4 Edu5(a)0.511.51983 1987−88 1993−94 1999−00 2004−05 2009−10 2011−12Muslims vs Hindus−MaleEdu1 Edu2 Edu3 Edu4 Edu5(b)Notes: The figure shows the pattern of educational distribution and ed-ucational gap between Hindu males and Muslim males over five broadeducational categories. The time points correspond to seven rounds ofNSS covering the period 1983-2012. The five broad educational cate-gories are not-literate (Edu1), literate but below primary (Edu2), primary(Edu3), middle (Edu4) and secondary and above (Edu5). Panel (a) showsthe distribution across the five broad education categories and the evolu-tion of the distribution over the period 1983-2012. Panel (b) shows the therelative gap in educational attainment between Hindu males and Muslimmales over the same period.120Figure 4.7: Distribution of education attainment-Female020406080100 Hindu Muslim19831987−881993−941999−002004−052009−102011−1219831987−881993−941999−002004−052009−102011−12Muslims vs Hindus−FemaleEdu1 Edu2 Edu3 Edu4 Edu5(a)0.511.521983 1987−88 1993−94 1999−00 2004−05 2009−10 2011−12Muslims vs Hindus−FemaleEdu1 Edu2 Edu3 Edu4 Edu5(b)Notes: The figure shows the pattern of educational distribution and edu-cational gap between Hindu females and Muslim females over five broadeducational categories. The time points correspond to seven rounds ofNSS covering the period 1983-2012. The five broad educational cate-gories are not-literate (Edu1), literate but below primary (Edu2), primary(Edu3), middle (Edu4) and secondary and above (Edu5). Panel (a) showsthe distribution across the five broad education categories and the evolu-tion of the distribution over the period 1983-2012. Panel (b) shows the therelative gap in educational attainment between Hindu females and Mus-lim females over the same period.121Figure 4.8: Distribution of occupational choices020406080100 Hindu Muslim19831987−881993−941999−002004−052009−102011−1219831987−881993−941999−002004−052009−102011−12Muslims vs HindusWhite Collar Blue Collar (Self−employed)Blue Collar (Wage employee) Agriculture (Self−employed)Agriculture (Wage employee)(a)0.511.51983 1987−88 1993−94 1999−00 2004−05 2009−10 2011−12Muslims vs HindusWhite Collar Blue Collar (Self−employed)Blue Collar (Wage employee) Agriculture (Self−employed)Agriculture (Wage employee)(b)Notes: The figure shows the pattern of occupational choice and gap inoccupational choice between Hindus and Muslims over five broad occu-pational categories. The time points correspond to seven rounds of NSScovering the period 1983-2012. Panel (a) shows the distribution acrossthe five broad occupational categories and the evolution of the distribu-tion over the period 1983-2012. Panel (b) shows the the relative gap inoccupational choice between Hindus and Muslims over the same period.122Figure 4.9: Distribution of occupational choices-Rural020406080100 Hindu Muslim19831987−881993−941999−002004−052009−102011−1219831987−881993−941999−002004−052009−102011−12Muslims vs Hindus−RuralWhite Collar Blue Collar (Self−employed)Blue Collar (Wage employee) Agriculture (Self−employed)Agriculture (Wage employee)(a)0.511.51983 1987−88 1993−94 1999−00 2004−05 2009−10 2011−12Muslims vs Hindus−RuralWhite Collar Blue Collar (Self−employed)Blue Collar (Wage employee) Agriculture (Self−employed)Agriculture (Wage employee)(b)Notes: The figure shows the pattern of occupational choice and gap inoccupational choice between Hindus and Muslims over five broad occu-pational categories for the rural sample. The time points correspond toseven rounds of NSS covering the period 1983-2012. Panel (a) shows thedistribution across the five broad occupational categories and the evolu-tion of the distribution over the period 1983-2012. Panel (b) shows thethe relative gap in occupational choice between Hindus and Muslims overthe same period.123Figure 4.10: Distribution of occupational choices-Urban020406080100 Hindu Muslim19831987−881993−941999−002004−052009−102011−1219831987−881993−941999−002004−052009−102011−12Muslims vs Hindus−urbanWhite Collar Blue Collar (Self−employed)Blue Collar (Wage employee) Agriculture (Self−employed)Agriculture (Wage employee)(a)0.511.521983 1987−88 1993−94 1999−00 2004−05 2009−10 2011−12Muslims vs Hindus−UrbanWhite Collar Blue Collar (Self−employed)Blue Collar (Wage employee) Agriculture (Self−employed)Agriculture (Wage employee)(b)Notes: The figure shows the pattern of occupational choice and gap inoccupational choice between Hindus and Muslims over five broad occu-pational categories for the urban sample. The time points correspond toseven rounds of NSS covering the period 1983-2012. Panel (a) shows thedistribution across the five broad occupational categories and the evolu-tion of the distribution over the period 1983-2012. Panel (b) shows thethe relative gap in occupational choice between Hindus and Muslims overthe same period.124Figure 4.11: Distribution of occupational choices-Muslims vs Upper Caste020406080100 Upper Caste Muslim19831987−881993−941999−002004−052009−102011−1219831987−881993−941999−002004−052009−102011−12Muslim vs Upper CasteWhite Collar Blue Collar (Self−employed)Blue Collar (Wage employee) Agriculture (Self−employed)Agriculture (Wage employee)(a)0.511.51983 1987−88 1993−94 1999−00 2004−05 2009−10 2011−12Muslim vs Upper CasteWhite Collar Blue Collar (Self−employed)Blue Collar (Wage employee) Agriculture (Self−employed)Agriculture (Wage employee)(b)Notes: The figure shows the pattern of occupational choice and gap in oc-cupational choice between Upper Castes and Muslims over five broad oc-cupational categories. The time points correspond to seven rounds of NSScovering the period 1983-2012. Panel (a) shows the distribution acrossthe five broad occupational categories and the evolution of the distribu-tion over the period 1983-2012. Panel (b) shows the the relative gap inoccupational choice between SC/STs and Muslims over the same period.125Figure 4.12: Distribution of occupational choices-Muslims vs SC/ST020406080100 SC/ST Muslim19831987−881993−941999−002004−052009−102011−1219831987−881993−941999−002004−052009−102011−12Muslim vs SC/STWhite Collar Blue Collar (Self−employed)Blue Collar (Wage employee) Agriculture (Self−employed)Agriculture (Wage employee)(a)0.511.522.51983 1987−88 1993−94 1999−00 2004−05 2009−10 2011−12Muslim vs SC/STWhite Collar Blue Collar (Self−employed)Blue Collar (Wage employee) Agriculture (Self−employed)Agriculture (Wage employee)(b)Notes: The figure shows the pattern of occupational choice and gap inoccupational choice between SC/STs and Muslims over five broad occu-pational categories. The time points correspond to seven rounds of NSScovering the period 1983-2012. Panel (a) shows the distribution acrossthe five broad occupational categories and the evolution of the distribu-tion over the period 1983-2012. Panel (b) shows the the relative gap inoccupational choice between SC/STs and Muslims over the same period.126Figure 4.13: Distribution of occupational choices-Male020406080100 Hindu Muslim19831987−881993−941999−002004−052009−102011−1219831987−881993−941999−002004−052009−102011−12Muslims vs Hindus−MaleWhite Collar Blue Collar (Self−employed)Blue Collar (Wage employee) Agriculture (Self−employed)Agriculture (Wage employee)(a)0.511.51983 1987−88 1993−94 1999−00 2004−05 2009−10 2011−12Muslims vs Hindus−MaleWhite Collar Blue Collar (Self−employed)Blue Collar (Wage employee) Agriculture (Self−employed)Agriculture (Wage employee)(b)Notes: The figure shows the pattern of occupational choice and gap in oc-cupational choice between Hindu males and Muslim males over five broadoccupational categories. The time points correspond to seven rounds ofNSS covering the period 1983-2012. Panel (a) shows the distributionacross the five broad occupational categories and the evolution of the dis-tribution over the period 1983-2012. Panel (b) shows the the relative gapin occupational choice between Hindu males and Muslim males over thesame period.127Figure 4.14: Distribution of occupational choices-Female020406080100 Hindu Muslim19831987−881993−941999−002004−052009−102011−1219831987−881993−941999−002004−052009−102011−12Muslims vs Hindus−FemaleWhite Collar Blue Collar (Self−employed)Blue Collar (Wage employee) Agriculture (Self−employed)Agriculture (Wage employee)(a)0.511.51983 1987−88 1993−94 1999−00 2004−05 2009−10 2011−12Muslims vs Hindus−FemaleWhite Collar Blue Collar (Self−employed)Blue Collar (Wage employee) Agriculture (Self−employed)Agriculture (Wage employee)(b)Notes: The figure shows the pattern of occupational choice and gap inoccupational choice between Hindu females and Muslim females overfive broad occupational categories. The time points correspond to sevenrounds of NSS covering the period 1983-2012. Panel (a) shows the dis-tribution across the five broad occupational categories and the evolutionof the distribution over the period 1983-2012. Panel (b) shows the therelative gap in occupational choice between Hindu females and Muslimfemales over the same period.128Figure 4.15: The Log Wage Distributions for 1983 and 2004-20050.2.4.6.81density0 1 2 3 4 5log wageHindu − 1983 Muslim − 1983Hindu − 2004−05 Muslim − 2004−05(a)−.1−.050.05.1.15.2.25.3Log wage (Hindu) − Log wage(Muslim)0 10 20 30 40 50 60 70 80 90 100percentile1983 2004−05(b)Notes: Panel A presents the kernel densities of the log wage distributionsfor Hindus and Muslims for 1983 and 2004-2005. Panel B shows thedifferences in log wages between Hindus and Muslims at different per-centiles of the wage distribution for the two time periods.129Figure 4.16: The Log Consumption Distributions for 1983 and 2004-20050.2.4.6.81density−1 0 1 2 3 4 5log consumptionHindu − 1983 Muslim − 1983Hindu − 2004−05 Muslim − 2004−05(a)−.1−.050.05.1.15.2.25.3Log consumption(Hindu)−Log consumption(Muslim)0 10 20 30 40 50 60 70 80 90 100percentile1983 2004−05(b)Notes: Panel A presents the kernel densities of the log consumption ex-penditure distributions for Hindus and Muslims for 1983 and 2004-2005.Panel B shows the differences in log consumption expenditure betweenHindus and Muslims at different percentiles of the consumption distribu-tion for the two time periods.130Table 4.1: Oaxaca-Blinder decomposition of log wage gaps(1) (2) (3) (4)Total Gap Explained UnexplainedTotal EducationPanel A: 198310th quantile -0.1224*** -0.0083 0.0189*** -0.1141***(0.0177) (0.0082) (0.0041) (0.0179)50th quantile -0.0183 -0.0026 0.0407*** -0.0157(0.0148) (0.0128) (0.0061) (0.0149)90th quantile 0.1047*** 0.1061*** 0.1011*** -0.0014(0.0233) (0.0128) (0.0099) (0.0192)mean 0.0015 0.0241** 0.0558*** -0.0226*(0.0150) (0.0100) (0.0069) (0.0116)Panel B: 2004-0510th quantile -0.0417 0.0208** 0.0218*** -0.0625**(0.0300) (0.0087) (0.0051) (0.0288)50th quantile 0.0268** 0.0523*** 0.0359*** -0.0255**(0.0125) (0.0096) (0.0063) (0.0121)90th quantile 0.2888*** 0.1855*** 0.1813*** 0.1034***(0.0385) (0.0267) (0.0221) (0.0361)mean 0.0813*** 0.0751*** 0.0675*** 0.0062(0.0150) (0.0118) (0.0097) (0.0116)Panel C: Change 1983 to 2004-0510th quantile 0.0807** 0.0291** 0.0028 0.0516(0.0339) (0.0127) (0.0059) (0.0333)50th quantile 0.0451** 0.0549*** -0.0048 -0.0098(0.0191) (0.0157) (0.0085) (0.0196)90th quantile 0.1841*** 0.0793*** 0.0802*** 0.1048**(0.0475) (0.0305) (0.0252) (0.0420)mean 0.0798*** 0.0511*** 0.0117 0.0288*(0.0217) (0.0156) (0.0117) (0.0172)Notes: Panel A and B presents the decomposition of the Hindu-Muslim wagegap in 1983 and 2004-2005 respectively. Panel C presents the change in theHindu-Muslim wage gap between 1983 and 2004-2005. The wage gaps aredecomposed into explained and unexplained components using RIF regressionapproach of FFL for the different quantiles and using a standard OLS decom-position for the mean. Education dummies, age, age squared, a rural dummyand state dummies are the covariates. * significant at 10%; ** significant at5%; *** significant at 1%.131Table 4.2: Oaxaca-Blinder decomposition of log wage gaps: Rural(1) (2) (3) (4)Total Gap Explained UnexplainedTotal EducationPanel A: 198310th quantile -0.1245*** -0.0263** 0.0035 -0.0982***(0.0296) (0.0133) (0.0026) (0.0337)50th quantile -0.0271 -0.0414*** 0.0106*** 0.0143(0.0243) (0.0119) (0.0040) (0.0226)90th quantile 0.0576 0.0716*** 0.0769*** -0.014(0.0453) (0.0259) (0.0193) (0.0390)mean -0.0537*** -0.0105 0.0228*** -0.0432**(0.0178) (0.0124) (0.0065) (0.0173)Panel B: 2004-0510th quantile -0.1549*** -0.0419*** 0.0067* -0.1130***(0.0209) (0.0106) (0.0035) (0.0240)50th quantile -0.0089 0.0198* 0.0130*** -0.0287**(0.0128) (0.0115) (0.0041) (0.0136)90th quantile 0.0079 0.1050*** 0.1375*** -0.0971**(0.0370) (0.0403) (0.0265) (0.0405)mean -0.0345** 0.0272** 0.0349*** -0.0617***(0.0165) (0.0135) (0.0079) (0.0146)Panel C: Change 1983 to 2004-0510th quantile -0.0304 -0.0156 0.0032 -0.0148(0.0310) (0.0206) (0.0041) (0.0416)50th quantile 0.0182 0.0612*** 0.0024 -0.043(0.0297) (0.0159) (0.0057) (0.0277)90th quantile -0.0497 0.0334 0.0606* -0.0831(0.0555) (0.0509) (0.0338) (0.0610)mean 0.0192 0.0377* 0.0121 -0.0185(0.0258) (0.0194) (0.0102) (0.0252)Notes: Panel A and B presents the decomposition of the Hindu-Muslim wagegap in 1983 and 2004-2005 respectively for the rural sample. Panel C presentsthe change in the Hindu-Muslim wage gap between 1983 and 2004-2005. Thewage gaps are decomposed into explained and unexplained components usingRIF regression approach of FFL for the different quantiles and using a standardOLS decomposition for the mean. Education dummies, age, age squared andstate dummies are the covariates. * significant at 10%; ** significant at 5%;*** significant at 1%.132Table 4.3: Oaxaca-Blinder decomposition of log wage gaps: Urban(1) (2) (3) (4)Total Gap Explained UnexplainedTotal EducationPanel A: 198310th quantile -0.0012 0.1640*** 0.1489*** -0.1651***(0.0306) (0.0201) (0.0162) (0.0348)50th quantile 0.2540*** 0.2743*** 0.2175*** -0.0203(0.0351) (0.0198) (0.0172) (0.0293)90th quantile 0.2236*** 0.1905*** 0.1583*** 0.0331(0.0204) (0.0165) (0.0109) (0.0207)mean 0.2023*** 0.2225*** 0.1806*** -0.0202(0.0226) (0.0156) (0.0131) (0.0175)Panel B: 2004-0510th quantile 0.1818** 0.1904*** 0.1248*** -0.0086(0.0709) (0.0218) (0.0170) (0.0713)50th quantile 0.2492*** 0.2860*** 0.1936*** -0.0368(0.0347) (0.0314) (0.0198) (0.0350)90th quantile 0.4280*** 0.2507*** 0.1862*** 0.1773***(0.0607) (0.0200) (0.0142) (0.0569)mean 0.3150*** 0.2731*** 0.1890*** 0.0419*(0.0264) (0.0229) (0.0161) (0.0252)Panel C: Change 1983 to 2004-0510th quantile 0.1830** 0.0265 -0.024 0.1565**(0.0809) (0.0317) (0.0226) (0.0796)50th quantile -0.0048 0.0116 -0.0239 -0.0164(0.0520) (0.0375) (0.0246) (0.0468)90th quantile 0.2044*** 0.0602** 0.0279* 0.1442**(0.0651) (0.0260) (0.0165) (0.0598)mean 0.1127*** 0.0506* 0.0085 0.0621**(0.0377) (0.0283) (0.0200) (0.0309)Notes: Panel A and B presents the decomposition of the Hindu-Muslimwage gap in 1983 and 2004-2005 respectively for the urban sample. Panel Cpresents the change in the Hindu-Muslim wage gap between 1983 and 2004-2005. The wage gaps are decomposed into explained and unexplained com-ponents using RIF regression approach of FFL for the different quantiles andusing a standard OLS decomposition for the mean. Education dummies, age,age squared and state dummies are the covariates. * significant at 10%; **significant at 5%; *** significant at 1%.133Table 4.4: Oaxaca-Blinder decomposition of log wage gaps: Muslim vs Up-per Caste(1) (2) (3) (4)Total Gap Explained UnexplainedTotal EducationPanel A: 198310th quantile -0.0458* 0.0465*** 0.0834*** -0.0924***(0.0274) (0.0118) (0.0073) (0.0334)50th quantile 0.1356*** 0.1844*** 0.1967*** -0.0488***(0.0172) (0.0168) (0.0123) (0.0158)90th quantile 0.2681*** 0.2053*** 0.1907*** 0.0628***(0.0246) (0.0131) (0.0095) (0.0208)mean 0.1416*** 0.1581*** 0.1715*** -0.0165(0.0162) (0.0115) (0.0084) (0.0133)Panel B: 2004-0510th quantile -0.0051 0.0232** 0.0569*** -0.0283(0.0304) (0.0099) (0.0063) (0.0297)50th quantile 0.1266*** 0.1312*** 0.1157*** -0.0045(0.0184) (0.0146) (0.0096) (0.0157)90th quantile 0.4970*** 0.3050*** 0.2900*** 0.1920***(0.0367) (0.0174) (0.0166) (0.0365)mean 0.2005*** 0.1708*** 0.1612*** 0.0297**(0.0169) (0.0140) (0.0110) (0.0136)Panel C: Change 1983 to 2004-0510th quantile 0.0407 -0.0234 -0.0265*** 0.0641(0.0393) (0.0156) (0.0084) (0.0427)50th quantile -0.009 -0.0532*** -0.0810*** 0.0442*(0.0227) (0.0191) (0.0150) (0.0231)90th quantile 0.2288*** 0.0997*** 0.0993*** 0.1291***(0.0450) (0.0211) (0.0194) (0.0418)mean 0.0589** 0.0127 -0.0103 0.0462**(0.0232) (0.0170) (0.0131) (0.0195)Notes: Panel A and B presents the decomposition of the Upper Caste-Muslimwage gap in 1983 and 2004-2005 respectively. Panel C presents the change inthe Upper Caste-Muslim wage gap between 1983 and 2004-2005. The wagegaps are decomposed into explained and unexplained components using RIFregression approach of FFL for the different quantiles and using a standardOLS decomposition for the mean. Education dummies, age, age squared andstate dummies are the covariates. * significant at 10%; ** significant at 5%;*** significant at 1%.134Table 4.5: Oaxaca-Blinder decomposition of log wage gaps: Muslim vsSC/ST(1) (2) (3) (4)Total Gap Explained UnexplainedTotal EducationPanel A: 198310th quantile -0.1389*** -0.0400*** -0.0386*** -0.0989***(0.0190) (0.0125) (0.0039) (0.0231)50th quantile -0.2002*** -0.1056*** -0.0755*** -0.0946***(0.0183) (0.0120) (0.0056) (0.0148)90th quantile -0.3363*** -0.2479*** -0.2233*** -0.0884***(0.0339) (0.0210) (0.0180) (0.0236)mean -0.2309*** -0.1273*** -0.1052*** -0.1036***(0.0187) (0.0114) (0.0064) (0.0128)Panel B: 2004-0510th quantile -0.0472 -0.0535*** -0.0288*** 0.0064(0.0294) (0.0101) (0.0037) (0.0305)50th quantile -0.0616*** -0.0447*** -0.0381*** -0.017(0.0161) (0.0103) (0.0038) (0.0155)90th quantile -0.3152*** -0.2545*** -0.1967*** -0.0608(0.0371) (0.0274) (0.0212) (0.0398)mean -0.1114*** -0.0862*** -0.0655*** -0.0252**(0.0141) (0.0095) (0.0065) (0.0128)Panel C: Change 1983 to 2004-0510th quantile 0.0917*** -0.0136 0.0098* 0.1053***(0.0344) (0.0142) (0.0053) (0.0369)50th quantile 0.1386*** 0.0609*** 0.0374*** 0.0777***(0.0250) (0.0168) (0.0071) (0.0236)90th quantile 0.021 -0.0066 0.0266 0.0276(0.0542) (0.0366) (0.0302) (0.0510)mean 0.1195*** 0.0411** 0.0397*** 0.0784***(0.0245) (0.0170) (0.0099) (0.0189)Notes: Panel A and B presents the decomposition of the SC/ST-Muslim wagegap in 1983 and 2004-2005 respectively. Panel C presents the change in theSC/ST-Muslim wage gap between 1983 and 2004-2005. The wage gaps aredecomposed into explained and unexplained components using RIF regressionapproach of FFL for the different quantiles and using a standard OLS decompo-sition for the mean. Education dummies, age, age squared and state dummiesare the covariates. * significant at 10%; ** significant at 5%; *** significant at1%.135Table 4.6: Oaxaca-Blinder decomposition of log wage gaps: Male(1) (2) (3) (4)Total Gap Explained UnexplainedTotal EducationPanel A: 198310th quantile -0.0320** -0.0024 0.0232*** -0.0296*(0.0147) (0.0073) (0.0023) (0.0164)50th quantile 0.0214 0.1035*** 0.0943*** -0.0822***(0.0180) (0.0136) (0.0083) (0.0191)90th quantile 0.1602*** 0.1434*** 0.1333*** 0.0168(0.0227) (0.0126) (0.0097) (0.0197)mean 0.0623*** 0.0871*** 0.0892*** -0.0248**(0.0145) (0.0089) (0.0065) (0.0112)Panel B: 2004-0510th quantile 0.0784*** 0.0325*** 0.0193*** 0.0460*(0.0211) (0.0093) (0.0025) (0.0245)50th quantile 0.1325*** 0.0929*** 0.0566*** 0.0396**(0.0170) (0.0105) (0.0053) (0.0164)90th quantile 0.3773*** 0.2612*** 0.2388*** 0.1161***(0.0339) (0.0244) (0.0176) (0.0324)mean 0.1717*** 0.1213*** 0.0964*** 0.0504***(0.0143) (0.0121) (0.0078) (0.0118)Panel C: Change 1983 to 2004-0510th quantile 0.1104*** 0.0349*** -0.0039 0.0755***(0.0244) (0.0125) (0.0031) (0.0267)50th quantile 0.1111*** -0.0107 -0.0376*** 0.1218***(0.0232) (0.0154) (0.0097) (0.0254)90th quantile 0.2171*** 0.1178*** 0.1056*** 0.0993***(0.0381) (0.0276) (0.0213) (0.0340)mean 0.1094*** 0.0342** 0.0072 0.0752***(0.0187) (0.0150) (0.0101) (0.0151)Notes: Panel A and B presents the decomposition of the Hindu-Muslim wagegap in 1983 and 2004-2005 respectively for males. Panel C presents the changein the Hindu-Muslim wage gap between 1983 and 2004-2005. The wage gapsare decomposed into explained and unexplained components using RIF re-gression approach of FFL for the different quantiles and using a standard OLSdecomposition for the mean. Education dummies, age, age squared and statedummies are the covariates. * significant at 10%; ** significant at 5%; ***significant at 1%.136Table 4.7: Oaxaca-Blinder decomposition of log wage gaps: Female(1) (2) (3) (4)Total Gap Explained UnexplainedTotal EducationPanel A: 198310th quantile 0.1973*** 0.0202 0.0053* 0.1771***(0.0653) (0.0168) (0.0032) (0.0645)50th quantile 0.003 -0.0146 -0.0026 0.0176(0.0297) (0.0154) (0.0043) (0.0279)90th quantile 0.0198 -0.0087 -0.0157 0.0285(0.0711) (0.0471) (0.0334) (0.0647)mean 0.0514* -0.0118 -0.0040 0.0632**(0.0308) (0.0174) (0.0097) (0.0275)Panel B: 2004-0510th quantile 0.2804*** -0.0099 -0.004 0.2903***(0.0745) (0.0091) (0.0033) (0.0744)50th quantile 0.0054 -0.0523*** -0.0059 0.0577(0.0457) (0.0128) (0.0077) (0.0429)90th quantile -0.2987 -0.149 -0.0945 -0.1497(0.2046) (0.1164) (0.1246) (0.1676)mean 0.0524 -0.0478** -0.0165 0.1002**(0.0455) (0.0201) (0.0204) (0.0400)Panel C: Change 1983 to 2004-0510th quantile 0.0831 -0.0301 -0.0093* 0.1133(0.0980) (0.0195) (0.0049) (0.0999)50th quantile 0.0023 -0.0377** -0.0033 0.0401(0.0595) (0.0174) (0.0086) (0.0604)90th quantile -0.3185 -0.1403 -0.0788 -0.1782(0.2199) (0.1291) (0.1299) (0.1838)mean 0.0010 -0.0360 -0.0125 0.0370(0.0573) (0.0269) (0.0227) (0.0543)Notes: Panel A and B presents the decomposition of the Hindu-Muslim wagegap in 1983 and 2004-2005 respectively for females. Panel C presents thechange in the Hindu-Muslim wage gap between 1983 and 2004-2005. Thewage gaps are decomposed into explained and unexplained components usingRIF regression approach of FFL for the different quantiles and using a stan-dard OLS decomposition for the mean. Education dummies, age, age squaredand state dummies are the covariates. * significant at 10%; ** significant at5%; *** significant at 1%.137Table 4.8: Oaxaca-Blinder decomposition of log consumption gaps(1) (2) (3) (4)Total Gap Explained UnexplainedTotal EducationPanel A: 198310th quantile 0.0763*** 0.0779*** 0.0306*** -0.0017(0.0147) (0.0043) (0.0024) (0.0140)50th quantile 0.0971*** 0.1115*** 0.0397*** -0.0144*(0.0093) (0.0046) (0.0028) (0.0085)90th quantile 0.1685*** 0.1652*** 0.0578*** 0.0032(0.0158) (0.0071) (0.0040) (0.0143)mean 0.1058*** 0.1138*** 0.0417*** -0.0080(0.0093) (0.0047) (0.0029) (0.0081)Panel B: 2004-0510th quantile 0.0841*** 0.1043*** 0.0397*** -0.0202*(0.0108) (0.0065) (0.0032) (0.0108)50th quantile 0.1328*** 0.1393*** 0.0569*** -0.0065(0.0102) (0.0070) (0.0038) (0.0099)90th quantile 0.2066*** 0.1867*** 0.0990*** 0.0199(0.0259) (0.0116) (0.0065) (0.0226)mean 0.1420*** 0.1428*** 0.0644*** -0.0007(0.0094) (0.0070) (0.0042) (0.0082)Panel C: Change 1983 to 2004-0510th quantile 0.0078 0.0263*** 0.0091** -0.0185(0.0190) (0.0078) (0.0037) (0.0170)50th quantile 0.0357** 0.0278*** 0.0172*** 0.0079(0.0152) (0.0081) (0.0046) (0.0148)90th quantile 0.0382 0.0215 0.0412*** 0.0167(0.0279) (0.0139) (0.0083) (0.0266)mean 0.0362*** 0.0290*** 0.0227*** 0.0073(0.0136) (0.0082) (0.0049) (0.0125)Notes: Panel A and B presents the decomposition of the Hindu-Muslimconsumption gap in 1983 and 2004-2005 respectively. Panel C presents thechange in the Hindu-Muslim consumption gap between 1983 and 2004-2005.The wage gaps are decomposed into explained and unexplained componentsusing RIF regression approach of FFL for the different quantiles and using astandard OLS decomposition for the mean. Household size, the number ofearning members in the household, rural dummy, state dummies, educationalattainment of the household head and highest level of education attained inthe household are the covariates. * significant at 10%; ** significant at 5%;*** significant at 1%.138Table 4.9: Oaxaca-Blinder decomposition of log consumption gaps: Rural(1) (2) (3) (4)Total Gap Explained UnexplainedTotal EducationPanel A: 198310th quantile 0.0558*** 0.0830*** 0.0266*** -0.0272*(0.0185) (0.0078) (0.0028) (0.0161)50th quantile 0.0583*** 0.1072*** 0.0295*** -0.0489***(0.0098) (0.0065) (0.0027) (0.0101)90th quantile 0.1250*** 0.1631*** 0.0419*** -0.0382**(0.0183) (0.0104) (0.0042) (0.0162)mean 0.0678*** 0.1135*** 0.0313*** -0.0457***(0.0107) (0.0079) (0.0029) (0.0097)Panel B: 2004-0510th quantile 0.0396*** 0.0856*** 0.0297*** -0.0460***(0.0131) (0.0065) (0.0025) (0.0125)50th quantile 0.0789*** 0.1192*** 0.0415*** -0.0403***(0.0102) (0.0069) (0.0030) (0.0099)90th quantile 0.1289*** 0.1855*** 0.0679*** -0.0565**(0.0216) (0.0130) (0.0054) (0.0221)mean 0.0849*** 0.1273*** 0.0461*** -0.0424***(0.0103) (0.0068) (0.0030) (0.0101)Panel C: Change 1983 to 2004-0510th quantile -0.0162 0.0025 0.0031 -0.0187(0.0222) (0.0109) (0.0031) (0.0183)50th quantile 0.0206 0.012 0.0120*** 0.0086(0.0156) (0.0095) (0.0036) (0.0162)90th quantile 0.0039 0.0223 0.0260*** -0.0184(0.0271) (0.0163) (0.0066) (0.0269)mean 0.0171 0.0138 0.0148*** 0.0033(0.0157) (0.0108) (0.0036) (0.0151)Notes: Panel A and B presents the decomposition of the Hindu-Muslim con-sumption gap in 1983 and 2004-2005 respectively for the rural sample. PanelC presents the change in the Hindu-Muslim consumption gap between 1983and 2004-2005. The wage gaps are decomposed into explained and unex-plained components using RIF regression approach of FFL for the differentquantiles and using a standard OLS decomposition for the mean. Householdsize, the number of earning members in the household, state dummies, educa-tional attainment of the household head and highest level of education attainedin the household are the covariates. * significant at 10%; ** significant at 5%;*** significant at 1%.139Table 4.10: Oaxaca-Blinder decomposition of log consumption gaps: Urban(1) (2) (3) (4)Total Gap Explained UnexplainedTotal EducationPanel A: 198310th quantile 0.1526*** 0.1758*** 0.1020*** -0.0232(0.0175) (0.0082) (0.0060) (0.0165)50th quantile 0.2435*** 0.2602*** 0.1347*** -0.0167(0.0167) (0.0111) (0.0070) (0.0158)90th quantile 0.2620*** 0.2920*** 0.1396*** -0.03(0.0296) (0.0178) (0.0094) (0.0294)mean 0.2384*** 0.2489*** 0.1281*** -0.0105(0.0159) (0.0100) (0.0059) (0.0153)Panel B: 2004-0510th quantile 0.1502*** 0.2274*** 0.1140*** -0.0772***(0.0196) (0.0174) (0.0111) (0.0202)50th quantile 0.3318*** 0.3413*** 0.1813*** -0.0095(0.0186) (0.0201) (0.0133) (0.0195)90th quantile 0.3791*** 0.3427*** 0.1878*** 0.0364(0.0377) (0.0240) (0.0127) (0.0285)mean 0.2977*** 0.3034*** 0.1615*** -0.0057(0.0199) (0.0180) (0.0110) (0.0157)Panel C: Change 1983 to 2004-0510th quantile -0.0023 0.0517*** 0.012 -0.0540**(0.0269) (0.0198) (0.0136) (0.0236)50th quantile 0.0883*** 0.0811*** 0.0466*** 0.0072(0.0267) (0.0254) (0.0160) (0.0244)90th quantile 0.1171*** 0.0507* 0.0482*** 0.0664*(0.0452) (0.0304) (0.0162) (0.0376)mean 0.0593** 0.0545** 0.0335** 0.0048(0.0264) (0.0223) (0.0134) (0.0191)Notes: Panel A and B presents the decomposition of the Hindu-Muslim con-sumption gap in 1983 and 2004-2005 respectively for the urban sample. PanelC presents the change in the Hindu-Muslim consumption gap between 1983and 2004-2005. The wage gaps are decomposed into explained and unex-plained components using RIF regression approach of FFL for the differentquantiles and using a standard OLS decomposition for the mean. Householdsize, the number of earning members in the household, state dummies, educa-tional attainment of the household head and highest level of education attainedin the household are the covariates. * significant at 10%; ** significant at 5%;*** significant at 1%.140Table 4.11: Oaxaca-Blinder decomposition of log consumption gaps: Mus-lim vs Upper Caste(1) (2) (3) (4)Total Gap Explained UnexplainedTotal EducationPanel A: 198310th quantile 0.1727*** 0.1086*** 0.0618*** 0.0641***(0.0135) (0.0073) (0.0029) (0.0131)50th quantile 0.1826*** 0.1363*** 0.0727*** 0.0463***(0.0092) (0.0062) (0.0030) (0.0087)90th quantile 0.2521*** 0.1943*** 0.0997*** 0.0577***(0.0162) (0.0076) (0.0047) (0.0165)mean 0.1945*** 0.1432*** 0.0768*** 0.0513***(0.0090) (0.0075) (0.0031) (0.0095)Panel B: 2004-0510th quantile 0.1615*** 0.1175*** 0.0568*** 0.0440***(0.0113) (0.0062) (0.0031) (0.0106)50th quantile 0.2196*** 0.1751*** 0.0905*** 0.0446***(0.0106) (0.0077) (0.0044) (0.0105)90th quantile 0.3179*** 0.2533*** 0.1583*** 0.0645***(0.0267) (0.0152) (0.0106) (0.0241)mean 0.2334*** 0.1801*** 0.0992*** 0.0533***(0.0098) (0.0073) (0.0046) (0.0086)Panel C: Change 1983 to 2004-0510th quantile -0.0113 0.0088 -0.0051 -0.0201(0.0183) (0.0104) (0.0040) (0.0166)50th quantile 0.0370** 0.0387*** 0.0178*** -0.0017(0.0144) (0.0096) (0.0052) (0.0148)90th quantile 0.0658** 0.0590*** 0.0585*** 0.0068(0.0286) (0.0165) (0.0118) (0.0262)mean 0.0389*** 0.0368*** 0.0224*** 0.0020(0.0131) (0.0103) (0.0054) (0.0135)Notes: Panel A and B presents the decomposition of the Upper Caste-Muslimconsumption gap in 1983 and 2004-2005 respectively. Panel C presents thechange in the Upper Caste-Muslim consumption gap between 1983 and 2004-2005. The wage gaps are decomposed into explained and unexplained com-ponents using RIF regression approach of FFL for the different quantiles andusing a standard OLS decomposition for the mean. Household size, numberof earning members in the household, state dummies, educational attainmentof the household head and highest level of education attained in the householdare the covariates. * significant at 10%; ** significant at 5%; *** significantat 1%.141Table 4.12: Oaxaca-Blinder decomposition of log consumption gaps: Mus-lim vs SC/ST(1) (2) (3) (4)Total Gap Explained UnexplainedTotal EducationPanel A: 198310th quantile -0.0836*** -0.0152** -0.0401*** -0.0684***(0.0172) (0.0069) (0.0026) (0.0198)50th quantile -0.1005*** 0.0017 -0.0496*** -0.1023***(0.0119) (0.0056) (0.0028) (0.0122)90th quantile -0.1165*** 0.0404*** -0.0678*** -0.1569***(0.0178) (0.0090) (0.0063) (0.0198)mean -0.1005*** 0.0054 -0.0537*** -0.1059***(0.0116) (0.0056) (0.0031) (0.0130)Panel B: 2004-0510th quantile -0.0325*** 0.0132** -0.0144*** -0.0457***(0.0109) (0.0064) (0.0019) (0.0110)50th quantile -0.0372*** 0.0244*** -0.0180*** -0.0616***(0.0113) (0.0056) (0.0024) (0.0112)90th quantile -0.1227*** 0.0260** -0.0342*** -0.1487***(0.0273) (0.0105) (0.0048) (0.0258)mean -0.0573*** 0.0222*** -0.0217*** -0.0795***(0.0105) (0.0059) (0.0028) (0.0097)Panel C: Change 1983 to 2004-0510th quantile 0.0511** 0.0284*** 0.0256*** 0.0227(0.0203) (0.0094) (0.0031) (0.0222)50th quantile 0.0633*** 0.0227*** 0.0316*** 0.0407**(0.0185) (0.0069) (0.0035) (0.0191)90th quantile -0.0062 -0.0144 0.0337*** 0.0082(0.0311) (0.0128) (0.0087) (0.0315)mean 0.0432*** 0.0168** 0.0320*** 0.0264(0.0166) (0.0078) (0.0042) (0.0180)Notes: Panel A and B presents the decomposition of the SC/ST-Muslim con-sumption gap in 1983 and 2004-2005 respectively. Panel C presents the changein the SC/ST-Muslim consumption gap between 1983 and 2004-2005. Thewage gaps are decomposed into explained and unexplained components usingRIF regression approach of FFL for the different quantiles and using a standardOLS decomposition for the mean. Household size, the number of earning mem-bers in the household, state dummies, educational attainment of the householdhead and highest level of education attained in the household are the covariates.* significant at 10%; ** significant at 5%; *** significant at 1%.142Chapter 5ConclusionThis thesis attempts to understand the Indian development experience through theprism of disadvantaged minorities and their interaction with the majority group.Political reservations in India were mandated with a view that it would providepolitical representation to Scheduled Castes and Tribes and thus make it possiblefor them not to be at the mercy of their upper castes. Chapter 2 shows that anyevaluation of such a policy of reservation would need to take into account theincentives of the politicians. It shows that Schedule Caste politicians tend to targetgoods which are more easily identifiable with a particular legislator and hencecan bring in political gains for the legislator. The absence of a comparable effectfor Scheduled Tribes points to the fact that such incentives might differ acrosspoliticians belonging to different ethnic groups.The third chapter which deals with the effect of colonization on post-IndependentHindu-Muslim violence tries to answer an important question-whether British143colonialism had led to a worsening of Hindu-Muslim relations in the Indian subcontinent. I find evidence contrary to the popular claim that British colonialismhad an adverse effect on Hindu-Muslim relations. One thing to note though in thatmy results is not a direct test of the ”divide and rule” policy allegedly followed bythe British in their dealings with their colonial subjects in the Indian subcontinent.It might well be the case that in addition to the ”divide and rule ” policy, the Britishwere responsible for other measures such as an improved law and order system orestablishment of democratic institutions which offsets the adverse effects of their”divide and rule” policy. Future research would try to gain an understanding ofthe precious channel through which British colonialism might have an effect onHindu-Muslim religious violence.Chapter 4 looks at the evolution of economic conditions of Muslims vis-a-visnon-Muslims in the last three decades. The analysis shows that not only relativedeprivation of Muslims is increasing with respect to upper caste Hindus, Sched-uled Castes/Tribes are also gaining compared to Muslims. Surprisingly this wors-ening of relative deprivation is more acute in case of Muslim men than women.My results are in contrast to what Hnatkovska, Lahiri, and Paul (2012) find incase of Scheduled Castes/Tribes. This difference in outcomes between Scheduledcastes and Muslims in the context of a liberalizing economy like India throws upa puzzle about how different marginalized communities have fared which requiresfurther research.144BibliographyAcemoglu, Daron, Simon Johnson, and James A Robinson. 2001. “The Colo-nial Origins of Comparative Development: An Empirical Investigation.” TheAmerican Economic Review 91 (5): 1369–1401.Alesina, Alberto. 1997. Political cycles and the Macroeconomy. MIT press.Alesina, Alberto, William Easterly, and Janina Matuszeski. 2011. “Artificial states.”Journal of the European Economic Association 9 (2): 246–277.Baber, Zaheer. 2004. “‘Race’, Religion and Riots: The ‘Racialization’ of Commu-nal Identity and Conflict in India.” Sociology 38 (4): 701–718.Banerjee, Abhijit, and Esther Duflo. 2006. “Addressing Absence.” Journal of Eco-nomic Perspectives 20 (1): 117.Banerjee, Abhijit, Selvan Kumar, Rohini Pande, and Felix Su. 2010. “Do informedvoters make better choices? Experimental evidence from urban India.” Work-ing Paper, Massachusetts Institute of Technology.Banerjee, Abhijit, and Rohini Somanathan. 2007. “The political economy of pub-lic goods: Some evidence from India.” Journal of Development Economics82 (2): 287–314.Bardhan, Pranab K, Dilip Mookherjee, and Monica Parra Torrado. 2010. “Impactof political reservations in West Bengal local governments on anti-povertytargeting.” Journal of Globalization and Development 1 (1).Bayly, Christopher A. 1985. “The Pre-history of ‘Communalism’? Religious Con-flict in India, 1700–1860.” Modern Asian Studies 19 (02): 177–203.145Berman, Peter A. 1998. “Rethinking health care systems: Private health care pro-vision in India.” World Development 26 (8): 1463–1479.Besley, Timothy, and Robin Burgess. 2002. “The Political Economy Of Govern-ment Responsiveness: Theory And Evidence From India.” Quarterly Journalof Economics 117 (4).Besley, Timothy, Rohini Pande, Lupin Rahman, and Vijayendra Rao. 2004. “Thepolitics of public good provision: Evidence from Indian local governments.”Journal of the European Economic Association 2 (2-3): 416–426.Bhalotra, Sonia, Irma Clots-Figueras, Guilhem Cassan, and Lakshmi Iyer. 2014.“Religion, politician identity and development outcomes: Evidence from In-dia.” Journal of Economic Behavior & Organization 104:4–17.Bhatia, Bela, and Jean Dreze. 2006. “Employment guarantee in Jharkhand: Groundrealities.” Economic and Political Weekly: 3198–3202.Bhaumik, Sumon Kumar, and Manisha Chakrabarty. 2009. “Is education the panaceafor economic deprivation of Muslims?: evidence from wage earners in India,1987–2005.” Journal of Asian Economics 20 (2): 137–149.Borooah, Vani K, and Sriya Iyer. 2005. “Vidya, Veda, and Varna: The influence ofreligion and caste on education in rural India.” The Journal of DevelopmentStudies 41 (8): 1369–1404.Burgess, Robin, Remi Jedwab, Edward Miguel, Ameet Morjaria, et al. 2014. TheValue of Democracy: Evidence from Road Building in Kenya. Technical re-port.Carroll, James. 2002. Constantine’s Sword: The Church and the Jews–A History.Houghton Mifflin Harcourt.Chattopadhyay, Raghabendra, and Esther Duflo. 2004. “Women as policy makers:Evidence from a randomized policy experiment in India.” Econometrica 72(5): 1409–1443.146Chin, Aimee, and Nishith Prakash. 2011. “The redistributive effects of politicalreservation for minorities: Evidence from India.” Journal of DevelopmentEconomics 96 (2): 265–277.Clots-Figueras, Irma. 2011. “Women in politics: Evidence from the Indian States.”Journal of Public Economics 95 (7): 664–690.Clots-Figuerasa, Irma. 2012. “Are Female Leaders Good for Education?” Ameri-can Economic Journal: Applied Economics 4 (1): 212–244.Das, Suranjan. 1990. “Communal Violence in Twentieth Century Colonial Ben-gal: An Analytical Framework.” Social Scientist: 21–37.Eisinger, Peter K. 1982. “Black employment in municipal jobs: The impact ofblack political power.” American Political Science Review 76 (02): 380–392.Engerman, Stanley L., and Kenneth L. Sokoloff. 1997. “Factor endowments, in-stitutions, and differential paths of growth among new world economies,”edited by Stephen Haber, 260–304. How Latin America Fell Behind. Stan-ford, CA: Stanford University Press.Field, Erica, Matthew Levinson, Rohini Pande, and Sujata Visaria. 2008. “Segre-gation, Rent Control, and Riots: The Economics of Religious Conflict in anIndian City.” The American Economic Review: 505–510.Firpo, Sergio, Nicole M Fortin, and Thomas Lemieux. 2009. “Unconditional quan-tile regressions.” Econometrica 77 (3): 953–973.Franck, Raphael, and Ilia Rainer. 2012. “Does the Leader’s Ethnicity Matter? Eth-nic Favoritism, Education, and Health in Sub-Saharan Africa.” American Po-litical Science Review 106 (02): 294–325.Freitag, Sandria B. 1991. “Crime in the social order of colonial North India.”Modern Asian Studies 25 (02): 227–261.Gordon, Nora. 2004. “Do federal grants boost school spending? Evidence fromTitle I.” Journal of Public Economics 88 (9): 1771–1792.147Goyal, Aparajita, and Jean Dre`ze. 2003. “Future of Mid-Day Meals.” Economicand Political Weekly 38 (44): 4673–4683.Hajnal, Zoltan L. 2001. “White residents, black incumbents, and a declining racialdivide.” In American Political Science Association, 95:603–617. 03. Cam-bridge Univ Press.Hardy, Peter. 1972. The Muslims of British India. 13. London: Cambridge Univer-sity Press.Hasan, Mushirul. 1980. “Communalism in the Provinces: A Case Study of Bengaland the Punjab, 1922-26.” Economic and Political Weekly: 1395–1406.Hnatkovska, Viktoria V, Amartya Lahiri, and Sourabh Paul. 2012. “Castes and La-bor Mobility.” American Economic Journal: Applied Economics 4 (2): 274–307.Hodler, Roland, and Paul A Raschky. 2014. “Regional favoritism*.” The Quar-terly Journal of Economics: qju004.Iyer, Lakshmi. 2010. “Direct versus indirect colonial rule in India: Long-termconsequences.” The Review of Economics and Statistics 92 (4): 693–713.Iyer, Lakshmi, and Abhijit Banerjee. 2005. “History, Institutions, and EconomicPerformance: The Legacy of Colonial Land Tenure Systems in India.” Amer-ican economic review 95 (3): 1190–1213.Iyer, Lakshmi, Anandi Mani, Prachi Mishra, and Petia B Topalova. 2011. “Thepower of political voice: Women’s political representation and crime in In-dia.” Harvard Business School BGIE Unit Working Paper, nos. 11-092.Jha, Saumitra. 2013. “Trade, institutions, and ethnic tolerance: Evidence fromSouth Asia.” American political Science review 107 (04): 806–832.Jha, Shikha, and Bharat Ramaswami. 2010. “How can food subsidies work bet-ter? Answers from India and the Philippines.” Asian Development Bank Eco-nomics Working Paper Series, no. 221.148Johnson, Doug. 2009. “Can Workfare Serve as a Substitute for Weather Insurance?The Case of NREGA in Andhra Pradesh.” The Case of NREGA in AndhraPradesh (December 17, 2009).Kabir, Humayun. 1969. Muslim Politics, 1906-47: And Other Essays. Calcutta:Firma KL Mukhopadhyay.Kasara, Kimuli. 2007. “Tax me if you can: Ethnic geography, democracy, andthe taxation of agriculture in Africa.” American Political Science Review 101(01): 159–172.Keefer, Philip. 2002. “Clientelism, Credibility and Democracy.” In Mimeo, De-velopment Research Group, The World Bank. Citeseer.Keefer, Philip, and Stuti Khemani. 2005. “Democracy, public expenditures, andthe poor: understanding political incentives for providing public services.”The World Bank Research Observer 20 (1): 1–27.Khalidi, Omar. 2006. Muslims in Indian economy. Three Essays Collective.Khemani, Stuti. 2004. “Political cycles in a developing economy: effect of elec-tions in the Indian states.” Journal of Development Economics 73 (1): 125–154.Khera, Reetika. 2011. “Trends in diversion of PDS grain.” Economic and PoliticalWeekly 46 (21): 21.Kremer, Michael, Nazmul Chaudhury, F Halsey Rogers, Karthik Muralidharan,and Jeffrey Hammer. 2005. “Teacher absence in India: A snapshot.” Journalof the European Economic Association 3 (2-3): 658–667.Krishnan, Nandini. 2007. “Political Reservations and Rural Public Good Provi-sion in India.” Working Paper, Boston University.Kudamatsu, Masayuki. 2009. “Ethnic favoritism: Micro evidence from Guinea.”IIES Working Paper.149Kumar, P. 2005. “Communal Riots in Mau Nath Bhanjan,” edited by Steven IWilkinson, 244–271. Religious Politics and Communal Violence. OxfordUniversity Press.Kuran, Timur, and Anantdeep Singh. 2013. “Economic Modernization in LateBritish India: Hindu-Muslim Differences.” Economic Development and Cul-tural Change 61 (3): 503–538.La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert WVishny. 1997. “Legal determinants of external finance.” Journal of Finance:1131–1150.Lange, Matthew K. 2004. “British colonial legacies and political development.”World Development 32 (6): 905–922.Lange, Matthew, and Andrew Dawson. 2009. “Dividing and ruling the world? Astatistical test of the effects of colonialism on postcolonial civil violence.”Social forces 88 (2): 785–817.Levitt, Steven D. 1996. “How do senators vote? Disentangling the role of voterpreferences, party affiliation, and senator ideology.” The American EconomicReview: 425–441.Maiorano, Diego. 2014. “The Politics of the Mahatma Gandhi National RuralEmployment Guarantee Act in Andhra Pradesh.” World Development 58:95–105.McLane, John R. 1965. “The decision to partition Bengal in 1905.” Indian Eco-nomic & Social History Review 2 (3): 221–237.Mehta, Asoka, and Achyut Patwardhan. 1942. The Communal Triangle in India.Kitabistan.Milligan, Kevin, and Michael Smart. 2005. “Regional grants as pork barrel poli-tics.” CES-ifo Working Paper, no. 1453.Mitra, Anirban, and Debraj Ray. 2014. “Implications of an Economic Theory ofConflict: Hindu-Muslim Violence in India.” Journal of Political Economy122 (4): 719–765.150Mookherjee, Dilip. 2014. “MNREGA: Populist leaky bucket or successful anti-poverty programme?” Ideas for India.Mukundan, Mullikottu-Veettil, and Mark Bray. 2006. “The decentralisation of ed-ucation in Kerala State, India: Rhetoric and reality.” In Decentralisation andPrivatisation in Education, 111–131. Springer.Munshi, Kaivan, and Mark Rosenzweig. 2006. “Traditional Institutions Meet theModern World: Caste, Gender, and Schooling Choice in a Globalizing Econ-omy.” The American Economic Review 96 (4): 1225–1252.Nayak, Radhika, NC Saxena, and John Farrington. 2002. “Reaching the Poor: TheInfluence of Policy and Administrative Processes on the Implementation ofGovernment Poverty Schemes in India.” Working Paper No. 175, OverseasDevelopment Institute, London.Nye, John VC, Ilia Rainer, and Thomas Stratmann. 2014. “Do Black Mayors Im-prove Black Relative to White Employment Outcomes? Evidence from LargeUS Cities.” Journal of Law, Economics, and Organization: ewu008.Olsson, Ola. 2009. “On the democratic legacy of colonialism.” Journal of Com-parative Economics 37 (4): 534–551.Pande, Rohini. 2003. “Can mandated political representation increase policy in-fluence for disadvantaged minorities? Theory and evidence from India.” TheAmerican Economic Review 93 (4): 1132–1151.Rao, M Govinda, and Nirvikar Singh. 2003. “The political economy of center-state fiscal transfers in India.” WORLD BANK TECHNICAL PAPER: 69–124.Ravallion, Martin, and Michael Lokshin. 2004. “Gainers and losers from tradereform in Morocco.” World Bank policy research working paper, no. 3368.Ray, Anil Baran. 1977. “Communal Attitudes to British Policy: The Case of thePartition of Bengal 1905.” Social Scientist: 34–46.Rehavi, M Marit. 2007. “Sex and politics: Do female legislators affect state spend-ing?” Working Paper, University of Michigan.151Robinson, Francis. 2007. Separatism Among Indian Muslims The Politics of theUnited Provinces’ Muslims, 1860–1923. Vol. 16. Cambridge University Press.Sachar Committee. 2006. Social, economic and educational status of the Muslimcommunity. New Delhi: Prime Minister’s High Level Committee, Govern-ment of India.Sahoo, Sarbeswar. 2008. “Ethno-Religious Identity and Sectarian Civil Society:A Case from India.” Studies in Ethnicity and Nationalism 8 (3): 453–480.Sass, Tim R, and Stephen L Mehay. 2003. “Minority representation, electionmethod, and policy influence.” Economics & Politics 15 (3): 323–339.Singh, Nirvikar. 2008. “Decentralization And Public Delivery Of Health Care Ser-vices In India.” Health Affairs 27 (4): 991–1001.Svedberg, Peter. 2012. “Reforming or replacing the public distribution systemwith cash transfers.” Economic and Political Weekly 47 (7): 53–62.Talbot, Ian. 2007. “Religion and violence: the historical context for conflict inPakistan.” Religion and Violence in South Asia: Theory and Practice, Rout-ledge, London: 154–172.Thorat, Sukhdeo, and Joel Lee. 2005. “Caste discrimination and food securityprogrammes.” Economic and Political Weekly: 4198–4201.Topalova, Petia. 2007. “Trade liberalization, poverty and inequality: Evidencefrom Indian districts.” In Globalization and Poverty, 291–336. Universityof Chicago Press.Tribune News Service. 2005. “Doctor on emergency duty found absent.” Decem-ber 13. http://www.tribuneindia.com/2005/20051213/cth3.htm.Van der Veer, Peter. 1994. Religious Nationalism: Hindus and Muslims in India.Berkeley, CA: University of California Press.Varshney, Ashutosh, and Steven Wilkinson. 2006. Varshney-Wilkinson Dataset onHindu-Muslim Violence in India, 1950-1995, Version 2. Ann Arbor, MI, -02-17. http://doi.org/10.3886/ICPSR04342.v1.152Washington, Ebonya L. 2008. “Female Socialization: How Daughters Affect TheirLegislator Fathers.” American Economic Review 98 (1): 311–32.Wilkinson, Steven I. 2006. Votes and violence: Electoral competition and ethnicriots in India. Cambridge University Press.153

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0166458/manifest

Comment

Related Items