UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Three essays on female and child outcomes in India Bhattacharjee, Shampa 2006

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

Item Metadata

Download

Media
24-ubc_2016_may_bhattacharjee_shampa.pdf [ 662.06kB ]
Metadata
JSON: 24-1.0300647.json
JSON-LD: 24-1.0300647-ld.json
RDF/XML (Pretty): 24-1.0300647-rdf.xml
RDF/JSON: 24-1.0300647-rdf.json
Turtle: 24-1.0300647-turtle.txt
N-Triples: 24-1.0300647-rdf-ntriples.txt
Original Record: 24-1.0300647-source.json
Full Text
24-1.0300647-fulltext.txt
Citation
24-1.0300647.ris

Full Text

Three Essays on Female and Child Outcomes inIndiabyShampa BhattacharjeeB. 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)February 2016© Shampa Bhattacharjee, 2016AbstractPoor child health outcomes and high fertility rates are viewed as major obstaclesto development in most developing countries. Chapters 2 and 3 of my thesisinvestigate the determinants of these outcomes in the Indian context.The second chapter looks at the impact of political cycles on infant mortalityin India. This study shows that children born 0-12 months before scheduled stateassembly elections have 13.4% lower mortality risks as compared to children notborn before scheduled elections and that the effect is higher for children bornin more politically competitive regions. In addition, the chapter presents someevidence that mothers who gave birth before elections have more regular antenatalcheck-ups and at least one tetanus injection during their pregnancy. Children bornbefore scheduled assembly elections are also less likely to suffer from low birthweight.My third chapter tests the effect of female employment on fertility in India.The results show that female employment in manufacturing has a negative impacton fertility once the endogeneity in female employment is accounted for. Howeverfemale employment in agriculture and aggregate female labour force participationiihas no such effects.The fourth chapter is joint work with Dr. Viktoria Hnatkovska and Dr. AmartyaLahiri. This chapter examines the evolution of gender gaps in India between 1983and 2010 in education, occupation choices and wages. The chapter shows that thegaps have shrunk quite sharply between men and women in education and choiceof occupations and wages. The gaps have narrowed most sharply for the youngestcohorts in the workforce, suggesting that measured gaps will continue to declineover the next two decades.iiiPrefaceThe first two chapters of this thesis are original, unpublished, independent workby the author, Shampa Bhattacharjee. The third chapter is joint work with Dr.Viktoria Hnatkovska and Dr. Amartya Lahiri at the at the Vancouver Schoolof Economics, University of British Columbia. A version of the third chapterhas been published [Bhattacharjee, Shampa, Viktoria Hnatkovska and AmartyaLahiri. 2015. “The Evolution of Gender Gaps in India”. In India Policy Forum2014-15, volume 11, Brookings Institution and National Council of Applied Eco-nomic Research, edited by Subir Gokarn, Arvind Panagariya, and Shekhar Shah,119-156. New Delhi, Sage publications.]. I did the most of the empirical analysisand contributed to preparing parts of the manuscript.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 The Timing of Elections and Infant Mortality: Evidence from India 42.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . 102.3 Institutional Background . . . . . . . . . . . . . . . . . . . . . . 122.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13v2.5 Empirical Strategy and Results . . . . . . . . . . . . . . . . . . . 172.5.1 Effect of Elections on Infant Mortality . . . . . . . . . . . 172.5.2 Neonatal and 2-12 Months Mortality . . . . . . . . . . . . 212.5.3 Closeness to Election and Infant Mortality . . . . . . . . . 222.5.4 Infant Mortality Just Before and Just After Scheduled Elec-tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232.5.5 Margin of Victory as a Determinant of Political Cycle . . . 242.5.6 Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . 252.6 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . 282.6.1 Placebo Tests . . . . . . . . . . . . . . . . . . . . . . . . 282.6.2 Birth of More Preferred Children and the Timing of Elec-tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282.6.3 Excluding Kerala . . . . . . . . . . . . . . . . . . . . . . 292.6.4 Including Politically Non-competitive States . . . . . . . . 292.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303 Fertility Rate and Female Employment in India . . . . . . . . . . . 513.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513.2 Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . 543.3 Trade Reform in India . . . . . . . . . . . . . . . . . . . . . . . . 563.4 Data Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . 583.5 Empirical Strategy and Results . . . . . . . . . . . . . . . . . . . 603.5.1 Effect of Female Labor Force Participation on Fertility . . 60vi3.5.2 Effect of Female Employment in Agriculture and Manu-facturing on Fertility . . . . . . . . . . . . . . . . . . . . 623.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 634 The Evolution of Gender Gaps in India . . . . . . . . . . . . . . . . 684.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 684.2 Empirical Regularities . . . . . . . . . . . . . . . . . . . . . . . . 744.2.1 Education Attainment . . . . . . . . . . . . . . . . . . . . 764.2.2 Occupation Choices . . . . . . . . . . . . . . . . . . . . . 794.3 Wage Outcomes and Gender Differences . . . . . . . . . . . . . . 824.4 The Young . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 874.5 Female Labor Force Participation . . . . . . . . . . . . . . . . . . 894.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 905 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110Appendix A Appendix to Chapter 2 . . . . . . . . . . . . . . . . . . . . 119Appendix B Appendix to Chapter 3 . . . . . . . . . . . . . . . . . . . . 128viiList of TablesTable 2.1a Summary Statistics (Child Characteristics) . . . . . . . . . . . 33Table 2.1b Summary Statistics (Electoral Variables) . . . . . . . . . . . . 34Table 2.1c Summary Statistics (Mother’s Characteristics) . . . . . . . . . 34Table 2.2 Elections and Infant Mortality . . . . . . . . . . . . . . . . . . 36Table 2.3 Accounting for Midterm Elections . . . . . . . . . . . . . . . . 37Table 2.4 Neonatal and Infant Mortality . . . . . . . . . . . . . . . . . . 38Table 2.5 Closeness to Election and Infant Mortality . . . . . . . . . . . 39Table 2.6 Slightly Before and After Scheduled Election . . . . . . . . . . 40Table 2.7 Role of Political Competition . . . . . . . . . . . . . . . . . . 42Table 2.8 Antenatal Visits . . . . . . . . . . . . . . . . . . . . . . . . . 43Table 2.9 Tetanus During Pregnancy . . . . . . . . . . . . . . . . . . . . 44Table 2.10 Low Birth Weight . . . . . . . . . . . . . . . . . . . . . . . . 45Table 2.11 Robustness check: Placebo Tests . . . . . . . . . . . . . . . . 46Table 2.12 Probability of Birth of Female Child . . . . . . . . . . . . . . 47Table 2.13 Excluding Kerala . . . . . . . . . . . . . . . . . . . . . . . . . 48Table 2.14 Election and Infant Mortality (All States) . . . . . . . . . . . . 49viiiTable 2.15 Role of Political Competition (All States) . . . . . . . . . . . . 50Table 3.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . 65Table 3.2 Female Labor Force Participation and Total Fertility Rate . . . 66Table 3.3 Effect of Female Employment in Agriculture and Manufactur-ing on Total Fertility Rate . . . . . . . . . . . . . . . . . . . . 67Table 4.1 Sample Summary Statistics . . . . . . . . . . . . . . . . . . . 92Table 4.2 Labor Market Characteristics by Gender: Rural and Urban Work-ers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94Table 4.3 Education Gaps: Years of Schooling . . . . . . . . . . . . . . 95Table 4.4 Marginal Effect of Female Dummy on Education Categories . 97Table 4.5 Marginal Effect of Female Dummy on Occupational Categories. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99Table 4.6 Changes in the Gender Wage Gap . . . . . . . . . . . . . . . . 101Table 4.7 Decomposition of the Changes in the Wage Gap . . . . . . . . 102Table A.1 Elections and Infant Mortality (Changing Omitted Category) . 122Table A.2 Elections and Infant Mortality (With Current GDP as Control) . 123Table A.3 Elections and Infant Mortality (Including Multiple Births) . . . 124Table A.4 Elections and Infant Mortality (Including NFHS I) . . . . . . . 125Table A.5 Presence of Doctor During Delivery . . . . . . . . . . . . . . . 126Table A.6 Mechanisms (All states) . . . . . . . . . . . . . . . . . . . . . 127ixList of FiguresFigure 2.1 Election Cycle Dummies . . . . . . . . . . . . . . . . . . . . 32Figure 2.2 Election and Infant Mortality . . . . . . . . . . . . . . . . . . 35Figure 2.3 Role of Political Competition . . . . . . . . . . . . . . . . . . 41Figure 3.1 Decline in the Average Tariff in Agriculture and Manufacturing 64Figure 4.1 Gender Gaps: Labor Market Participation Rates . . . . . . . . 93Figure 4.2 Distribution of Workforce Across Education Categories . . . 96Figure 4.3 Distribution of Workforce Across Occupation Categories . . . 98Figure 4.4 Gender Wage Gaps Since 1983 . . . . . . . . . . . . . . . . . 100Figure 4.5 Gender Wage Gaps by Education and Occupation Categories . 103Figure 4.6 Education Gaps in Years by Birth Cohorts . . . . . . . . . . . 104Figure 4.7 Gap in Years of Education: 16-25 Year Olds . . . . . . . . . . 105Figure 4.8 Occupational Distribution of 16-25 Year Olds . . . . . . . . . 106xAcknowledgmentsI am greatly indebted to my supervisors, Dr. Siwan Anderson and Dr. PatrickFrancois for their constant support and guidance throughout the course of theprogram. It would have been impossible to finish my PhD without their help.They always encouraged me which gave me confidence at times when things gottough. I would also like to express my gratitude to my committee member, Dr.Marit Rehavi for her supervision and invaluable advice regarding my research.I am grateful to my co-authors Dr. Viktoria Hnatkovska and Dr. AmartyaLahiri for working with me. I am also thankful to Dr. Ashok Kotwal, who hasconstantly helped and guided me throughout the course of my PhD.I want to thank Dr. Mukesh Eswaran, Dr. Nicole Fortin, Dr. Keith Head,Dr. Vadim Marmer and the participants at the UBC Economics Empirical Lunch,UBC Economics PhD job market workshop, Econ 640 workshop and CanadianEconomics Association conference for their helpful comments and suggestions.I am indebted to the U.S. Census Bureau for making their software PopulationAnalysis System (PASEX) available to me. I am also thankful to Maureen Chinfor administrative support throughout my PhD.xiI am grateful to my parents and my brother for their continuous support andencouragement. Finally I want to especially thank my partner, Arka for sharinghis tariff dataset with me and for all his help including valuable comments andsuggestions which greatly improved my work.xiiChapter 1IntroductionPoor child health outcomes coupled with rapid population growth are viewed asmajor obstacles to development in most developing countries. The problems be-come more acute in the Indian context, where infant mortality was 46 per thou-sand live births in 2013. Not only is this figure higher than developed countries(Japan: 2 per 1000, United Kingdom 4 per 1000, United States 6 per 1000), butit is also higher in comparison to other south Asian countries like Bangladesh (39per 1000), Nepal (36 per 1000) and Bhutan (34 per 1000). Likewise, high fer-tility is an important concern in India, where the total fertility rate was 2.5 perwoman in 2013. This is again higher compared to Bangladesh (2.2), Nepal(2.3)and Bhutan (2.2). Chapters 2 and 3 of my thesis investigate the determinants ofthese outcomes in the Indian context.Chapter 2 of my thesis tests the impact of political cycles on development out-comes in India. It has been widely established in the political economy literature1that politicians are concerned about winning elections and thus they adopt poli-cies to maximize their chances of being elected. Such opportunistic behaviouron the part of politicians often results in the creation of favorable economic con-ditions before elections (Nordhaus (1975); Rogoff (1990); Alesina and Roubini(1992)). This can have large implications for the welfare of individuals in devel-oping countries since the poor electorate in these countries is highly sensitive toeconomic fluctuations.In this chapter, I focus on child health outcomes; infant mortality in particular.Infant mortality is an extremely important issue from the point of view of devel-oping countries and is associated with a number of factors like maternal health,quality and access to medical care and socioeconomic conditions. According toCutler, Deaton, and Lleras-Muney (2006) deaths in childhood comprise a signif-icant percentage of all deaths in poor countries (30%) as compared to rich coun-tries (less than 1%). Given that child health is an acute problem in poor countries,a politician may want to exert higher effort to improve health conditions beforeelections in order to secure more votes. Thus public health systems may functionmore efficiently during election years. The existence of political cycles also im-plies that the overall economic conditions improve during election years and thiscan show up in child health. Various development schemes (like temporary em-ployment generation schemes) may work more efficiently under pressure from thegovernment which translates into higher family income and improved nutritionalstatus leading to better health outcomes.This chapter shows that children born 0-12 months before scheduled state as-2sembly elections have 13.4% lower mortality rate as compared to children notborn before scheduled elections. My results also show that the effect of the timingof scheduled elections is higher for children born in more politically competitiveregions. In addition, the chapter presents some evidence that health care utilizationimproves for mothers who give birth before scheduled elections. Children bornbefore scheduled assembly elections are also less likely have low birth weight.Chapter 3 tests the effect of female labor force participation on fertility rates inIndia using district level data between the period 1987-2001. I have also computedthe effects of female employment in the agricultural and the manufacturing sectorson total fertility rates. The results show that female labor force participation inmanufacturing has a significant negative impact on fertility once the endogeneityin female labor force participation is accounted for. However female employmentin agriculture and overall female labor force participation has no significant impacton fertility.Chapter 4 is joint work with Dr. Viktoria Hnatkovska and Dr. Amartya Lahiri.In this chapter we have examined the evolution of gender gaps in India between1983 and 2010 in education, occupation choices and wages. We find that thegender gaps have shrunk quite sharply in education and choice of occupations.Our analysis also shows that wage gaps have declined across most percentiles ofthe income distribution. The gaps have narrowed most sharply for the youngestcohorts in the workforce suggesting that measured gaps will continue to declineover the next two decades.3Chapter 2The Timing of Elections and InfantMortality: Evidence from India2.1 IntroductionPoliticians are concerned with winning elections and thus they adopt policies tomaximize their chances of being elected. Such opportunistic behaviour on the partof politicians often results in the creation of favorable economic conditions beforeelections (Nordhaus (1975); Rogoff (1990); Alesina and Roubini (1992)). Thiscan have large implications for the welfare of individuals in developing countriessince the poor electorate in these countries is highly sensitive to economic fluctu-ations.In this chapter I investigate the effect of the timing of state assembly electionson child health in India. I focus on child health, in particular infant mortality, for4the following reasons. Firstly, child health is an extremely important issue fromthe point of view of developing countries, where deaths in childhood typicallycomprise a significant percentage of all deaths (roughly 30%) as compared torich countries (less than 1%) (Cutler, Deaton, and Lleras-Muney (2006)). Giventhat child health is such an acute problem in poor countries, a politician maywant to exert higher effort to improve these health conditions before elections inorder to secure more votes. Thus public health systems can potentially functionmore efficiently during election years. Secondly, even if policy makers do nottarget public health directly, the existence of political cycles implies that overalleconomic conditions improve during election years and this can show up in childhealth. Various development schemes (like temporary employment generationschemes) might work more efficiently under pressure from the government whichtranslates into higher family income and improved nutritional status leading tobetter child health outcomes.The existing literature has extensively studied opportunistic political cycles.Some studies have examined the impact of the election cycle on government pol-icy. Veiga and Veiga (2007) showed that Portuguese local governments increasetotal expenditures in the before elections. Other studies have analyzed political cy-cles in macro level outcomes such as inflation, growth or employment (McCallum(1978); Alesina (1988)). However, very few studies have focused on whether theinjection of public services before elections produces better social and economicindividual level outcomes.There is a small but emerging literature on political cycles in India (Cole5(2009), Khemani (2004)). However, their research mostly focused on the effect ofpolitical cycles on government policy. In this chapter, instead of looking at gov-ernment policies, I investigate the effect of state elections on child heath which isoften considered an index of welfare in developing countries.This study also contributes to the literature on the implications of aggregateeconomic conditions on child health. Previous research has found that political cy-cles create temporary economic booms. The results obtained in this chapter implythat the improvement in overall conditions during elections contribute positivelytowards child survival during infancy. The results also seem to suggest that theremight be some long run implications of being born before elections on adult out-comes. There has been evidence of factors that jointly determine infant mortalityand adult life outcomes. For example birth weight in the short run influences in-fant mortality while in the long run has significant impact on adult characteristicslike height, IQ, earnings and education (Black, Devereux, and Salvanes (2007)).However, further research is required on this topic.India is a particularly relevant place to analyze the implications of electioncycles. It is a developing country with high participation in democratic elections.The states in my sample have an average voter turnout of about 58%. Also, Indianstate elections are not perfectly synchronized enabling me to identify the impact ofelections separately from time invariant effects. The constitution of India requiresstate elections to be held every five years. However, there have been some midtermelections where elections were held before the completion of the full five yearterms. These elections take place one, two, three and four years after the previous6election. Since scheduled elections take place after the completion of full fiveyear terms, the electoral calendar is exogenous and known perfectly in advanceby all agents. Hence, opportunistic manipulation by politicians before elections ispossible. On the other hand, midterm elections are more likely to be sudden andthus allow politicians much less time to manipulate policies. The identificationstrategy used in this chapter rests upon comparing the mortality risks of siblingsborn before scheduled elections and those born in off-election years.An advantage of this study compared to previous research on political cyclesin India is that this study uses a significantly larger sample; more than 150,000children spanning over 23 years and covering 91 elections, 58 of which are sched-uled elections1. Moreover, since the data have information on birth histories ofmothers over time, I have been able include mother fixed effects in the estima-tions 2. Thus, the results here essentially compare across children born to thesame mothers at different points of time (those born before scheduled electionyears and those not born before scheduled elections).Using data from the 14 major Indian states which have at least some degreeof political competition; that is, states where an effective opposition party exists,I find that children born 0-12 months before scheduled state assembly electionshave 13.4% lower mortality risks as compared to children not born before sched-uled elections. The timing of midterm elections has no influence on infant mor-1. One problem with Cole (2009) was that it covered only a period of eight years which mightbe considered insufficient for finding the effect of political cycles.2. In Cole (2009) the unit of observation was at the district level analysis and in Khemani(2004) the unit of observation was at the state level. Thus Cole (2009) and Khemani (2004) couldinclude only district and state fixed effects respectively7tality.It can be argued that those children who are born just after scheduled electionshave been exposed to improved prenatal care before elections. Thus these childrenshould also experience some improvement in infant mortality. My results showthat those children, who are born 0-1 month and 0-2 months after elections, gainfrom their exposure to improved prenatal care. However, children born 0-6 monthsafter election do not experience any improvement in infant mortality.It has been emphasized in the literature that political cycles depend on thedegree of political competition (Dahlberg and Johansson (2002), Cole (2009)).Thus, we expect the impact of the timing of scheduled elections on infant mortalityto be higher in more politically competitive regions. My results provide evidenceof such targeting and show that the effect of the timing of scheduled elections ishigher for children born in close election districts3.Infant mortality is sensitive to a number of conditions at the time of birth suchas environment, sanitation, access to clean water, prenatal and neonatal healthservices, calorie intake and diseases (Ross (2006)). Governments can influencehealth conditions though public health initiatives like reducing the absenteeism ofdoctors and nurses in government hospitals, encouraging parents to visit publichealth facilities more frequently, filtering drinking water supplies, building sani-3. Close election districts are those districts where incumbents had a narrow margin of victoryin the previous election. In India, elections are held at constituency level. However, the data usedfor generating the infant mortality variable is observed only at the district level and not at theconstituency level. A district is composed of a number of constituencies (9 on average) and so theelection data has to be aggregated to the district level. Thus, the absolute margin is defined at thedistrict level.8tation systems and draining swamps. Estimating the impact of election timing onthe efficiency of the public health system (for example whether the election timingis synchronized with reduced absenteeism of public health workers) is a direct testof whether elections have a positive impact on government health policy. How-ever, in the absence of data on these variables, I check whether there is improvedhealth care utilization by mothers who gave birth before elections. In particular,I test whether the mothers of children born just before elections had more regularantenatal check-ups and had at least one tetanus injection during pregnancy. Thischapter provides evidence that both of these things in fact occurred.Government initiatives can also improve the nutritional status of children.This can be achieved by reducing leakages in the food security network, gen-erating increased temporary employment opportunities via public works program(Schuknecht (1996)), or by providing direct monetary payments to voters (Akhme-dov and Zhuravskaya (2004)) before elections. This means that consumption islikely to go up before elections. However, consumption data in available fromthe National Sample Survey (NSS) dataset and the survey is conducted every fiveyears. Thus yearly consumption data is not available which makes it difficult tocompare between election years and non-election years. In order to test the im-pact of the timing of elections on the nutritional standards of children, I analyzethe impact of timing of elections on the incidence of low birth weight.This chapter is divided into seven sections. The next section provides a briefreview of the existing literature related to my work. Section 2.3 briefly outlinesthe institutional background. Section 2.4 describes the data used in this study.9Section 2.5 outlines the empirical strategy and discusses the results. Section 2.5.1outlines the main results on the impact of the timing of elections on infant mor-tality. In section 2.5.2, I have estimated the impact of the timing of scheduledelections on neonatal and 2-12 month mortality separately, since in India mostof the deaths during infancy occur in the first month of life. In sections 2.5.3and 2.5.4, I have tried to identify the effect of the timing of elections on mortalityrisks of children born just before or after scheduled elections. Section 2.5.5 showswhether the impact of being born before scheduled elections is higher for elec-torally more competitive districts. Section 2.5.6 discusses some possible mecha-nisms which might describe the main results. I have done some robustness checksin Section 2.6. Section 2.7 finally concludes the chapter.2.2 Literature ReviewThere is substantial research on whether politicians manipulate policy for electoralgains in developed countries. The first models of political cycles were developedby Nordhaus (1975) and Lindbeck (1976). They argue that with myopic votersopportunistic incumbent politicians stimulate the economy before elections. Aseparate set of models by Persson and Tabellini (1990) and Rogoff and Sibert(1988) predict political budget cycles based on rational expectations on the partof voters and asymmetric information between incumbents and voters. In suchcases, policy makers signal their abilities by creating favorable economic situa-tions before elections which lead to the emergence of political cycles.Empirical evidence on opportunistic political cycles in developed countries10is mixed. McCallum (1978), Alesina (1988) and Klein (1996) reject the claimthat the timing of elections influences macroeconomic outcomes such as GDPgrowth and output in the United States. However, Berger and Woitek (1997) andGrier (2008) found that the timing of elections exerts a significant influence onaggregate output for Germany and the United States respectively. Apart frommacroeconomic targets, several authors have claimed that politicians target pol-icy instruments just before elections. Veiga and Veiga (2007) evaluated a panelcomposed of Portuguese municipalities during the 1979-2001 period and identi-fied decreases in budget balances and local taxes and increases in expenditures inelection years.There is also a growing literature documenting the presence of political cy-cles in developing countries. Gonzalez (2002) showed that the Mexican govern-ment systematically used fiscal policy before elections as a means to secure votes.Akhmedov and Zhuravskaya (2004) investigated a panel of local Russian politicaljurisdictions and found an increase in public expenditures before elections and adecrease right afterwards. Drazen and Eslava (2010) showed that infrastructurespending increases prior to elections in Colombia.In the Indian context, Cole (2009) shows that bank lending follows the elec-toral cycle, with agricultural credit increasing by 5-10% points in an election year.His paper also demonstrates that election year credit booms do not affect agricul-tural output. Khemani (2004) developed a career concerns model and empiricallyshowed that state elections in India has a positive effect on public service deliv-11ery 4. She also showed that elections have a negative effect on some commoditytaxes and concluded that fiscal instruments are targeted to provide favors to piv-otal groups of voters. My work is similar in spirit to the growing literature onpolitical cycles in India. However, unlike most of the previous work which ex-plores the manipulation of government policy instruments such as expenditure,taxes and credit during election years, my study analyses the effect of elections onindividual level health outcomes.A number of recent papers have highlighted the role of politicians’ electoralincentives on service delivery which influences public health. Most of these stud-ies have focused on interventions and institutions5 that strengthen the electorate’svoice leading to better public good provision and improved health outcomes (Fuji-wara (2014), Chattopadhyay and Duflo (2004), Besley and Burgess (2001)). Thischapter makes an important contribution to this literature by showing how the tim-ing of policy is determined by the electoral incentives of the politicians and thatthis has a significant impact on health outcomes.2.3 Institutional BackgroundThe constitution of India requires that elections for state assemblies be held at fiveyear intervals. However, unscheduled elections are possible, when alignmentsshift within the ruling party or the coalition government breaks down. Politi-4. She used a specific public service, road construction5. Interventions include changes in electoral rules like the introduction of new voting technolo-gies (Fujiwara (2014)) and mandated political representations (Chattopadhyay and Duflo (2004)).Besley and Burgess (2001) showed that democratic institutions and mass media play a significantrole in increasing the government’s responsiveness to the electorate.12cal pressure from the central government is another important reason for midtermelections. The party governing at the center can dissolve a state legislature follow-ing the imposition of Presidential rule in a state. In the period 1975-1998, therewere 91 state elections. Of these, 33 elections (36%) were unscheduled elections.The presence of midterm elections implies that elections across states do not takeplace at the same time.The constitution of India assigns the powers and functions of the center andstates6. The delivery of public health services is essentially a state responsibil-ity7 (Berman (1998)). In addition, health care workers are almost always stateemployees (Singh (2008)).2.4 DataThe micro-data used in this survey come from the second round of the NationalFamily Health Survey of India (NFHS-2) conducted in 19988. This data-set con-tains complete fertility histories for ever-married women aged 15-49 in 1998-99,including the retrospective time and incidence of child deaths. The data has infor-mation on the district of residence of the household during the time of the survey.6. The central government is responsible and can pass legislation on the services mentionedin the Union list (like defense and foreign affairs). Similarly, state governments have exclusivepowers to pass legislation on services mentioned in the state list (like public order, police, andagriculture). There are also areas of joint jurisdiction of the center and the states like education.These items are mentioned in the concurrent list. The states do have jurisdiction over the con-current list but in the case of conflict between the center and the states, the former has overridingpower.7. The health-related provisions in the union list relate mostly to research, and scientific andtechnical education. The concurrent list includes prevention of infectious diseases from spreadingover state boundaries and other issues with wider national ramifications (Gupta and Rani (2004))8. The last round of the NFHS (NFHS III, 2006) does not have district information and so thesample could not be extended beyond 1998.13While it is possible that the district of birth of the child might not be same as thedistrict of the residence of the household in 1998-99, this is unlikely to be an issuein India. Spatial mobility is low in India(Munshi and Rosenzweig (2006); Desh-ingkar and Anderson (2004); Cutler et al. (2010)) and this is particularly true forwomen after marriage. Migration at the time of marriage is the main reason forgeographical movement among women in India (Rosenzweig and Stark (1989),Deshingkar and Akter (2009)).Using this dataset, we can construct individual-level indicators of infant mor-tality across time. The estimation sample contains more than 175,000 childrenborn to more than 56,000 mothers born over the period 1975-1998 across 14 majorIndian states. Table 2.1a shows that the average infant mortality over the sample is89 per 1000 individuals9. There is substantial variation in infant mortality acrossstates. Kerala has the lowest incidence of infant deaths at 29 per 1000 while theUttar Pradesh has the highest rate at 117 per 1000 The sample averages of otherindividual level controls (listed in Appendix A) are also reported in Table 2.1a.The election data come from the official website of the Election Commissionof India. This data include the identity of the contestant, party affiliation and pollpercentage of the electoral unit 10. Table 2.1b shows the distribution of birthsacross election and non-election years, for both scheduled and all elections. Fig-ure 2.1 captures this cycle, where SE denotes scheduled election and ME denotesmidterm election. 0 indicates children born 0-12 months before elections, -1 in-9. This can be compared with the global average of 63 per 1000 in 199010. Assembly elections are held at the constituency level which is typically smaller than a dis-trict.14dicates children born 12-24 months before an election, -2 indicates children bornmore than 24 months before an election, +1 indicates children born 1-12 monthsafter an election, +2 indicates children born 13-24 months before an election. Byconstruction, the proportion of children born in all categories, except those bornmore than two years before, is higher for all elections as compared to the sched-uled elections. The figures shown in Table 2.1b confirm this. Also by construction,the proportion of children born more than 24 months before scheduled electionsis much higher than children born in the other categories.Table 2.1b also shows a relatively higher proportion of children born 0-12months before scheduled elections. This might imply that the mothers who givebirth during election years are different from mothers who give birth in non-election years. However, comparing across children born to the same motheraddresses this problem.I have also tried to test whether the effect of the timing of elections on infantmortality is higher in more politically competitive regions. I define a region tobe politically competitive if the incumbent party/coalition had a narrow margin ofvictory in the previous election in that region. The definition of regions howeveris a little problematic. In India elections are held at the constituency level andconstituency level information is not available in the NFHS data-set. However,the NFHS data report the district of residence. A district is composed of a numberof constituencies (usually 9). Thus, I have defined the margin of victory of theincumbent ruling party at the district level. This is given by the absolute differencebetween the proportion of seats obtained by the ruling party or ruling coalition15and the proportion of seats obtained by the biggest opposition. Mathematically,the variable measuring political competition is given by:Adst = |pwdst− podst | (2.1)where Adst is the variable measuring political competition in the district d attime t. The variable pwdst and podst measure the proportion of seats obtained bythe ruling party (or ruling coalition) and the biggest opposition respectively, in thelast assembly election, corresponding to time t in district d of state s.The Election Commission of India website provides information on party af-filiations of individual contestants in an election. However, the name of the rulingparty or ruling coalition is not available from the election commission websitewhich is essential for computing the political competition variable. I have usedthe Times of India database to ascertain the name of the party (or all the parties incase of coalition governments) that ruled a state.In order to analyze the possible mechanisms behind the fall in infant mortalityI have estimated the impact of the timing of elections on the following outcomes:the number of antenatal checkups during pregnancy, whether the mother had atetanus injection during pregnancy and the incidence of low birth weight. Infor-mation on these variables is available only for children born after 1995 in NFHS IIdata. Three years is an extremely short period for estimating the impact of electioncycles. At most one election per state will be covered and some states might nothave a single scheduled election during this time. In order to address this problem16of a small sample size, I have used NFHS I (1992-93) in addition to NFHS II forestimating these. Thus, the sample in this case contains information on childrenborn between 1987-1992 from NFHS I and 1995-1998 from NFHS II11.2.5 Empirical Strategy and Results2.5.1 Effect of Elections on Infant MortalityI estimate the effect of being born in an election year on infant mortality. Thebasic estimating equation is:yimdst = α+βEist +φXimdst + τt +µm+ εidst (2.2)Where yimdst is a dummy variable that indicates whether the index child i, bornto mother m, in district d of state s in year t died by the age of 12 months. Eist is adummy variable equal to 1 if the child i is born between 0 and 12 months before anelection. µm and τt denote mother and year of birth fixed effects. Ximdst includesthe controls used throughout the chapter. Child specific controls included in Ximdstare dummies for the order of birth, month of birth of the child, and a dummyvariable indicating whether the child is female. These controls account for thevariation in death risk within children born to the same mother. In addition realstate domestic product lagged by two years and average voter turnout in districtsin the previous election are also included as controls. Real state domestic product11. The main results are also estimated using both rounds of NFHS as shown in Table A.4 givenin Appendix A. The results are consistent with estimation using only NFHS II17controls for the level of prosperity across states12.Including mother fixed effects is particularly important. A problem of com-paring children born in election years with children born in non-election years isthat mothers who give birth just before elections might be different from motherswho give birth in other years. The differences can be due to differential livingstandards, fertility, contraception preferences and awareness of the availability ofhealth-related technology and services. Thus, health improvements of childrenborn before elections might be due to selection issues rather than changes in gov-ernment policies. This will overestimate the effects if relatively better off mothersgive birth in election years as compared to non-election years. Similarly, if poorermothers choose to give birth before elections13, the results will be underestimated.Mother fixed effects take account of the selection issues associated with the typeof mothers who give birth before elections.Here it is important to distinguish between scheduled elections and midtermelections. The scheduled elections are those which are mandated by the Consti-tution of India and occur five years after the previous election. Whereas midtermelections are those that occur one, two, three or four years after the last election(either scheduled or midterm), that is, before the completion of the full term ofthe present elected government in office. The timing of midterm elections is lesslikely to be exogenous. They are also likely to be sudden and so the governmenthas less time to adjust policies. I have estimated the results separately for all elec-12. The results are also robust to the inclusion of contemporaneous GDP as shown in Table A.2in Appendix A.13. This is more likely since these mothers are most in need of resources18tions, midterm elections and scheduled elections. We would expect the coefficientof the election dummy, β in equation (2.2), to be negative and statistically signif-icant for children born before scheduled elections. We also expect smaller effectsfor children born before midterm elections.Columns 1, 3 and 5 of Table 2.2 present the estimates of equation (2.2) forall elections, scheduled elections and midterm elections respectively. Column 1shows that infant mortality is lower for children born between 0-12 months beforeall elections. The result is significant at the 10% level. The fall is much higherfor children born 0-12 months before scheduled elections as shown in Column3. Children born 0-12 months before scheduled elections have over 13% moresurvival chances. The estimate for scheduled election is highly significant at the1% level. Consistent with the fact that midterm elections provide less scope foropportunistic manipulation, the results are insignificant for children born 0-12months before midterm elections as shown in Column 5.I have also estimated the effect of the entire election cycle on infant mortality.The estimating equation is:yimdst =α+β−1E−1ist+β0Eist+β+1E+1ist+β+2E+2ist+γXimdst+δm+τt+ξimdst(2.3)Where Eist is a dummy variable which is equal to 1 if the child i is born 0-12 months before an election, E−1ist is a dummy variable which is equal to 1if the child i is born between 13 and 24 months before an election, E+1ist is a19dummy variable equal to 1 if the child i is born between 1 and 12 months after anelection, E+2ist is a dummy variable equal to 1 if the child i is born between 13and 24 months after an election14. The omitted category includes children born 2or more years before an election15.Figure 2.2 shows the predicted relationship estimated by equation (2.3). Panel(a) shows the relationship for all elections and panel (b) shows the relationship forscheduled elections. While panel (a) shows that infant mortality is more or lessflat over the election cycle for all elections, panel (b) shows that infant mortalityfalls significantly for children born 0-12 months before scheduled electionsColumns 2, 4 and 6 of Table 2.2 show the estimates of equation (2.3) forall elections, scheduled elections and midterm elections respectively. Column 2shows that children born 0-12 months before all elections have lower mortalityrisk. The estimates are all statistically insignificant. On the other hand, childrenborn 0-12 months before scheduled elections have significantly lower mortalityrisk as shown in Column 4. The effect of being born before a midterm election issmall and statistically insignificant.One problem of comparing children born before scheduled election years withchildren not born before a scheduled election, is that the comparison group con-sists of children born before midterm elections and those born in off-electionyears. I have estimated the results dropping all children born 0-12 months be-fore midterm elections so that the control group consists of children born in the14. Sections 2.5.3 and 2.5.4 will show the results using different time frames.15. The results obtained when other categories are treated as omitted categories are shown inTable A.1 in Appendix A.20off-election years. These results are presented in Table 2.3. Columns 1 and 2 ofTable 2.3 correspond to Columns 3 and 4 of Table 2.2. They present the baselineresults. Columns 3 and 4 of Table 2.3 present results from regressions estimatedby excluding children born 0-12 months before midterm elections. It can be seenthat the results are similar in sign and significance to the baseline results. Columns5 and 6 of Table 2.3 show estimates of equation (2.3) using the full sample but con-trolling for midterm elections (that is including a dummy variable equal to 1 if thechild i is born 0-12 months before midterm elections.). The estimates are againsimilar in sign and magnitude to the baseline results.2.5.2 Neonatal and 2-12 Months MortalityInfant mortality can be disaggregated into two components: neonatal mortalityand mortality during the remaining 2-12 months of an infant’s life. Neonatalmortality is mainly influenced by the mother’s health and prenatal care. It is alsoextremely significant for India because most of the deaths during infancy occur inthe first month of life. In my data the average neonatal mortality over the sampleis 58 per 1000 while average infant mortality is 88.9 per 1000.I have estimated the results separately for neonatal and 2-12 months mortality.The results are shown in Table 2.4. The first two columns of Table 2.4 showsthe baseline results for infant mortality and are the same as Columns 3 and 4of Table 2.2. Columns 3 and 4 show estimates for neonatal mortality and thelast two columns contain estimates for 2-12 months mortality. The results showthat children born before scheduled elections gain both from reduced neonatal21and 2-12 month mortality. Children born 0-12 months before scheduled electionyears have a 13% and 14% reduced chance of neonatal and 2-12 month mortalityrespectively.2.5.3 Closeness to Election and Infant MortalityIt is plausible that the children born closer to scheduled elections compared tothose born earlier are more likely to benefit from policy manipulation before elec-tions. In order to test this, I have divided the children born 0-12 months beforescheduled elections into two groups: those who are born 0-6 months before elec-tions and those who are born 6-12 months before elections. I have also regressedinfant mortality on the number of months between birth of a child and the nextscheduled election. The specification for these regressions are similar to equation(2.2).Table 2.5 shows the estimates from the above regression. Column 1 shows theeffect of being born 0-6 months before a scheduled election. The regression esti-mates corresponding to Column 2 include a dummy variable indicating whether achild is born 6-12 months before a scheduled election in addition to the dummyvariable indicating whether a child is born 0-6 months before a scheduled elec-tion. Column 3 is similar to Column 3 of Table 2.2 and shows the effect of beingborn 0-12 months before a scheduled election. The results show that children whoare born 0-6 months before a scheduled election experience a greater fall in infantmortality compared to children born 6-12 months before a scheduled election. Theestimates of Column 4 are obtained by regressing the infant mortality variable on22the difference in months between the birth of a child and the next scheduled elec-tion. Column 4 also shows that children born closer to elections are more likelyto survive their infancy.2.5.4 Infant Mortality Just Before and Just After ScheduledElectionsIt might be argued that the mothers of children born just after scheduled electionshave been exposed to better prenatal care for the most part of their pregnancy.Thus, these children should also experience better health outcomes.I have estimated the effect of being born just before and just after scheduledelections. The results are shown in Table 2.6. The first 3 columns show the impactof being born 0-6, 0-2 and 0-1 months before a scheduled election respectively.Column 4, 5 and 6 show the impact of being born 0-1, 0-2 and 0-6 months aftera scheduled election. The table shows that the effect of being born in electionyears is strongest for children born 0-1 months before scheduled elections fol-lowed by children born 0-1 months after scheduled elections. However, the effectdrops if we consider children born 0-2 months before and 0-2 months after theelection. Again compared to children born 0-2 months before an election, the ef-fect is lower for children born 0-2 months after an election since the cohort born0-2 months after election is less exposed to improved prenatal care. The effectbecomes statistically insignificant for children born 0-6 months after a scheduledelection.232.5.5 Margin of Victory as a Determinant of Political CycleIt is likely that politicians will have an incentive to behave more opportunistically,in terms of manipulating service provision, when their margin of victory is small.To address this, I have examined whether the extent of political cycles dependson the margin of victory enjoyed by the current ruling party or coalition. Asmentioned before, elections in India take place at the constituency level whichhas to be aggregated to the district level to be merged with the individual levelNFHS data. The margin of victory is given by the absolute difference betweenthe proportion of seats obtained by the ruling party/coalition and the proportionof seats obtained by the largest opposition in a district.The simplest test estimating the role of political competition would be to lookat the sign and significance of the interaction between the election dummy and thedistrict margin of victory variable. Thus, I have estimated the following equation:Yimdst = α+βEist +µEist ∗Adst +θAdst + γXimdst +δm+ τt +ζimdst (2.4)If infant mortality falls in election years and the fall is greater in areas ofhigher political competition, we expect the coefficient on the election dummy, β ,to be negative and statistically significant and for µ to be positive and statisticallysignificant.Figure 2.3 shows the predicted relationship between infant mortality and theabsolute margin for children born before a scheduled election and children not24born before scheduled elections. While panel (a) shows the relationship betweeninfant mortality and the absolute margin of victory for children born 0-12 monthsbefore scheduled elections, panel (b) shows the same relationship for childrenborn 0-6 months before scheduled elections. Figure 2.3 shows that among childrenborn before scheduled elections, infant mortality is lower in districts where thestate ruling party had a narrower margin of victory in the previous election. Infantmortality is flat over the absolute margin for children not born before scheduledelections.Table 2.7 shows the estimates of equation (2.4). Columns 1 and 2 show theresults for children born 0-12 months before a scheduled election and Columns 3and 4 show the results for children born 0-6 months before scheduled elections.The results show that the fall in infant mortality before a scheduled election ishigher for children born in districts where the previous election was particularlyclose. This is true for both children born 0-12 months before an election andchildren born 0-6 months before an election. However, the effect of targeting ismore pronounced for children born 0-6 months before a scheduled election, asshown by the coefficient of the interaction term.2.5.6 MechanismsThe government can influence infant mortality in a number of ways. Firstly, Thepublic health system might work more efficiently and so the usage of medicalservices might go up. I have estimated the impact of being born before scheduledelections on the number of antenatal checkups of the mother during pregnancy25and whether the mother had at least one tetanus injection during pregnancy16.Secondly children born just before elections are likely to be better nourished.This might be the result of two factors. Firstly, if government job schemes (foodwork schemes) run more efficiently before elections, employment rates will go upin years preceding elections. The increased income is likely to have an impact onchild health and child nutrition. Secondly, the public distribution system whichis the food security network in India might function better before scheduled elec-tions. To estimate the impact of the timing of elections on nutritional status ofchildren I estimate the impact of the timing of elections on the incidence of lowbirth weight.As pointed out earlier, the information on these variables is not available forthe majority of children born to a mother. Thus, I could not include mother fixedeffect in these regressions. I have included district fixed effects and some addi-tional controls like parental education, membership in a disadvantaged socioeco-nomic group and a dummy variable for urban residence in these regressions.Tables 2.8-2.10 show the estimated results. Table 2.8 shows the results forantenatal visits during pregnancy by mothers of children born 0-6 and 0-12 monthsbefore elections. Columns 1 and 2 of Table 2.8 show the estimates for childrenborn 0-6 months before pregnancy while Columns 3 and 4 contain estimates forchildren born 0-12 months before an election. Columns 1 and 3 show the baseline16. I have shown the impact of the timing of scheduled elections on the presence of doctorsduring delivery in Table A.5 of Appendix A. However due to the discrepancies in the definitionof this variable between the coding documents of NFHS I, I had to use data from NFHS II only.Thus the sample is further small in this case and consists of children born between 1995-1998.26estimates without interaction effects. Columns 2 and 4 include the interaction termbetween the election dummy and the absolute margin. The estimates of Columns1 and 3 show that mothers of children born before scheduled elections made morefrequent antenatal visits during pregnancy. Columns 2 and 4 show that the impactis higher in politically more competitive districts. The estimates also show thatthe impact on antenatal visits is higher for children born 0-6 months before anelection as compared to children born 0-12 months before an election.Table 2.9 shows that results for the probability of having at least one tetanusinjection during pregnancy. Column 1 shows that mothers of children born 0-6months before scheduled elections were more likely to have at least one tetanus in-jection during pregnancy. The effect is insignificant for children born 0-12 monthsbefore an election. Columns 2 and 4 however show evidence of targeting for both0-6 and 0-12 months.Table 2.10 shows the impact of being born in election years on the probabilityof having a low birth weight. The estimates show that the probability of low birthweight falls just before elections. The effect becomes insignificant if we considerchildren born 0-12 months before an election. The coefficient of the interactionbetween the election dummy and the margin of victory is however insignificantfor both children born 0-12 and 0-6 months before elections.272.6 Robustness Checks2.6.1 Placebo TestsI have done a falsification exercise to check the robustness of my results. As-suming that the election took place 12 months, 24 months and 36 months beforethe real election, I have estimated equations (2.2) and (2.4). If the previous esti-mates were driven by pre-existing state-specific trends, then the estimates of theseplacebo treatment effects on infant mortality would be similar in sign and mag-nitude to the main estimates and statistically significant. Table 2.11 shows theestimated results. The coefficients on all of the placebo treatment dummies aresmall and insignificant2.6.2 Birth of More Preferred Children and the Timing ofElectionsI also explore the possibility that my results are driven by the gender compositionof births. Given the extent of son preference that exists in India (Pande and Astone(2007); Clark (2000)), the reduction in infant mortality before scheduled electionscould be driven by the fact that families have more boys then. I have estimated theimpact of being born 0-6 or 0-12 months before an election on the sex of the child.Table 2.12 shows the estimated results. The results are statistically insignificant.This implies that the gender composition remains unchanged for children bornclose to elections.282.6.3 Excluding KeralaThe reduction in infant mortality before scheduled elections should be higher instates which have low infant mortality rates compared with states in which there islittle scope for reduction. In my sample one state, Kerala has significantly lowerinfant mortality rates as compared to the others. The infant mortality rate in Keralais 29 per 1000 compared to the sample average of 89 per 1000. All other statesexcept Kerala have infant mortality rates over of 60 per 1000. Thus, the effect ofthe timing of elections should be higher if Kerala is excluded from analysis. Table2.13 shows the results excluding Kerala. It can be seen that the magnitude of theestimates improves with the exclusion of Kerala.2.6.4 Including Politically Non-competitive StatesThe main argument for the decline in infant mortality just before elections is thatpoliticians are opportunistic and want to remain in power. Thus, we expect theeffect of the timing of elections on infant mortality to be non-existent or weak instates with very little political competition. For the 14 states considered so far, thegovernment changed at least once during the sample period.17 I have estimatedthe results including all 15 major states of India. Table 2.14 presents results withall states using specification 2.2 and 2.3. Columns 1 and 2 of Table 2.14 repeatColumns 3 and 4 of Table 2.2. Column 3 and 4 of Table 2.14 shows the results17. Among the 15 major states in India, West Bengal is the only state where government changedjust once during the sample period. The left front came into power in West Bengal in 1977 andremained in power until 2011. Thus, one party ruled for almost the entire sample period (1975-1998). West Bengal has very little political competition and so we expect political cycles to besignificantly muted there.29with all states. As expected, the magnitude and the level of significance of theestimates drop if we consider all states.If it is indeed the lack of political competition that results in the estimatesbecoming insignificant if we consider all states, then including the interactionterm between the election dummy and the political competition variable as inspecification (2.4) should return significant estimates. Table 2.15 presents theestimates of equation (3) for all states. The results are statistically significantonce again.182.7 ConclusionThe economics and political science literatures provide evidence that politicianshave strong incentives to improve the economic conditions of voters in electionyears. These incentives to provide special favors can be higher in more politicallycompetitive regions. However, there has been very little empirical research on thewelfare implications of these political cycles. The results presented in this chaptershow that the impacts of electoral cycles are not confined to economic policiesand macroeconomic outcomes alone. Rather, they can have important effects onindividual level developmental outcomes.This chapter shows that infant mortality is significantly lower for children bornjust before scheduled elections in states with at least some political competition.The results further show that the reduction in infant mortality is higher in politi-18. The regressions for mechanisms are also estimated using data from all states. The resultsare presented in Table A.6 in Appendix A and the results are similar to the results with only thepolitically competitive states.30cally more competitive districts.My results also provide evidence that medical care utilization goes up beforeelections. Mothers of children born just before elections are likely to have moreregular antenatal checkups and have at least one tetanus injection during preg-nancy. Antenatal visits can have significant positive effects on both mother andchild health. Kitzman et al. (1997) showed that increased consultation with nursesduring pregnancy reduces pregnancy induced hypertension and childhood injuries.Children born just before elections are also less likely to be of low birth weight.Birth weight determines a number of adult life outcomes like educational attain-ment and earnings. Thus, my results suggest that periods of prosperity created bypolitical cycles can potentially have long-term impacts on the lives of individualsin poor countries.Health in general and particularly child health is an extremely important is-sue in developing countries. Child deaths occur due to insufficient health carefacilities or inadequate nutrition (Jones et al. (2003)). The availability of healthcare facilities and nutrition both depend on government policies and thus politi-cal variables play an important role in influencing child survival. These variablescan be manipulated by politicians to signal their ability to improve the economicconditions of voters before elections, or simply because voters are myopic. Myresults are consistent with the maximization of political gains by politicians lead-ing to extremely different results for similar children (born to the same mother)separated only by their timing of birth.31Figure 2.1: Election Cycle DummiesSE+1ME+20SE-1 0+1-2SE ME0 +2SE0 +1 +2 -2 -1 0Scheduled Election CycleAll Election Cycle-1Notes: The figure shows creation of the election dummies. The top figure corresponds to all elections and the bottomfigure corresponds to scheduled elections.32Table 2.1a: Summary Statistics (Child Characteristics)Variable Mean (Std. Dev.) NInfant Mortality (Scaled 0-100) 8.923 ( 28.510) 175804Neonatal Mortality (Scaled 0-100) 5.814 (23.402) 1758042-12 month Mortality (Scaled 0-100) 3.11 (17.359) 175804Female Child 0.48 (0.5) 175804Birth Order 1 0.278 (0.448) 175804Birth Order 2 0.246 (0.431) 175804Birth Order 3 0.184 (0.387) 175804Birth Order 4 0.121 (0.327) 175804Birth Order 5 0.075 (0.264) 175804Birth Order 6 0.045 (0.208) 175804Birth Order 7 0.025 (0.157) 175804Birth Order 8 0.014 (0.115) 175804Birth Order 9 0.007 (0.083) 175804Birth Order 10 0.005 (0.073) 175804Month of Birth 1 0.068 (0.253) 175804Month of Birth 2 0.063 (0.243) 175804Month of Birth 3 0.078 (0.269) 175804Month of Birth 4 0.076 (0.265) 175804Month of Birth 5 0.083 (0.276) 175804Month of Birth 6 0.087 (0.282) 175804Month of Birth 7 0.087 (0.282) 175804Month of Birth 8 0.106 (0.308) 175804Month of Birth 9 0.089 (0.285) 175804Month of Birth 10 0.097 (0.296) 175804Month of Birth 11 0.088 (0.284) 175804Month of Birth 12 0.076 (0.266) 175804Notes: Standard deviations are in parentheses. Infant mortality, Neo-natalmortality and 2-12 month mortality are defined as percentages. Female de-notes the gender of the child. Month of birth, order of birth refer to the monthand order of birth of a child. Sample includes children born in the period1975-1998 from 14 major states in India.33Table 2.1b: Summary Statistics (Electoral Variables)Variable Mean (Std. Dev.) NBorn 0-12 Months before Election 0.265 (0.441) 175804Born 13-24 Months before Election 0.182 (0.386) 175804Born 1-12 Months after Election 0.226 (0.418) 175804Born 13-24 Months after Election 0.208 (0.406) 175804Born more than 24 Months before Election 0.120 (0.325) 175804Born 0-12 Months before Scheduled Election 0.183 (0.387) 175804Born 13-24 Months before Scheduled Election 0.147 (0.355) 175804Born more than 24 Months before Scheduled Election 0.434 (0.496) 175804Born 1-12 Months after Scheduled Election 0.119 (0.324) 175804Born 13-24 Months after Scheduled Election 0.125 (0.331) 175804Absolute Margin 0.497 (0.285) 175804Average District Turnout 0.583 (0.108) 175804Notes: Standard deviations are in parentheses. Born m to n months before or after election isa dummy indicating whether the child is born m to n months before or after an election. Thefigures are tabulated for all elections and scheduled elections. Absolute margin of victory isdefined as the proportion of seats by which the ruling party/coalition in the state won/lost inthe district of birth of the child during the previous election. Sample includes children born inthe period 1975-1998 from 14 major states in India.Table 2.1c: Summary Statistics (Mother’s Characteristics)Variable Mean (Std. Dev.) NScheduled Caste .173 ( .379) 74580Scheduled Tribe 0.094 (0.292) 74580Muslim 0.133 (0.339) 74580Urban 0.22 (0.414) 74580Mother’s Years of Education 2.932 (4.322) 74580Father’s Years of Education 5.79 (5.04) 74580Number of Antenatal Visits 1.973 (2.862) 56140Tetanus during pregnancy 0.651 (.477) 62004Notes: Standard deviations are in parentheses. Scheduled Caste,Scheduled Tribe and Muslim are dummies indicating whether themother belonged to historically disadvantaged caste (ScheduledCaste) or tribal groups (Scheduled Tribe) or the second largest re-ligious community in India (Muslim). Tetanus during pregnancy isa dummy equal to 1 if the mother had at least one tetanus injectionduring pregnancy. The sample consists of mothers of children bornin the period 1987-1992 (NFHS I) and 1995-1999 (NFHS II). Thissample is used to estimate Tables 2.8-2.10.34Figure 2.2: Election and Infant Mortality77.588.599.510Infant Mortality>24 mnths bef12−24 mnths bef0−12 mnths bef1−12 months aft12−24 mnths afterMonths before/after electionAll Elections(a)7.588.599.510Infant Mortality>24 mnths bef12−24 mnths bef0−12 mnths bef1−12 months aft12−24 mnths afterMonths before/after scheduled electionScheduled Elections(b)Notes: The panels in the figure shows the predicted relationship between infant mortalityand years relative to scheduled election and all election years. While Panel (a) graphs therelationship for all elections, panel (b) shows the relationship for scheduled elections.35Table 2.2: Elections and Infant Mortality(1) (2) (3) (4) (5) (6)All All Scheduled Scheduled Midterm MidtermBorn 0-12 Months before Election -0.560∗ -0.639(0.290) (0.388)Born 13-24 Months before Election -0.106(0.366)Born 1-12 Months after Election 0.0413(0.429)Born 13-24 Months after Election -0.318(0.389)Born 0-12 Months before Scheduled Election -1.194∗∗∗ -1.137∗∗∗(0.351) (0.398)Born 12-24 Months before Scheduled Election 0.0866(0.390)Born 1-12 Months after Scheduled Election 0.371(0.436)Born 12-24 Months after Scheduled Election -0.203(0.469)Born 0-12 Months before Midterm Election 0.267 0.295(0.433) (0.452)Born 12-24 Months before Midterm Election 0.0667(0.471)Born 1-12 Months after Midterm Election 0.0823(0.441)Born 12-24 Months after Midterm Election -0.222(0.438)Observations 175804 175804 175804 175804 175804 175804r2 0.341 0.341 0.341 0.341 0.341 0.341Notes: Standard errors in parentheses. Each column represents a separate regression. Infant mortality is the dependant variablein all the columns. Columns 1 and 2 report coefficients on dummies for all election, columns 3 and 4 report coefficients ondummies for scheduled elections and columns 5 and 6 report coefficients on dummies for midterm elections. In addition to thereported variables, all regressions include mother’s fixed effect, dummies for year of birth, month of birth, order of birth, sex ofthe child. Other controls include average district turnout, real state domestic product per capita lagged by two years and monthof polling dummies. Sample includes children born in the period 1975-1998 from 14 major states in India. Errors are clusteredat district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0136Table 2.3: Accounting for Midterm Elections(1) (2) (3) (4) (5) (6)Baseline Dropping midterm Including midterm controlsBorn 0-12 Months before Scheduled Election -1.194∗∗∗ -1.137∗∗∗ -1.258∗∗∗ -1.176∗∗∗ -1.179∗∗∗ -1.099∗∗∗(0.351) (0.398) (0.364) (0.427) (0.353) (0.411)Born 12-24 Months before Scheduled Election 0.0866 -0.121 0.120(0.390) (0.451) (0.407)Born 1-12 Months after Scheduled Election 0.371 0.517 0.410(0.436) (0.473) (0.444)Born 12-24 Months after Scheduled Election -0.203 0.00429 -0.177(0.469) (0.524) (0.474)Born 0-12 Months before Midterm Election 0.111 0.171(0.436) (0.454)Observations 175804 175804 157442 157442 175804 175804r2 0.341 0.341 0.370 0.370 0.341 0.341Notes: Standard errors in parentheses. Each column represents a separate regression. Infant mortality is the dependant variable inall columns. Columns 1 and 2 report coefficients on dummies for children born before or after scheduled elections. The sample hereincludes children born before midterm election (same as columns 3 and 4 of Table 2.2). Columns 3 and 4 also report coefficients ondummies for children born before and after scheduled elections but the sample now does not include children born 0-12 months beforemidterm election. Columns 5 and 6 include a dummy for whether a child is born 0-12 months before midterm election in addition tothe scheduled election dummies. Apart from the reported variables, all regressions include mother’s fixed effect, dummies for year ofbirth, month of birth, order of birth, sex of the child. Other controls include average district turnout, real state domestic product percapita lagged by two years and month of polling dummies. Errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0137Table 2.4: Neonatal and Infant Mortality(1) (2) (3) (4) (5) (6)Infant Mortality Neonatal Mortality 2-12 Month MortalityBorn 0-12 Months before Scheduled Election -1.194∗∗∗ -1.137∗∗∗ -0.746∗∗ -0.730∗∗ -0.445∗∗ -0.407∗(0.351) (0.398) (0.289) (0.322) (0.210) (0.237)Born 12-24 Months before Scheduled Election 0.0866 0.00852 0.0781(0.390) (0.340) (0.246)Born 1-12 Months after Scheduled Election 0.371 0.231 0.140(0.436) (0.352) (0.283)Born 12-24 Months after Scheduled Election -0.203 -0.153 -0.0498(0.469) (0.350) (0.269)Observations 175804 175804 175804 175804 175804 175804r2 0.341 0.341 0.341 0.341 0.295 0.295Notes: Standard errors in parentheses. Each column represents a separate regression. Infant mortality is the dependant variablein column 1 and 2, neonatal mortality in column 3 and 4 and 2-12 month mortality in column 5 and 6. The coefficientsreported correspond to dummies for children born before and after scheduled elections. In addition to the reported variables,all regressions include mother’s fixed effect, dummies for year of birth, month of birth, order of birth, sex of the child. Othercontrols include average district turnout, real state domestic product per capita lagged by two years and the dummies for monthof polling. Sample includes children born in the period 1975-1998 from 14 major states in India. Errors are clustered at districtlevel.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0138Table 2.5: Closeness to Election and Infant Mortality(1) (2) (3) (4)0-6 months 0-6,6-12 months 0-12 months Distance Nextbefore before before Sch ElecBorn 0-6 Months before Scheduled Election -1.035∗∗ -1.227∗∗∗(0.419) (0.428)Born 6-12 Months before Scheduled Election -1.145∗∗(0.484)Born 0-12 Months before Scheduled Election -1.194∗∗∗(0.351)Dist of the next Scheduled Election from the MOB 0.010∗∗(0.005)Observations 175804 175804 175804 175804r2 0.341 0.341 0.341 0.341Notes: Standard errors in parentheses. Each column represents a separate regression. Infant mortality is the dependantvariable. Column 1 reports the coefficient on the dummy indicating whether a child is born 0-6 months before scheduledelection. Column 2 reports the coefficients on dummies indicating whether a child is born 0-6 months or 6-12 months beforescheduled election. Column 3 reports coefficients on a dummy indicating whether a child is born 0-12 months before election.The coefficient reported in columns 4 corresponds to the distance between the month-year of birth of the child and the monthyear of next scheduled election. In addition to the reported coefficients, all regressions include mother’s fixed effect, dummiesfor year of birth, month of birth, order of birth, sex of the child. Other controls include average district turnout, real statedomestic product per capita and the month of polling dummies. Sample includes children born in the period 1975-1998 from14 major states in India. Errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0139Table 2.6: Slightly Before and After Scheduled Election(1) (2) (3) (4) (5) (6)0-6 0-2 0-1 0-1 0-2 0-6months months months months months monthsbefore before before after after afterBorn 0-6 Months before Scheduled Election -1.035∗∗(0.419)Born 0-2 Months before Scheduled Election -1.305∗∗(0.571)Born 0-1 Months before Scheduled Election -1.688∗∗(0.659)Born 0-1 Months after Scheduled Election -1.663∗∗(0.663)Born 0-2 Months after Scheduled Election -1.186∗∗(0.594)Born 0-6 Months after Scheduled Election 0.0294(0.451)Observations 175804 175804 175804 175804 175804 175804r2 0.341 0.341 0.341 0.341 0.341 0.341Notes: Standard errors in parentheses. Each column represents a separate regression. Infant mortality is the dependantvariable. Columns 1 and 6 report coefficients on dummies indicating whether a child is born 0-6 months before and afterscheduled election respectively. Columns 2 and 5 report coefficients on dummies for children born 0-2 months beforeand after scheduled election respectively. Columns 3 and 4 report coefficient on dummies indicating whether a child isborn 0-1 months before and after scheduled election respectively. In addition to the reported variables, all regressionsinclude mother’s fixed effect, dummies for year of birth, month of birth, order of birth, sex of the child. Other controlsinclude average district turnout, real state domestic product per capita lagged by two years and month of polling dummies.Column 1 includes coefficient of dummy indicating whether the child is born 0-6 Sample includes children born in theperiod 1975-1998 from 14 major states in India. Errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0140Figure 2.3: Role of Political Competition678910Infant Mortality0 .2 .4 .6 .8 1Absolute MarginBorn 0−12 months before Others0−12 months before(a)678910Infant Mortality0 .2 .4 .6 .8 1Absolute MarginBorn 0−6 months before Others0−6 months before(b)Notes: The panels in the figure shows the predicted relationship between infant mortalityand absolute margin of victory for children born before scheduled election and other years.While panel (a) graphs the relationship for children born 0-12 months before scheduledelection, panel (b) shows the relationship for children born 0-6 months before scheduledelection.41Table 2.7: Role of Political Competition(1) (2) (3) (4)0-12 months before 0-6 months beforeWith WithBaseline Interactions Baseline InteractionsBorn 0-12 Months before Scheduled Election -1.194∗∗∗ -2.180∗∗∗(0.351) (0.598)Absolute Margin x Born 0-12 Months before Scheduled Election 2.048∗∗(1.002)Born 0-6 Months before Scheduled Election -1.035∗∗ -2.526∗∗∗(0.419) (0.824)Absolute Margin x Born 0-6 Months before Scheduled Election 2.989∗∗(1.380)Absolute Margin 0.0146 0.0279(0.482) (0.477)Observations 175804 175804 175804 175804r2 0.341 0.341 0.341 0.341Notes: Standard errors in parentheses. Each column represents a separate regression. Infant mortality is the dependantvariable. Columns 1 and 3 report the coefficients on dummies for children born 0-6 months and 0-12 months before scheduledelections. Columns 2 and 4 report the coefficients on dummy variables indicating whether a child is born 0-6 or 0-12 monthsbefore scheduled elections and interactions between the election dummy and the absolute margin of victory of the rulingparty/coalition in the previous election. Columns 2 and 4 also report the coefficient on the absolute margin of victory variable.In addition to the reported variables, all regressions include mother’s fixed effect, dummies for year of birth, month of birth,order of birth, sex of the child. Other controls include average district turnout, state domestic product lagged by two yearsand dummies for the month of polling. Sample includes children born in the period 1975-1998 from 14 major states in India.Errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0142Table 2.8: Antenatal Visits(1) (2) (3) (4)0-6 months before 0-12 months beforeBorn 0-6 Months before Scheduled Election 0.126∗∗∗ 0.285∗∗∗(0.0468) (0.0864)Absolute Margin x Born 0-6 Months before Scheduled Election -0.317∗∗(0.152)Born 0-12 Months before Scheduled Election 0.0693∗ 0.166∗∗(0.0418) (0.0766)Absolute Margin x Born 0-12 Months before Scheduled Election -0.199(0.140)Absolute Margin -0.0555 -0.0536(0.0652) (0.0669)Observations 56140 56140 56140 56140r2 0.493 0.493 0.493 0.493Notes: Standard errors in parentheses. Each column represents a separate regression. The dependant variable isthe number of antenatal visits. Columns 1 and 3 report the coefficients on dummies for children born 0-6 monthsand 0-12 months before scheduled elections. Columns 2 and 4 report the coefficients on dummy variables indicatingwhether a child is born 0-6 or 0-12 months before scheduled elections and interactions between the election dummyand the absolute margin of victory of the ruling party/coalition in the previous election. Columns 2 and 4 also reportthe coefficient on the absolute margin of victory variable. In addition to the reported variables, all regressions includedistrict fixed effect, dummies for year of birth, month of birth, order of birth, sex of the child, mother’s and father’s yearof schooling and dummies for rural residence, membership in SC/ST. Other controls include average district turnout,real state domestic product per capita lagged by two years and month of polling dummies. Sample includes childrenborn in the period 1987-1992 and 1995-1998 from 14 major states in India. Errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0143Table 2.9: Tetanus During Pregnancy(1) (2) (3) (4)0-6 months before 0-12 months beforeBorn 0-6 Months before Scheduled Election 0.0184∗∗ 0.0511∗∗∗(0.00844) (0.0150)Absolute Margin x Born 0-6 Months before Scheduled Election -0.0704∗∗(0.0296)Born 0-12 Months before Scheduled Election 0.00504 0.0275∗∗(0.00743) (0.0115)Absolute Margin x Born 0-12 Months before Scheduled Election -0.0494∗∗(0.0200)Absolute Margin 0.00327 0.00450(0.0146) (0.0149)Observations 62004 62004 62004 62004r2 0.262 0.262 0.262 0.250Notes: Standard errors in parentheses. Each column represents a separate regression. The dependant variable is a dummyindicating whether the mother had at least one tetanus injection during pregnancy. Columns 1 and 3 report the coefficients ondummies for children born 0-6 months and 0-12 months before scheduled elections. Columns 2 and 4 report the coefficientson dummy variables indicating whether a child is born 0-6 or 0-12 months before scheduled elections and interactionsbetween the election dummy and the absolute margin of victory of the ruling party/coalition in the previous election.Columns 2 and 4 also report the coefficient on the absolute margin of victory variable. In addition to the reported variables,all regressions include district fixed effect, dummies for year of birth, month of birth, order of birth, sex of the child,mother’s and father’s year of schooling and dummies for rural residence, membership in SC/ST. Other controls includeaverage district turnout, real state domestic product per capita lagged by two years and month of polling dummies. Sampleincludes children born in the period 1987-1992 and 1995-1998 from 14 major states in India. Errors are clustered at districtlevel.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0144Table 2.10: Low Birth Weight(1) (2) (3) (4)0-6 months before 0-12 months beforeBorn 0-6 Months before Scheduled Election -0.0214∗∗ -0.0312∗(0.00848) (0.0186)Absolute Margin x Born 0-6 Months before Scheduled Election 0.0195(0.0313)Born 0-12 Months before Scheduled Election -0.00363 -0.00797(0.00783) (0.0152)Absolute Margin x Born 0-12 Months before Scheduled Election 0.00891(0.0251)Absolute Margin -0.00423 -0.00370(0.0115) (0.0118)Observations 52817 52817 52817 52817r2 0.0428 0.0429 0.0427 0.0427Notes: Standard errors in parentheses. Each column represents a separate regression. The dependant variable is a dummyindicating whether a child is of low birth weight is the dependant variable. Columns 1 and 3 report the coefficientson dummies for children born 0-6 months and 0-12 months before scheduled elections. Columns 2 and 4 report thecoefficients on dummy variables indicating whether a child is born 0-6 or 0-12 months before scheduled elections andinteractions between the election dummy and the absolute margin of victory of the ruling party/coalition in the previouselection. Columns 2 and 4 also report the coefficient on the absolute margin of victory variable. In addition to the reportedcoefficients, all regressions include district fixed effect, dummies for year of birth, month of birth, order of birth, sex ofthe child, mother’s and father’s year of schooling and dummies for rural residence, membership in SC/ST. Other controlsinclude average district turnout, real state domestic product per capita lagged by two years and month of polling dummies.Sample includes children born in the period 1987-1992 and 1995-1998 from 14 major states in India. Errors are clusteredat district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0145Table 2.11: Robustness check: Placebo Tests(1) (2) (3) (4) (5) (6)12 months 24 months 36 monthsBorn 0-12 Months before Election -0.450 0.0558 0.427 -0.543 -0.129 0.0317(0.468) (0.687) (0.483) (0.752) (0.476) (0.714)Absolute Margin x Born 0-12 Months before Elecion -1.079 2.113 -0.330(1.130) (1.354) (1.213)Absolute Margin 0.812 0.291 0.678(0.633) (0.656) (0.640)Observations 136200 136200 136200 136200 136200 136200r2 0.397 0.397 0.397 0.397 0.397 0.397Notes: Standard errors in parentheses. Each column represents a separate regression. Infant mortality is the dependantvariable. Columns 1, 3 and 5 report the coefficients on dummies for children born 0-12 months before scheduled electionsif the scheduled election was held 12 months, 24 months and 36 months before the actual scheduled election respectively.Columns 2, 4 and 6 report the coefficients on dummy variables indicating whether a child is born 0-12 months before sched-uled elections and interactions between the election dummy and the absolute margin of victory of the ruling party/coalition inthe previous election. Columns 2, 4 and 6 also report the coefficient on the absolute margin of victory variable. In addition tothe reported variables, all regressions include mother’s fixed effect, dummies for year of birth, month of birth, order of birth,sex of the child. Other controls include average district turnout, real state domestic product per capita lagged by two yearsand dummies for the month of polling. Sample includes children born in the period 1975-1998 from 14 major states in India,excluding children born 0-12 months before actual scheduled election. Errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0146Table 2.12: Probability of Birth of Female Child(1) (2) (3) (4)0-6 months before 0-12 months beforeBorn 0-6 Months before Scheduled Election -0.00648 -0.0211(0.00786) (0.0134)Absolute Margin x Born 0-6 Months before Scheduled Election 0.0298(0.0207)Born 0-12 Months before Scheduled Election -0.00442 -0.00674(0.00665) (0.0104)Absolute Margin x Born 0-12 Months before Scheduled Election 0.00507(0.0169)Absolute Margin -0.0157∗ -0.0133(0.00866) (0.00882)Observations 175804 175804 175804 175804r2 0.314 0.314 0.314 0.314Notes: Standard errors in parentheses. Each column represents a separate regression. Dependant variable is a dummy equalto 1 if the child is female. Columns 1 and 3 report the coefficients on dummies for children born 0-6 months and 0-12months before scheduled elections. Columns 2 and 4 report the coefficients on dummy variables indicating whether a childis born 0-6 or 0-12 months before scheduled elections and interactions between the election dummy and the absolute marginof victory of the ruling party/coalition in the previous election. Columns 2 and 4 also report the coefficient on the absolutemargin of victory variable. In addition to the reported variables, all regressions include mother’s fixed effect, dummiesfor year of birth, month of birth, order of birth, sex of the child. Other controls include average district turnout, real statedomestic product per capita lagged by two years and dummies for the month of polling. Sample includes children born inthe period 1975-1998 from 14 major states in India. Errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0147Table 2.13: Excluding Kerala(1) (2) (3) (4)Including Kerala Excluding Kerala0-12 months 0-6 months 0-12 months 0-6 monthsbefore before before beforeBorn 0-12 Months before Scheduled Election -1.194∗∗∗ -1.284∗∗∗(0.351) (0.365)Born 0-6 Months before Scheduled Election -1.035∗∗ -1.081∗∗(0.419) (0.430)Observations 175804 175804 170391 170391r2 0.341 0.341 0.340 0.340Notes: Standard errors in parentheses. Each column represents a separate regression. Infant mortality is the de-pendant variable. Columns 1 and 2 report results including Kerala, the state where infant mortality is significantlylower than the average infant mortality over the sample. Columns 2 and 4 report results excluding Kerala fromthe sample. Columns 1 and 3 report coefficients on dummies indicating whether the child is born 0-12 monthsbefore scheduled election. Columns 2 and 4 report coefficients on dummies for children born 0-6 months beforescheduled election. Apart from the reported variables all regressions include mother’s fixed effect, dummies foryear of birth, month of birth, order of birth, sex of the child. Other controls include average district turnout, realstate domestic product per capita and month of polling dummies. Errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0148Table 2.14: Election and Infant Mortality (All States)(1) (2) (3) (4)Politically Competitive States All StatesBorn 0-12 Months before Scheduled Election -1.194∗∗∗ -1.137∗∗∗ -0.675∗ -0.554(0.351) (0.398) (0.367) (0.414)Born 12-24 Months before Scheduled Election 0.0866 0.250(0.390) (0.380)Born 1-12 Months after Scheduled Election 0.371 0.589(0.436) (0.460)Born 12-24 Months after Scheduled Election -0.203 -0.133(0.469) (0.440)Observations 175804 175804 185976 185976r2 0.341 0.341 0.343 0.343Notes: Standard errors in parentheses. Each column represents a separate regression. Infant mortality isthe dependant variable. The sample used in regressions corresponding to columns 1 and 2 include datafrom 14 major politically competitive states in India (details in section 6.4). Columns 3 and 4 include all15 major states in India. The coefficients reported correspond to dummies for children born before andafter scheduled elections. In addition to the reported coefficients, all regressions include mother’s fixedeffect, dummies for year of birth, month of birth, order of birth, sex of the child. Other controls includeaverage district turnout, real state domestic product per capita lagged by two years and dummies for monthof polling. Errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0149Table 2.15: Role of Political Competition (All States)(1) (2) (3) (4)0-12 months before 0-6 months beforeWith WithBaseline Interactions Baseline InteractionsBorn 0-12 Months before Scheduled Election -0.675∗ -1.532∗∗(0.367) (0.613)Absolute Margin x Born 0-12 Months before Scheduled Election 1.761∗(0.967)Born 0-6 Months before Scheduled Election -0.655 -1.958∗∗(0.432) (0.803)Absolute Margin x Born 0-6 Months before Scheduled Election 2.594∗(1.326)Absolute Margin -0.100 -0.0883(0.479) (0.475)Observations 185976 185976 185976 185976r2 0.343 0.343 0.343 0.343Notes: Standard errors in parentheses. Each column represents a separate regression. Infant mortality is the dependantvariable. Columns 1 and 3 report coefficients on dummies for children born 0-6 months and 0-12 months before scheduledelections. Columns 2 and 4 report the coefficients on dummy variables indicating whether a child is born 0-6 or 0-12months before scheduled elections and interactions between the election dummy and the absolute margin of victory of theruling party/coalition in the previous election. Columns 2 and 4 also report the coefficient on the absolute margin of victoryvariable. In addition to the reported variables, all regressions include mother’s fixed effect, dummies for year of birth, monthof birth, order of birth, sex of the child. Other controls include average district turnout, real state domestic product per capitalagged by two years and dummies for the month of polling. Sample includes children born in the period 1975-1998 from 15major states in India. Errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0150Chapter 3Fertility Rate and FemaleEmployment in India3.1 IntroductionRapid population growth is viewed as one of the major obstacles to developmentin several developing countries. Incentives for having small families often do notexist in developing countries (Po¨rtner (2001); Caldwell and Caldwell (1987)), andas a result their population increases at a very fast rate. Population growth reducesthe per capita income and lowers the pace of poverty reduction (Eswaran (2006)).Moreover, high population growth has adverse effects on the environment throughdeforestation and air and water pollution (Eswaran (2006); Cropper and Griffiths(1994)).This chapter explores whether providing economic incentives to women af-fects the number of children they choose to have, and whether there are differ-ences across sectors, in the Indian context. In particular, this study estimates theeffect of female labor force participation on the total fertility rates. In addition, I51compare the impact of female employment in the agricultural and manufacturingsectors since women working in different sectors could have different demandsfor children (Goh (1999)).It is generally held that there exists a negative relationship between fertilityand female employment due to the incompatibility between child care and laborforce participation (Cramer (1980), Felmlee (1993), Smith-Lovin and Tickamyer(1981)). Since child-rearing is intensive in the mother’s time, female labor-forceparticipation may decrease the number of desired children. Alternatively withhigher income a woman might be able to afford better childcare services and thiscan increase the demand for children. Thus, the question of how female employ-ment impacts fertility is an empirical one.When exploring the relationship between fertility and female labor force par-ticipation, an endogeneity problem must be accounted for between these two si-multaneous decisions. OLS is inconsistent because the employment of women ispotentially correlated with unobservable factors affecting fertility decisions likethe preference for children and child care costs.The standard econometric approach to the endogeneity of regressors is instru-mental variables. Thus if we want to estimate the impact of employment opportu-nities on fertility, we need to find some exogenous determinant of female employ-ment which is uncorrelated with the unobservable factors affecting fertility.Such an exogenous change took place in India during 1991. In response to asevere balance of payments crisis in 1991, India turned to the International Mon-etary Fund (IMF) for assistance. As a condition of IMF support, India had to52drastically reduce tariffs in different sectors over a short period of time. The dis-persion in tariffs across sectors was also substantially reduced. This chapter usesvariation in the exposure to tariff levels across districts in India as an instrumentfor analyzing the impact of female labor force participation on fertility. I alsoseparately examine the influence of female employment in different sectors onfertility. Agriculture, manufacturing and mining are the three broad traded goodsectors. However, the proportion of women employed in mining is extremelysmall. Thus I have confined my analysis to the first two sectors. Figure 3.1 com-pares the average tariff in agriculture and manufacturing in 1987 and 1998. Thefigure shows that the tariff decline has been quite substantial in both of these sec-tors.Reduction in tariffs is likely to lower the prices of tradable goods in the domes-tic market. Thus the demand for workers employed in these sectors will fall. So,tariff decline is likely to reduce the demand for female labour who are employedin sectors in which tariff decline took place.The data used for estimating the total fertility rate is obtained from the Censusof India. The estimates for fertility are at the district level. However, the tariff datais at the national level. In order to get a district level estimate, I have calculated adistrict level tariff measure. I have also computed district level tariff measures sep-arately for the agricultural and manufacturing sectors. This is done by interactingthe share of the workers in a district who are employed in the various industriesin the pre-reform period and then taking the sum across all the subsectors underagriculture and manufacturing separately. This derivation is based on Topalova53(2007). Thus my instrument exploits the variation in the degree of trade liberal-ization across the various industries and the variation in districts to the degree towhich each is affected by the tariff reduction.The result obtained in this chapter is that female labor force participation onthe whole, and female employment in agriculture, has no statistically significanteffect on fertility. However, increasing female employment in manufacturing sig-nificantly reduces the fertility rate.The remainder of this chapter is organized as follows. Section 3.2 gives a briefliterature review. In section 3.3, I have given a description of the tariff reformthat took place in India to see whether it was indeed exogenous and externallyimposed. Section 3.4 describes the data used in this study. Section 3.5 outlinesthe empirical strategy and the results obtained in this chapter. Finally, Section 3.6concludes the chapter.3.2 Related LiteratureIt has been argued that there exists a negative relationship between fertility and fe-male labor force-participation and wages (Becker (1960), Galor and Weil (1996)).Crafts (1989) used data from the 1911 Census of England and Wales and foundthat fertility is lower in districts which has better female employment opportuni-ties. Using an instrumental variables approach, Schultz (1985) tried to find thecausal impact of increasing male wages and female wages relative to male wageson fertility for Sweden. He used output prices as instruments and found that an in-crease in the female-to-male wage results in a decline in observed fertility. Other54studies have found find an insignificant influence of womans employment on fe-male fertility choices (Santow and Bracher (2001)). The evidence of a positiveeffect of womens employment on fertility is found for East Germany (Kreyenfeld(2004))There are many papers which have estimated the impact of female labor forceparticipation on fertility rates using cross-country data for the OECD countries.Most of these studies found that the female labor force participation rate changedits sign from a negative value before 1980, to a positive value thereafter for theOECD countries (Ahn and Mira (2002); Rindfuss, Guzzo, and Morgan (2003)).There has been limited evidence on the impact of female employment on fertil-ity for developing countries. Fang et al. (2010) studied how off-farm employmentaffects actual and desired fertility in rural China. They used two instrumental vari-ables. The first one is whether there is a bus stop in the village and the second oneis the proportion of the labor force in the village that is employed in enterpriseshaving at least 20 employees. They found that female employment significantlyreduces both actual and desired fertility levels in rural China. However, they con-sidered only rural China and off-farm employment. As far as the composition offemale employment is concerned, Goh (1999) analysed, with the help of a theoret-ical model, how women working in the manufacturing sector desire fewer childrenthan women working in agricultural sector, since the former group faces a highertime cost.In Indian context, Dre`ze and Murthi (2001) examined the role of female lit-eracy as a determinant of fertility. More recently Jensen (2012) provided experi-55mental evidence on the relationship between women’s employment opportunitiesand fertility outcomes. The author conducted an experiment where women in ran-domly selected rural villages in four Indian states (Haryana, Punjab, Rajasthan,and Uttar Pradesh) were given recruiting services for a period of three years. Theauthor found that women were significantly more likely to work in treated villagesand there was substantial delay in marriage and child bearing. Another recent pa-per by Anukriti and Kumler (2015) has analysed the impact of trade reforms onfertility in India.3.3 Trade Reform in IndiaUntil the mid 1980s, India pursued inward-looking development strategies. In par-ticular, India had high nominal tariffs and nontariff barriers, and a complex importlicensing system. During the mid 1980’s, India started to gradually implementsome reforms. Import and industrial licensing were eased. Also tariffs replacedsome quantitative restrictions (Topalova (2007)). However a rise in macroeco-nomic imbalances, accompanied with this gradual liberalization policy adopted inIndia, lead to balance of payments problems. As a result of the Gulf War, India'soil bill increased. There was also a reduction in demand from some of its tradingpartners and a drop in remittances from the Middle East. This lead to a deterio-ration of investor confidence and investors started to withdraw money. Politicalinstability during this period exacerbated the situation. The Congress Party led byRajiv Gandhi lost the 1989 election and a coalition government came into power.However two subsequent coalitions could not survive long and a fresh election56was announced in 1991. Political stability was further disrupted by the assassi-nation of the then chairman of the Congress Party, Rajiv Gandhi in 1991. Theseexogenous events led to large capital outflow.In order to overcome the crisis, the Indian government secured an emergencyloan from the IMF. However, the IMF’s support was conditional on macroeco-nomic stabilization policies and structural reforms. The structural reforms focusedon the dismantling of industrial and import licenses, reforms in the financial sec-tor and loosening trade policy. Thus the Indian government was forced to sharplyreduce import and export control systems. There were drastic tariff reductions:average tariffs fell from over 90% in 1987 to about 30% in 1997. The standarddeviation of tariffs dropped by about 50 percent over the same period. Industrieswhich initially had higher tariffs faced higher tariff reductions and the structure oftariffs across industries changed.From the perspective of households, such changes were exogenous in nature.Households were unlikely to have anticipated the reforms in the late 1980s. Thegovernment had to meet strict compliance deadlines and implemented the tar-iff reforms abruptly. Thus India drastically reduced tariffs over a short periodof time. Topalova and Khandelwal (2011) documents that tariff changes werenot correlated with pre-reform industry characteristics, including productivity andskill intensity, at least until 1997.573.4 Data SourcesThe data used in my analysis come from a number of sources. Firstly, the data forestimating the total fertility rate come from two Indian censuses held in 1991 and2001. In this chapter, I have measured fertility in a district by the total fertility rate(TFR), which represents the number of children that would be born to a womanif she lived to the end of her childbearing years and bore children at each age inaccordance with the prevailing age-specific fertility rates. The total fertility rate isa more useful measure of the fertility level than other measures (for example crudebirth rate), since it is independent of the age structure of the population. The totalfertility rate for the year 1991 are obtained from the census report on district levelestimates of fertility published in 1997. In this report, the total fertility rate hasbeen computed using the Arrianga method (described in detail in Appendix B).I have computed the total fertility rate for the year 2001 by this method usingthe 2001 census data on the number of children ever born to women by 5-yearage group. The construction of this variable is explained in Appedix B. Table 3.1contains the summary statistics of the variables used in this chapter. The averageTFR reduced from 4.4 in 1991 to 3.9 in 2001.The second data set used in this study consists of tariff data. The tariff datawas initially collected by Petia Topolova from various publications of the Ministryof Finance. This data was at the 6 digit level of the Indian Trade ClassificationHarmonized System. Arka Roy Chaudhuri matched the product lines of the HScode to the 3 digit National Industrial Classification (NIC) code using the concor-58dance of Debroy and Santhanam (2001) for the paper Roy Chaudhuri (2012). Inthis chapter I have used the data-set compiled by him. I have used the tariff datafor the years 1987 and 1998 in my analysis.Using this dataset, I have constructed a district level tariff variable. The vari-able is also constructed separately for the agricultural and manufacturing sectors.This district level tariff variable is a weighted aggregate of the tariffs faced byeach sub sector of the particular sectors in a district. The weights are equal to theproportion of workers in each district who were employed in that industry beforethe reforms took place. The information about the proportion of workers em-ployed in each industry is obtained from the 43rd (1987-88) round of the IndianNational Sample Survey (NSS). Table 3.1 shows that the agricultural tariff usedhas reduced from 27% to 8%. It is to be noted here that the sector specific tariffsare not scaled by the share of industry employment in that particular sector butby the share of industry employment in the district labor force. Thus the sum ofthe weights attached to the tariff is less than 1. The manufacturing tariff similarlyreduced from 6.5% to 2%.The district level information on the female labor force participation as well asthe proportion of females employed in agriculture and manufacturing comes fromthe 43rd (1987-88) and 55th (1999-2000) rounds of the NSS. We can see fromTable 3.1 that the proportion of females employed in agriculture by district and inage group 16-25, increased on average from 28.5% to 30%. However on average,the proportion of females employed in manufacturing reduced slightly from 3.4%to 3%. Overall female labor force participation also slightly reduced from 40% to5939% during the same period.I have included controls such as the proportion of people belonging to theScheduled Caste or Scheduled Tribe1, female literacy, proportion of populationthat is urban in the district and proportion of Muslims2 in the district. The data onthese variables are obtained from the 1991 and 2001 censuses.3.5 Empirical Strategy and Results3.5.1 Effect of Female Labor Force Participation on FertilityThe trade liberalization in India was externally imposed and comprehensive. Thusthe variations in tariff levels are unlikely to be correlated with the unobservablefactors that determine the fertility decisions of women. On the other hand, sincetariffs affect the productivity of industries; it is likely to influence the labor forceparticipation of women.In order to find the causal impact of female labor force participation on thefertility rate of a given district, I have used the district level tariffs as instrument.The first stage is given by:Pdt = α+βXdt + γTdt +δd + τt + εdt (3.1)where Pdt denotes the labor force participation of women in district d at timet as a percentage of total women in the age group 16-65 in the district d at time t.1. The constitution of India recognizes some traditionally disadvantaged castes (ScheduledCastes) and certain culturally distinct tribes (Scheduled Tribes) as requiring some additional con-sideration.2. Muslims form the largest religious minority in India60Xdt denotes other controls like the proportion of Scheduled Castes and ScheduledTribes, proportion of urban population, proportion of Muslims, female literacyrate and poverty rate. δd and τt denote the district fixed effects and a time dummyrespectively. The district fixed effects capture differences across districts that arefixed over time while the time dummies capture unobserved countrywide changesover the time period. Tdt is the instrument used in this study and is given byTdt =∑i∑ jWFjid1987×Tjit∑i∑ jWFjid1987where WFjid1987 denotes the total workforce employed in industry j of sectori in district d in 1987 (pre-reform period) and Tjit is the tariff level in industry j ofsector i at time t.The following equation is estimated in the second stage:TFRdt = λ +θXidt +ΨPdt +∆d +Tt +ξdt (3.2)where TFRdt is the total fertility rate in district d at time t. Pdt denotes theproportion of females in the age group 16-65 in the district d who participates inthe labor force. The relevant coefficients of interest is Ψ.Table 3.2 shows the impact of female labor force participation on the totalfertility rate. Column 1 shows the OLS results. We can see that female laborforce participation has no significant effect on the total fertility rate. Column2 shows the first stage and shows that the instrument is statistically significant.Columns 3 and 4 present the reduced form and the IV results. It can be seen that61the instrument has no significant effect on total fertility rate. The IV results givenin Column 4 show that female labor force participation has no effect on the totalfertility rate once the endogeneity in female labor force participation is accountedfor.3.5.2 Effect of Female Employment in Agriculture andManufacturing on FertilityIn order to find the causal impact of female employment in the agriculture andmanufacturing sectors on the fertility rate of a district, I have used the districtlevel tariffs for agriculture and manufacturing as instruments. The first stage forfemale labor force participation is given by:ρidt = α+βXidt + γaTadt + γmTmdt +δd + τt + εidt (3.3)where ρidt denotes the proportion of women who are employed in sector i(i = agriculture (a) or manufacturing (m)) of district d in state s at time t as aproportion of the total women in the age group 15-65 in the district d at time t.The instruments Tadt and Tmdt are given byTidt =∑ jWFjid1987×Tjit∑i∑ jWFjid1987In the second stage, I have estimated the following equation:TFRdt = λ +θXidt +ψaρadt +ψmρmdt +∆d +Tt +ξdt (3.4)62The relevant coefficients of interest are ψa and ψm.Table 3.3 shows the results obtained by regressing the proportion of femalesemployed in agriculture and manufacturing in a district on the district total fertil-ity rate. Column 1 presents the OLS results, which do not show any significantrelationship between sector wise female employment and fertility. Columns 2 and3 show the first stage results for the proportion of females employed in agricultureand manufacturing. We can see that the endogenous variables are significantlycorrelated with the instruments. Columns 4 and 5 show the reduced form andthe IV results. The IV results show that increased female employment in manu-facturing leads to a lower fertility rate. However, increasing female employmentin agriculture has no significant impact on female labor force participation. Thismight be due to the fact that agricultural jobs can be more easily combined withchild rearing.3.6 ConclusionThis chapter has been devoted to an examination of the hypothesis that there ex-ists a causal link from female labor-force participation, and sector-wise femaleemployment in the agriculture and the manufacturing sectors to the total fertilityrate. The key finding is that female employment in manufacturing at the districtlevel reduces the district fertility rate once the endogeneity of female employmentis accounted for. However, overall female labor force participation and femaleemployment in agriculture do not have any significant effects on fertility.63Figure 3.1: Decline in the Average Tariff in Agriculture and Manufacturing020406080100Agriculture ManufacturingMean Tariff 1987 Mean Tariff 1998Notes: The figure shows the mean tariff in agriculture and manufacturing for the years 1987and 1998. The left set of bars correspond to agriculture while the right set of bars correspondto manufacturing64Table 3.1: Summary StatisticsVariable Mean Std. Dev. NYear 1991Total Fertility Rate 4.373 0.959 319Female labor force participation 0.398 0.194 319Proportion of female in agriculture 0.285 0.183 319Proportion of female in manufacturing 0.034 0.044 319Agricultural Tariff 27.201 9.208 319Manufacturing Tariff 6.426 4.919 319Proportion of Scheduled Castes 0.166 0.072 319Proportion of Scheduled Tribes 0.1 0.169 319Proportion of Muslims 0.107 0.108 319Proportion of urban population 0.223 0.165 319Proportion of females literate 0.302 0.162 319Poverty Rate 0.348 0.174 319Year 2001Total Fertility Rate 3.879 0.733 319Female labor force participation 0.39 0.186 319Proportion of female in agriculture 0.298 0.188 319Proportion of female in manufacturing 0.03 0.045 319Agricultural Tariff 7.977 3.134 319Manufacturing Tariff 2.062 1.979 319Proportion of Scheduled Castes 0.165 0.072 319Proportion of Scheduled Tribes 0.101 0.167 319Proportion of Muslims 0.112 0.112 319Proportion of urban population 0.239 0.174 319Proportion of females literate 0.521 0.154 319Poverty rate 0.215 0.123 319Notes: The table reports the mean and standard deviations of the variablesused in this analysis. The top panel corresponds to the year 1991 and thebottom panel corresponds to the year 2001.65Table 3.2: Female Labor Force Participation and Total Fertility Rate(1) (2) (3) (4)OLS First Stage Reduced Form IVFemale labor force participation -0.0227 -0.845(0.185) (0.661)District Tariff 0.00650∗∗∗ -0.00549(0.00232) (0.00586)Observations 638 638 638 638F-stat(First Stage) 8.02Notes: The table presents the impact of female labor force participation on total fertilityrate. Columns 1, 2, 3 and 4 report the OLS, first stage, reduced form and IV results. Thecontrols used in the regressions are proportion of Scheduled Castes and Scheduled Tribes,proportion of urban population, proportion of Muslims, female literacy rate and poverty rate.The errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0166Table 3.3: Effect of Female Employment in Agriculture and Manufacturing onTotal Fertility Rate(1) (2) (3) (4) (5)First Stage First Stage ReducedOLS Agriculture Manufacturing Form IVProportion of female in agriculture 0.127 -0.169(0.186) (0.749)Proportion of female in manufacturing -0.940 -4.313∗∗(0.754) (2.065)Agricultural Tariff 0.00803∗∗∗ 0.00133∗∗ -0.00710(0.00205) (0.000541) (0.00597)Manufacturing Tariff 0.0144∗∗∗ 0.00631∗∗∗ -0.0296∗∗(0.00433) (0.00152) (0.0124)Observations 638 638 638 638 638F-stat(First Stage) 9.63 15.83Notes: The table presents the impact of female employment in agriculture and manufacturing on total fertility rate.Column 1 reports the OLS results. Columns 2 and 3 reports the first stage results for agriculture and manufacturingrespectively. Columns 4 and 5 reports the reduced form and the IV results. The controls used in the regressionsare proportion of Scheduled Castes and scheduled tribes, proportion of urban population, proportion of Muslims,female literacy rate and poverty rate. The errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.0167Chapter 4The Evolution of Gender Gaps inIndia4.1 IntroductionThe gender gap in employment in India is very large. In 1983, barely 31 percentof Indian women in the working age group of 16-64 years worked in the laborforce. By 2005, this number had risen, but barely, to 40 percent. The correspond-ing numbers for men were around 94 percent. Amongst the Indian workforce thatis illiterate, around one-third were women, both in 1983 and in 2005. At the otherextreme, in 1983 barely 11 percent of workers with middle school or higher edu-cation were women. This number rose to 22 percent by 2005. On the employmentside, in 1983 only 10 percent of white collar jobs in India were performed bywomen. This rose by a bare 5 percentage points to 15 percent in 2005.To summarize, a large share of working age Indian women choose not to par-ticipate in the labor market. When they do, they find themselves very poorlytrained with most of them having very little education. Consequently, most women68workers end up working in low skill and low return agrarian jobs while the higherskill white collar jobs are typically performed by men. Starting with the basicpremise that there are no innate differences between the genders in ability, thesestatistics tell a rather disheartening overall story of the allocation of talent in thecountry. They suggest large scale under-utilization of productive resources alongwith misallocation of labor inputs across occupations that potentially have seriousproductivity consequences for the country.While the statistics cited above are disappointing, the period since 1983 hasalso seen sharp declines in the gender wage gap. The median male wage was90 percent above the median female wage in 1983. By 2010 this premium haddeclined to about 50 percent. To put these numbers in perspective, in the US themedian gender wage premium declined from 55 percent to 18 percent between1979 and 2011 (see Kolesnikova and Liu (2011)).1 In China on the other hand,the gender gap has been reported to be rising over the past two decades. Nationalsurveys in China report that the average male-to-female wage mark-up has risenfrom 28 to 49 percent in urban areas and from 27 to 79 percent in rural areasbetween 1990 and 2010. The Indian performance is thus quite encouraging whenexpressed in this relative context.In this chapter we examine the factors underlying the sharp decline in thegender wage gap. Did the gender wage gap fall across all income groups? Didit decline due to a decline in the gender gaps in the proximate determinants of1. The OECD average for the median wage premium of full-time male workers over their fe-male counterparts in 2009 was 23 percent. There is a lot of variation though with the male premiumvarying from 35 percent in Austria and the Czech Republic to just around 5 percent in Italy.69wages such as education attainment rates and occupation choices of the work-force? We examine this using household level survey data from successive roundsof the National Sample Survey (NSS) from 1983 to 2010. The period since 1983is a particularly interesting phase in India since it has been characterized by sharpmacroeconomic changes. Whether such sharp macroeconomic changes have alsocoincided with better harnessing and allocation of talent in the country is a ques-tion of independent interest.Our primary finding is that there has been broad-based and significant de-creases in gender gaps across a number of indicators. Both education attainmentrates and occupation choices of men and women have been broadly convergingsince 1983. Moreover, a large part of the decline in the gender wage gap is ac-counted for by convergence in these attributes of wages. We also find that thegender wage gap has declined across most of the income distribution. However,while for the 10th and 50th percentiles of the wage distribution, the decline inthe gender wage gap was accounted for by convergence in measured attributes(primarily education), the gender wage convergence in the 90th percentile of thewage distribution was mostly due to unmeasured factors. Strikingly, changes inthe measured attributes of this group tended to widen the gender wage gap. Thiseffect is particularly strong in urban India which could reflect reductions in genderdiscrimination in urban areas though this requires more detailed investigation.Our results on gender gaps suggest a general pattern of declining socioe-conomic gaps across a number of different groups in India over the past threedecades. In Hnatkovska, Lahiri, and Paul (2012) and Hnatkovska, Lahiri, and70Paul (2013), it has been shown that gaps between Scheduled Castes and Tribesand the rest have narrowed sharply since 1983 along a number of different indi-cators. Similarly, Hnatkovska and Lahiri (2012) found an even sharper narrowingof socioeconomic gaps between rural and urban workers between 1983 and 2010.Taken together, these results suggest that the period since 1983 which has beenmarked by rapid economic transformation and growth in India has also been aperiod that has seen disadvantaged groups sharply reducing their large historicalsocioeconomic disparities relative to others.We should note that inequality in society can be measured as within-groupinequality or between-group inequality. Our approach in this chapter focuses onbetween-group inequality. Our finding of declining inequality between groups inthese papers is not inconsistent with findings of widening within-group inequalityin India during some sub-periods since 1983. It is plausible that there is moreinequality within and less inequality across groups. More generally, the resultssuggest that more work is required to determine the overall pattern of inequalityin India during the last 30 years of market oriented reforms and growth take-off.This chapter is related to some existing literature on the gender difference inlabor market outcomes in India. Tilak (1980), used survey data from East Go-davari district of Andhra Pradesh analyzed the difference in return to educationacross gender in India. The paper provides evidence that gender wage gap is rel-atively less for higher education groups. Using survey data from the Lucknowdistrict of Uttar Pradesh, Kingdon (1998) found that women face significantlylower economic rates of returns to education than men. Kingdon and Unni (2001)71found that women face high level of wage discrimination in the Indian labor mar-ket using 1987-1988 NSS data on Tamil Nadu and Madhya Pradesh. However,education contributes little to this wage disadvantage of women.A key limitation of these studies is that they are concentrated in specific dis-tricts or states and do not produce national level estimates. Using national level“Employment and Unemployment” surveys of the NSS for the years 1983 and1993, Duraisamy (2002) found that the return to female post-primary educationis higher than that for men in 1983 and also in 1993-94. A study by Bhaumikand Chakrabarty (2008) using 1987 and 1999 rounds of the NSSO employment-unemployment survey found that the gender wage gap narrowed considerably be-tween years 1987 and 1999. The narrowing of the earnings gap was attributedlargely to a rapid increase in the returns to the labor market experience of women.Using nationally representative data from India Human Development Survey (IHDS)2005, Agrawal (2013) found that the wage differential between males and femalescan largely be attributed to discrimination in the labor market. Differences in en-dowments plays a more prominent role in explaining wage difference betweensocial groups.Most of the papers in gender gap literature in Indian context focused on av-erage gap in male-female wages. Khanna (2012) analyzed whether male-femalewage gap differs for different wage levels. Using data from the 2009-10 employment-unemployment schedule of the National sample survey, this paper shows thatmale-female wage gap is higher at the lower end of the wage distribution.It is important to recognize at the outset that the focus of this chapter is on the72evolution of gender gaps amongst full-time workers in the workforce. This hastwo important consequences. First, the evolution of gender gaps amongst part-time workers is outside the ambit of the chapter. While part-time workers arean important component of the workforce, the measurement issues surroundingthis category are too serious to tackle within the confines of this chapter. Second,the chapter is silent about the trends in the labor force participation decisions ofwomen. This is a very important issue, not just for India but for all economies.Indeed, there is a significant amount of work in this area focusing on the USAand other industrial economies that has found evidence of a U-shaped pattern inthe evolution of female labor-force participation rates with participation initiallydeclining with development and rising later on in the development process. Indiatoo has seen a decline in the labor force participation rates of women over the lastten years. Whether or not this is part of the same syndrome that one has observedelsewhere in the west or is it due to some other India-specific factor is somethingthat deserves a paper on its own right. In this chapter we confine ourselves tosummarizing some of this literature in a separate sub-section.The next section presents our results on education and occupation attainmentrates and gender gaps in those indicators. Section 4.3 describes the evolutionof gender wage gaps and their decomposition into measured and unmeasured at-tributes. Section 4.4 analyses the trends in the gender gap of the young workers.Section 4.5 briefly discusses female labor force participation. The last sectionconcludes.734.2 Empirical RegularitiesOur data comes from successive quinquennial rounds of the National Sample Sur-vey (NSS) from 1983 to 2009-10. Specifically, we use rounds 38, 43, 50, 55,61 and 66 of the Employment and Unemployment surveys of the NSS. Givenour interest in labor market characteristics and outcomes, we restrict the sampleto working age adults in the age-group 16-64 who belong to households with amale head of household, who are working full-time and for whom we have in-formation on their education and occupation choices.2 While the overall NSSquinquennial surveys typically sample around 100,000 households (equivalently,around 460,000 individuals on average), our sample restriction reduces the sampleto around 160,000 on average. Table 4.1 gives the demographic characteristics ofthe workforce. Clearly, men and women differ very marginally along these demo-graphic characteristics.Our primary interest lies in examining the evolution of gender gaps in Indiasince 1983 along three dimensions: education, occupation and wages. Given thateducation and occupation choices are two fundamental ingredients in wage out-comes, we start with a closer examination of patterns on these two indicators.Before proceeding we would like to address a potential concern regarding oursample selection. Given that we are going to analyze outcomes of those in thelabor force, one might have legitimate concerns that our findings may be affectedby changes in the gender composition of the labor force. This could occur if there2. We leave out female-led households from the analysis since these households are likely tobe atypical in the generally patriarchal Indian family set-up.74were a differential changes in the proportion of women working full-time relativeto men, in the labor force participation rates of women relative to men or in therelative employment rates of women during the sample period. Figure 4.1 showsthe ratio of male to female rates in labor force participation, employment, fill-time workers and part-time workers. The key point to note is that there are noclear trends in any of these ratios which suggests that our finding are unlikely tobe driven by gender-based differential changes in the participation rates.The characteristics of the workforce in terms of their labor force participationchoices and outcomes may differ across the genders along a number of other mar-gins. One key factor of interest are potential differences between rural and urbanworkers. With a large majority of workers still living in rural India, it is importantto document any differences in labor force behavior between these two sectors.Table 4.2 describes the gender differences in the labor force characteristics ofworkers broken down by rural and urban workers. The key variables we reportare labor force participation rates (LFP), proportion of workers working full time(FULL), proportion working part-time (PART), proportion self employed (SELF),and proportion unemployed (UNMP).The numbers in the table show that the patterns are similar for rural and urbanworkers on most measures. The two key features worth noting are: (a) in both ru-ral and urban areas women are more likely to be working part-time relative to theirmale counterparts; (b) labor force participation rates are higher for rural womenrelative to urban women. In terms of our focus on full-time workers in the analysisbelow, the key point that we would like to emphasize is that the composition of75full-time and part-time workers has not changed much across gender lines duringthe sample period.4.2.1 Education AttainmentEducation attainments of sampled individuals in the NSS survey are reported ascategories: Illiterate, Primary, Secondary, etc.. While we use the category levelinformation for our analysis below, we also generated statistics on years of educa-tion by converting the categories into years of education. This conversion allowsus to represent the trends in a more parsimonious manner. The education cate-gories are mapped to years of education following Hnatkovska, Lahiri, and Paul(2012): not literate = 0 years, literate but below primary = 2 years, primary = 5years, middle = 8 years, secondary and higher secondary =10 years, graduate =15years, and postgraduate =17 yearsTable 4.3 reports the average years of education of the male and female work-force in India across all the rounds. While the overall education level of the work-force was a dismally low 3 years in 1983, the disparity between men and womenworkers was even more dramatic with men having on average around 3.5 years ofeducation while women had less than a year’s schooling. The relative gap in yearsof education between men and women of the Indian workforce was almost 4. By2010, the situation had improved, albeit slightly. The relative gap had declined toabout 1.7 with men having on average about 6.2 years of schooling while womenhad 3.6 years. There clearly has been some decline in the education gender gap.The evidence on years of education does not reveal where and how the catch-76up in education levels has been occurring. Did the decline in the gender gap inyears of education happen primarily due to women moving out of illiteracy ordue to more women moving past middle and secondary school? This question isimportant to since the addition of a year of education is likely to have very dif-ferent effects depending on what kind of education is that extra year acquiring.We collect the education levels reported in the NSS survey into five categories:illiterate (Edu1), some education (Edu2), primary (Edu3), middle (Edu4) and sec-ondary and above (Edu5). The last category collects all categories from secondaryand above. Given the relatively limited representation of workers in some of thehigher education categories at the college and beyond, this allows a relatively evendistribution of individuals across categories.Panel (a) of Figure 4.2 shows the distribution of men by education categoryon the left and the corresponding distribution of women on the right. The figureillustrates the direness of the education situation in India. In 1983, 70 percentof male workers had primary or below education levels while the correspondingnumber for women workers was 90 percent. The period since then has witnessedimprovements in these with the proportion of men with primary or lower educationlevel declining to 40 percent by 2010 while that for women it fell to around 60percent. At the other end of the education spectrum, in 1983 around 15 percent ofmen and 5 percent of women workers had secondary or higher education levels.By 2010 the share of this category had risen to 40 percent for men and 25 percentfor women.Panel (b) of Figure 4.2 looks at the change in the share of women in each77education category over time. The figure makes clear that women have been in-creasing their share in every education category except for Edu1 (illiterate) wherethe share has stayed unchanged. The fastest rise in the share of women occurredin education categories 2, 3 and 4 (some education, primary and middle school).Overall, the figure suggests that the education catch-up has been fairly uniformacross categories.Are the measured narrowing of the gender education gaps as suggested by thedata on years of education as well as categories of education statistically signif-icant? We examine this by estimating an ordered probit regression of educationattainment (measured by education category) on a constant and a female dummy.We do this for each sample round. Table 4.4 gives the marginal effect of the fe-male dummy in each round, the changes in the marginal effect across specifiedrounds as well as the statistical significance of the estimates. The estimates indi-cate that being female significantly increased the probability of being illiterate andsignificantly reduced the probability of being in all other education categories in1983. Over the subsequent 27 years, this over-representation of females amongstilliterate workers and under-representation in other categories declined for all cat-egories except for the secondary and above category. Moreover, the changes overtime were statistically significant.3In summary, our review of the education attainment levels of men and women3. We should note that the marginal effect of the female dummy measures its effect on theabsolute gap between the probability of that category between the genders. Hence, this is differentfrom the relative gap numbers reported in Figure 4.2 which reports trends in the relative gap inthe probabilities. This explains the difference in our results for the convergence patterns in Edu5category in Figure 4.2 and Table 4.4.78in the Indian labor force suggests that gender gaps in education have declinedsignificantly over the past three decades though the absolute levels of educationin the country remain unacceptably low. Additionally, while more women arejoining the labor force with secondary school or higher education, they have beennot done this fast enough to consistently raise their share of secondary and aboveeducated workers. This partly also be reflecting the fact that secondary educatedwomen in India are still not joining the labor force at high enough rates.4.2.2 Occupation ChoicesOur next indicator of interest is the occupational choice of the workforce. Specif-ically, we want to examine differences in the occupational choices between menand women workers in the workforce and how those differences have evolved overtime. We use the 3-digit occupation classification reported in NSS and aggregatethem into three broad occupational categories: Occ1: white-collar occupationslike administrators, executives, managers, professionals, technical and clericalworkers; Occ2: blue-collar occupations such as sales workers, service workersand production workers; and Occ3: Agrarian occupations which collects farmers,fishermen, loggers, hunters etc..Figure 4.3 shows the key features of the occupation distribution patterns of theworkforce broken down by gender. Panel (a) shows the distribution of the maleworkforce across the three occupation categories and the corresponding distribu-tion of female members of the workforce. The two graphs in panel (a) clearlyshow a robust pattern of occupational churning in the entire labor force: work-79ers of both genders have been switching out of agrarian occupations. The shareof agriculture in male full-time employment declined from around 50 percent in1983 to 30 percent in 2010. Correspondingly, the share of agriculture in femalefull-time employment also fell, albeit more tepidly, from 70 to 55 percent duringthe same period. The share of blue-collar employment for males rose from around40 to 50 percent while that of white-collar employment rose from 10 to around 20percent. Women, by contrast, saw blue-collar employment’s share in their totalemployment in 2010 rise slightly above its 1983 level of just under 25 percent.White collar employment of women however rose sharply from 5 to just under 20percent between 1983 and 2010.Panel (b) of Figure 4.3 shows the share of women in total full-time employ-ment in each occupation. Note that this is in contrast to Panel (a) which showed theshare of each occupation in total full-time female employment. The most visiblechange in the share of women is in Occ1 which is white-collar employment wherewomen’s share has increased from 10 to 15 percent between 1983 and 2010. Theshare of women in total employment in the other two occupations has not changedmuch during this period.The trends documented above suggest that women have been changing occu-pations during this period. Has this resulted in a decline in the gender disparitiesin the occupation distribution of the labor force? We answer this question byrunning a multinomial logit regression of occupational choice on a constant anda female dummy for each round. We then compute changes in the effect of thefemale dummy across the rounds. Table 4.5 shows the results. In a confirma-80tion of the visual suggestion above, in 1983 being female significantly increasedthe probability of being employed in agriculture while significantly reducing theprobability of employment in blue and white collar jobs (Occ2 and Occ1, respec-tively). While this basic pattern has not changed between 1983 and 2010, thenegative marginal effect of the female dummy on the probability of white-collaremployment declined significantly during this period indicating that there wasstatistically significant reduction in the under-representation of women in theseoccupations during this period. The other two broad occupation categories how-ever, showed a worsening of the initial disparity of representation with the over-representation of women in agricultural employment and under-representation inblue-collar occupations marginally worsening between 1983 and 2010.In summary, our review of the trends in the disparity between the gendersin their occupation distribution suggests a mixed picture. On the positive side,women have been moving out of agricultural jobs into blue and white collar jobsthereby behaving more like their male counterparts in the workforce. However,in terms of the share of women in the different occupations, only white-collarjobs have seen a significant expansion of the share of women while the under-representation in blue-collar jobs and over-representation in agrarian jobs has in-creased. This latter effect suggests to us that women have been moving out ofagricultural jobs and into blue-collar jobs at a slower rate than their male counter-parts.814.3 Wage Outcomes and Gender DifferencesWe now turn our attention to the third indicator of interest – gender gaps inwages. In terms of background, it is worth reiterating that two key determinantsof wages of individual workers are their education levels and the occupations thatthey choose. In the previous section we have shown that gender gaps in educationhave tended to narrow for all but the highest education groups. This trend is likelyto be a force towards raising the relative wage of women. We have also shown thatwomen’s share of employment has only increased in white-collar occupations. Inas much as women are getting disproportionately more represented in agriculturaljobs, one might expect this force to lower the relative wage of women if agriculturepays relatively lower wages. Clearly, there are offsetting underlying forces here.The NSS only reports wages from activities undertaken by an individual overthe previous week (relative to the survey week). Household members can under-take more than one activity in the reference week. For each activity we knowthe “weekly” occupation code, number of days spent working in that activity, andwage received from it. We identify the main activity for the individual as the onein which he spent maximum number of days in a week. If there are more thanone activities with equal days worked, we consider the one with paid employment(wage is not zero or missing). Workers sometimes change the occupation due toseasonality or for other reasons. To minimize the effect of transitory occupations,we only consider wages for which the weekly occupation code coincides withusual occupation (one year reference). We calculate the daily wage by dividing82total wage paid in that activity over the past week by days spent in that activity.Figure 4.4 shows the evolution of the gender wage gaps since 1983. Panel (a)shows the mean and median wage gaps across the rounds while Panel (b) showsthe wage gap across all percentiles for three different years: 1983, 2004-05 and2009-10. Two points are worth noting from the figure. First, the gender wagegap has shrunk secularly since 1983 for all groups except the very richest groups.In other words, the decline in the gender wage gap has been broad-based andinclusive. Second, there has been a very sharp decrease in the gender wage gapbetween 2004-05 and 2009-10. Uncovering the reasons behind this phenomenonis interesting in its own right.Are the measured decreases in the wage gap statistically significant? We testthis by running regressions of the log wage on a constant, a female dummy andcontrols for age and age squared (to control for potential lifecycle differences be-tween men and women related to their labor supply choices). We run the regres-sion for different quantiles as well as for the mean.4 Table 4.6 shows the results.The regression results show that the decline in the wage gaps were significant forall income groups except the 90th percentile for whom there was no significantchange in the wage gap between 1983 and 2010. Moreover, there was also astatistically significant decrease in the wage gap between 2004-05 and 2009-10.So, what is driving the wage convergence between the genders? Specifically,how much of the decrease in the gender wage gap is due to convergence in mea-4. We use the Recentered Influence Function (RIF) regressions developed by Firpo, Fortin,and Lemieux (2009) to estimate the effect of the female dummy for different points of the wagedistribution.83sured attributes of workers? To understand the time-series evolution of the genderwage gaps we use the Oaxaca-Blinder decomposition technique to decompose theobserved changes in the mean and quantile wage gaps between 1983 and 2010into explained and unexplained components as well as to quantify the contribu-tion of the key individual covariates. We employ Ordinary Least Squares (OLS)regressions for the decomposition at the mean, and Recentered Influence Function(RIF) regressions for decompositions at the 10th, 50th, and 90th quantiles.5 Ourexplanatory variables are demographic characteristics such as individual’s age,age squared, caste, and geographic region of residence. Additionally, we con-trol for the education level of the individual by including dummies for educationcategories.6The results of the Oaxaca-Blinder decomposition exercise are reported in Ta-ble 4.7. The table shows that all of the gender wage convergence for the medianand around 75 percent of it for the mean can be accounted for by measured covari-ates. For the 10th percentile measured covariates explain around 50 percent of theobserved convergence. Encouragingly, convergence in education was a key con-tributor to the observed wage convergence for all these groups.7 The convergence5. The inter-temporal decomposition at the mean is in the spirit of Smith and Welch (1989).All decompositions are performed using a pooled model across men and women as the referencemodel. Following Fortin (2006) we allow for a group membership indicator in the pooled regres-sions. We also used 1983 round as the benchmark sample.6. We do not include occupation amongst the explanatory variables since it is likely to be en-dogenous to wages.7. As we show below, adding occupation choices to the list of explanatory variables does notsignificantly raise the share of the explained component in the observed wage convergence. Thisis not unusual. Blau and Kahn (2007) report that over 40 percent of the gender wage gap in theUSA remains unexplained even after accounting for a rich set of explanatory variables includingeducation, race, occupation, industry, union status, experience, etc..84at the 90th percentile between 1983 and 2010 however cannot be explained bymeasured covariates. In fact, the observable covariates of wages predict that thegender wage gap should have actually widened rather than narrowed The sourceof the wage convergence at the 90th percentile is thus a puzzle as it is almostentirely unexplained.To gain greater perspective on the underlying forces driving the contractionin the gender wage gap, Panel (a) of Figure 4.5 shows the gender wage gaps byeducation category. Examining panel (a) it is clear that the dispersion in the wagegap by education category has declined perceptibly since 1983. Moreover, genderwage gaps have declined sharply for groups with some education (edu2), primaryeducation (edu3) and those with middle school education (edu4) while increas-ing slightly for illiterates and those with secondary and above education. Sincewomen have been increasing their representation in education categories 2,3 and4 while reducing their relative representation in categories 1 and 5, the behaviorof the wage gaps by education category in panel (a) of Figure 4.5 suggests whyeducation accounts for a large part of the observed gender wage convergence.Panel (b) of Figure 4.5 gives the median wage gaps by occupation category.The median wage gaps were the highest in blue-collar jobs (occ2) and used to bethe lowest in white collar jobs (occ1) in 1983. By 2010, the wage gaps in thesetwo occupations had converged while the wage gap in agrarian jobs remained rel-atively unchanged. Recall from Table 4.5 that between 1983 and 2010 womenreduced their under-representation in white-collar occupations. At the same timetheir over-representation in agrarian jobs rose and the under-representation in85blue-collar occupations worsened.The effect of occupation choices on the wage gap is thus ambiguous. Onthe one hand, the movement of women towards white-collar occupations that hadlower average wage gaps would have tended to lower the wage gap. The increasedunder-representation in blue-collar jobs, typically characterized by high genderwage gaps, would also tend to lower the overall wage gap as would the declinein the wage gap over time in that occupation. However, the increase in the wagegap in white-collar occupation over time would have had the opposite effect ofwidening the wage gap.In summary, our results on wage outcomes of the workforce indicate that thegender wage gap has narrowed significantly. Most of this convergence was dueto convergence in measured covariates of wages. Additionally, there has been avery sharp convergence in male and female wages between 2004-05 and 2009-10.While the reasons behind this require more careful examination, our preliminaryexamination of the issue suggest that a narrowing of the gender gap in educationwas a key contributing factor. It is tempting to attribute the convergence to fac-tors such as the National Rural Employment Guarantee Program (NREGA) whichguarantees 100 days of work in the off-season to every rural household. However,we don’t believe that our results are driven by NREGA for a couple of reasons.First, as Figure 4.4 illustrates clearly, the convergent trends pre-date the introduc-tion of NREGA (which was only introduced in 2006). Second, the convergentpatterns characterize both rural and urban areas whereas NREGA only applied torural areas. Clearly, some factors that were common to both rural and urban ar-86eas are likely to have been at play rather than a rural India specific program likeNREGA.4.4 The YoungThe trends documented above do suggest significant narrowing in gender gapsacross multiple categories. However, a key reason for examining these trends isto also anticipate what might one expect to see over the next couple of decades interms of gender disparities. While forecasting such trends are very difficult, onemeasure which usually provide windows into future trends would be the trends inthe gender gaps of the young workers.To probe this more closely, Figure 4.6 shows that the primary force driving thecatch-up in education is the increasing education levels of younger cohorts. Thus,in 1983 the relative gender gap in years of education between men and womenworkers aged 16-22 was 3. By comparison, in 2005, the education gap was 1.4for the 17-23 year old cohort who were born between 1982-88. Clearly the gap islower for younger birth cohorts.We take a closer look at the gaps amongst younger workers by concentratingon the characteristics of 16-25 year olds in each survey round. We start with ed-ucation. Figure 4.7 reports the years of education of the 16-25 year olds, brokendown by females and males, and by rural and urban. As can be seen from theFigure, young workers in the 16-25 age group have been increasing their yearsof education in both rural and urban India. Moreover, in both areas the gap be-tween men and women has narrowed sharply. Perhaps, most impressively, in 201087women workers in urban areas had more years of education on average than theirmale counterparts. Even in rural India, in 2010 the gap was just above 1 year forthis group. These trends suggest that over the next two decades, the gender gapin education should become very small. These trends would get even stronger asmore and more educated women begin participating in the labor force.How have the 16-25 year olds been behaving in terms of their occupationchoices? Are there significant differences between the genders on this dimen-sion? Figure 4.8 shows the occupation choices of women (Panel (a)) and men(Panel (b)). The patterns are very similar for the two. The share of agriculturaloccupation have declined while the share of the other two occupations have risenfor both men and women between 1983 and 2010. In terms of comparisons of theoccupation distribution, by 2010, the share of the female workforce in the 16-25age group that was engaged in white-collar jobs was marginally higher than thecorresponding proportion for male workforce in the 16-25 age-group. On the otherhand, while women in this age group have been switching out of agriculture intoblue-collar occupations, their male counterparts in the same age group have beendoing so at a faster rate. Consequently, even in 2010 almost 60 percent of youngfemale workers were engaged in agrarian jobs while blue-collar jobs accountedfor only 30 percent of their employment. The corresponding numbers for youngmale workers on the other hand were 50 percent and 40 percent, respectively. Thekey though is that the gaps have narrowed much faster for these younger workersas compared to their older counterparts.The rapidly shrinking gender gaps amongst younger workers suggests to us88that going forward gender gaps are likely to narrow even faster as more and moreof the older cohorts drop out of the labor force and more younger cohorts withsimilar education and occupation choices) replace them in the workforce.4.5 Female Labor Force ParticipationA number of existing studies found that a U-shaped relationship exists between fe-male labor force participation and economic development (Goldin (1995); Mam-men and Paxson (2000); Kottis (1990); Fatima and Sultana (2009) ). They arguethat in low income societies, women work on family farms or enterprises and thusfemale labor force participation is high. As society gets richer there is higher focuson industrialization. Thus blue collar jobs becomes more important and woman’sparticipation in the labor market falls accordingly. This can be explained by risingfamily income, incompatibility of factory work with child care and social stigmaassociated with working outside home. With further economic development, fe-male labor force participation increases once again due to the expansion of highereducation among females and the emergence of a white-collar jobs. The stigmasassociated with jobs disappear overtime and at such advanced stages of develop-ment, the substitution effect on account of higher female wages dominates theincome effect.Empirical support for the U-hypothesis is primarily based on cross-countryanalysis (Mammen and Paxson (2000), C¸ag˘atay and O¨zler (1995)). Panel analy-ses, on the other hand, have produced mixed results. While Luci (2009) and Tam(2011) have argued that the U-shaped LFP hypothesis has support within coun-89tries over time, Gaddis and Klasen (2014) found that the evidence of a U-shapedrelationship is weak and extremely sensitive to underlying data.In the Indian context, there is mixed evidence on the U-shaped relationship.On the one hand, Olsen and Mehta (2006) found that a U-shaped relationship ex-ists between female employment and educational status. Using 1999-2000 NSSdata, they found that women with low education as well as those with univer-sity degrees more likely to work than middle educated women. Using panel databetween 1983-2010 from the National Sample survey, Lahoti and Swaminathan(2013) however did not find a significant relationship between level of economicdevelopment and woman’s participation rates in the labor force. Female laborparticipation rates tend to also vary between rural and urban areas and across sub-rounds of the NSS data, as shown by Bardhan (1984).As the discussion above makes clear, female labor force participation is a com-plicated subject that requires a separate paper on its own. We hope to return tothis issue in future work.4.6 ConclusionAllocating talent is one of the major challenges for any country. It is an evenbigger issue in rapidly developing economies with their changing economic struc-ture. This chapter has examined one aspect of this talent allocation process byexamining the evolution of gender gaps in India since 1983. The absolute differ-ences between males and females in the Indian labor force are huge in a numberof different indicators including education attainment rates, labor force participa-90tion rates, occupation choices as well as wages. However, the gaps have narrowedalong all these indicators in the last 27 years. Most encouragingly, the major-ity of the wage convergence is accounted for by measured covariates of wages,particularly education.This study has ignored some key areas that can shed greater light on the evo-lution of gender gaps. First, the study has focused on aggregate India-wide trends.Given the huge variation in policies and outcomes across states in India since1983, one profitable approach would be to exploit the cross-state differences tobetter identify the causal channels at work. This left for future research. Second,trends in female labor force participation rates in India need to be explored. Thishas first-order implications for gender disparities but comes with a host of dataand conceptual issues that render a full-scale examination of it difficult in thischapter.91Table 4.1: Sample Summary StatisticsMale FemaleSample SampleAge SCST Married share Rural Age SCST Married share Rural1983 35.55 0.25 0.79 0.79 0.75 33.69 0.36 0.85 0.21 0.861987-88 35.82 0.26 0.8 0.79 0.77 33.82 0.35 0.87 0.21 0.871993-94 36.11 0.27 0.8 0.79 0.75 34.62 0.35 0.86 0.21 0.861999-00 36.27 0.28 0.8 0.76 0.74 35.22 0.38 0.88 0.24 0.862004-05 36.63 0.27 0.8 0.78 0.73 35.91 0.35 0.86 0.22 0.842009-10 37.68 0.28 0.81 0.81 0.71 36.71 0.36 0.86 0.19 0.81Notes: This table reports summary statistics for the sample. The statistics are reported at the individuallevel.92Figure 4.1: Gender Gaps: Labor Market Participation RatesNotes: The figure shows the ratio of male to female rates in labor force participation, em-ployment, filltime workers and part-time workers for the six survey rounds.93Table 4.2: Labor Market Characteristics by Gender: Rural and Urban Work-ersMale FemaleRound LFP FULL PART SELF UNMP LFP FULL PART SELF UNEMPPanel A: Rural1983 0.9365 0.9578 0.0422 0.6131 0.0354 0.3567 0.8557 0.1443 0.6001 0.04381987-88 0.9417 0.966 0.034 0.5844 0.0396 0.3449 0.8965 0.1035 0.5692 0.04121993-94 0.9512 0.9665 0.0335 0.5836 0.0291 0.4188 0.8246 0.1754 0.614 0.02981999-00 0.9439 0.9626 0.0374 0.5561 0.0365 0.4163 0.8323 0.1677 0.5927 0.03512004-05 0.9528 0.9567 0.0433 0.5873 0.0354 0.4557 0.7912 0.2088 0.6661 0.03982009-10 0.9511 0.97 0.03 0.5361 0.0297 0.3477 0.8127 0.1873 0.5849 0.0357Panel B: Urban1983 0.9352 0.977 0.023 0.3941 0.06 0.1819 0.8933 0.1067 0.412 0.08081987-88 0.9345 0.9834 0.0166 0.4026 0.0614 0.1877 0.9162 0.0838 0.4213 0.09841993-94 0.9366 0.9858 0.0142 0.4074 0.0467 0.2173 0.8634 0.1366 0.4515 0.09011999-00 0.9275 0.984 0.016 0.4015 0.0518 0.1981 0.8745 0.1255 0.4453 0.07812004-05 0.931 0.9808 0.0192 0.4396 0.0475 0.2383 0.8561 0.1439 0.4957 0.0922009-10 0.9279 0.9876 0.0124 0.4085 0.0302 0.198 0.8804 0.1196 0.4217 0.079Notes: This table reports the labor force characteristics of men and women separately for rual and urban workers. LFPindicates Labor Force Participation rates, FULL is proportion of workers working full-time, PART are proportions ofpart-time workers, SELF indicate proportion of self-employment and UNEMP denotes the unemployent rate.94Table 4.3: Education Gaps: Years of SchoolingRelativeAverage years of education education gapOverall Male Female1983 2.99 3.54 0.93 3.83***(0.01) (0.01) (0.02) (0.08)1987-88 3.19 3.75 1.15 3.25***(0.01) (0.01) (0.02) (0.06)1993-94 3.82 4.42 1.55 2.86***(0.01) (0.02) (0.02) (0.04)1999-00 4.32 5.05 2 2.53***(0.02) (0.02) (0.03) (0.04)2004-05 4.82 5.44 2.64 2.06***(0.02) (0.02) (0.03) (0.02)2009-10 5.71 6.21 3.59 1.73***(0.03) (0.03) (0.06) (0.03)Notes: This table presents the average years of education forthe overall sample and separately for males and females; aswell as the gap in the years of education. The reported statisticsare obtained for each NSS survey round which is shown in thefirst column. Standard errors are in parenthesis. * p<0.1, **p<0.05, *** p<0.0195Figure 4.2: Distribution of Workforce Across Education Categories(a)(b)Notes: Panel (a) of this Figure presents the education distribution of each gender into thedifferent education categories. Panel (b) shows the share of women in all workers in eachcategory.96Table 4.4: Marginal Effect of Female Dummy on Education CategoriesPanel A. Marginal effects of female dummy Panel B.Changes1983 1987-88 1993-94 1999-00 2004-05 2009-10 1983-2005 1983-2010Edu 1 0.3760*** 0.3641*** 0.3582*** 0.3460*** 0.3011*** 0.2482*** -0.0749*** -0.1278***(0.003) (0.003) (0.0035) (0.0036) (0.004) (0.0062) (0.005) (0.0069)Edu 2 -0.0607*** -0.0531*** -0.0367*** -0.0180*** 0.0008* 0.0165*** 0.0615*** 0.0772***(0.001) (0.0009) (0.0008) (0.0006) (0.0005) (0.0006) (0.0011) (0.0012)Edu 3 -0.0971*** -0.0879*** -0.0648*** -0.0460*** -0.0335*** -0.0099*** 0.0636*** 0.0872***(0.0012) (0.0011) (0.001) (0.0009) (0.0009) (0.0009) (0.0015) (0.0015)Edu 4 -0.0935*** -0.0851*** -0.0884*** -0.0883*** -0.0790*** -0.0555*** 0.0145*** 0.038***(0.0011) (0.001) (0.0011) (0.0012) (0.0013) (0.0018) (0.0017) (0.0021)Edu 5 -0.1247*** -0.1380*** -0.1683*** -0.1937*** -0.1895*** -0.1992*** -0.0648*** -0.0745***(0.0011) (0.0011) (0.0014) (0.0018) (0.0021) (0.004) (0.0024) (0.0041)N 164979 182384 163126 173309 176968 133926Notes: Panel (a) reports the marginal effects of the female dummy in an ordered probit regression of education categories 1 to 5 ona constant and a female dummy for each survey round. Panel (b) of the table reports the change in the marginal effects over statedperiods and over the entire sample period. N refers to the number of observations. Standard errors are in parenthesis. * p-value≤0.10,** p-value≤0.05, *** p-value≤0.01.97Figure 4.3: Distribution of Workforce Across Occupation Categories(a)(b)Notes: Panel (a) of this Figure presents the occupation distribution of each gender into thedifferent occupation categories. Panel (b) shows the share of women in each category.98Table 4.5: Marginal Effect of Female Dummy on Occupational CategoriesPanel A. Marginal effects of female dummy Panel B.Changes1983 1987-88 1993-94 1999-2000 2004-05 2009-10 1983-2005 1983-2010Occ1 -0.0564*** -0.0488*** -0.0407*** -0.0512*** -0.0370*** -0.0394*** 0.0194*** 0.017***(0.0016) (0.0015) (0.002) (0.0022) (0.0024) (0.004) (0.0029) (0.0043)Occ2 -0.1172*** -0.1155*** -0.1481*** -0.1756*** -0.1670*** -0.1592*** -0.0498*** -0.042***(0.0031) (0.0031) (0.0031) (0.0034) (0.0037) (0.0055) (0.0048) (0.0063)Occ3 0.1736*** 0.1644*** 0.1888*** 0.2268*** 0.2040*** 0.1986*** 0.0304*** 0.025***(0.0033) (0.0033) (0.0035) (0.0037) (0.0041) (0.0064) (0.0053) (0.0072)N 164979 182384 163126 173309 176968 133926Notes: Panel (a) of the table presents the marginal effects of the female dummy from a multinomial probit regression of occupationchoices on a constant and a female dummy for each survey round. Panel (b) reports the change in the marginal effects of the ruraldummy over the relevant time periods. Agrarian jobs is the reference group in the regressions. N refers to the number of observations.Standard errors are in parenthesis. * p-value≤0.10, ** p-value≤0.05, *** p-value≤0.01.99Figure 4.4: Gender Wage Gaps Since 19831.51.61.71.81.921983 1993−94 1999−00 2004−05 2009−10 mean gap median gapRelative wage gap(a).2.3.4.5.6.7.8.91lnwage(Male)−lnwage(Fem)0 10 20 30 40 50 60 70 80 90 100percentile1983 2004−05 2009−10(b)Notes: Panel (a) of this Figure presents the relative male to female wage for full-time work-ers. Panel (b) shows the log ratio of male to female wages for each percentile.100Table 4.6: Changes in the Gender Wage GapPanel A: Female dummy coefficient Panel B: Changes1983 1993-1994 1999-2000 2004-2005 2009-2010 1983-2005 1983-201010th Perc -0.8851*** -0.6020*** -0.4727*** -0.7737*** -0.6035*** 0.1114*** 0.2816***(0.0193) (0.0157) (0.0129) (0.0199) (0.0277) (0.0277) (0.0338)50th Perc -0.6872*** -0.6064*** -0.6115*** -0.5164*** -0.3690*** 0.1708*** 0.3182***(0.0097) (0.0089) (0.009) (0.0086) (0.0112) (0.013) (0.0148)90th Perc -0.3543*** -0.3506*** -0.4141*** -0.4073*** -0.3841*** -0.0530*** -0.0298(0.01) (0.0132) (0.0184) (0.0235) (0.0354) (0.0255) (0.0368)Mean -0.6604*** -0.5641*** -0.5810*** -0.5777*** -0.4622*** 0.0827*** 0.1982***(0.0083) (0.0095) (0.0095) (0.01) (0.0139) (0.013) (0.0162)N 63981 63364 67322 64359 57339Notes: Panel (a) of this table reports the coefficient on the female dummy in a regression of log wages on a constant,a female dummy and controls for age (age and age squared). Panel (b) reports changes in the coefficient across therelevant rounds. N refers to the number of observations. Standard errors are in parenthesis. * p-value≤0.10, **p-value≤0.05, *** p-value≤0.01.101Table 4.7: Decomposition of the Changes in the Wage GapMeasured gap Explained Unexplained Explained by educationChange (1983 to 2009-10)10th Perc -0.1220*** -0.0638*** -0.0582** -0.0241***(0.0267) (0.0097) (0.0273) (0.0078)50th Perc -0.2102*** -0.2452*** 0.0349 -0.1378***(0.0287) (0.0143) (0.0257) (0.0099)90th Perc -0.1665*** 0.1484*** -0.3148*** 0.0455*(0.0569) (0.0352) (0.0544) (0.0259)Mean -0.2157*** -0.1512*** -0.0645*** -0.0891***(0.0169) (0.0105) (0.0158) (0.0083)Notes: This table presents the change in the rural-urban wage gap between 1983 and2009-10 and its decomposition into explained and unexplained components using theRIF regression approach of Firpo, Fortin, and Lemieux 2009 for the 10th, 50th and 90thquantiles and using OLS for the mean. The table also reports the contribution of educationto the explained gap (column (iv)). Bootstrapped standard errors are in parenthesis. * p-value≤0.10, ** p-value≤0.05, *** p-value≤0.01.102Figure 4.5: Gender Wage Gaps by Education and Occupation Categories(a)(b)Notes: Panel (a) of this Figure presents the relative male to female median wage gap byeducation category while Panel (b) shows the median wage gap between men and women indifferent occupations.103Figure 4.6: Education Gaps in Years by Birth Cohorts12345671983 1987−88 1993−94 1999−00 2004−05 2009−10 1926−32 1933−39 1940−46 1947−53 1954−601961−67 1968−74 1975−81 1982−88 1989−94OverallNotes: The figure shows the relative gap in years of education between males and femalesfor the six survey rounds for different birth cohorts.104Figure 4.7: Gap in Years of Education: 16-25 Year Olds(a)(b)Notes: The figure reports the overall years of education as well as the years of education offemale and male workers in the 16-25 age-age-group. Panel (a) of this Figure correspondsto the rural sector and panel (b) corresponds to the urban sector.105Figure 4.8: Occupational Distribution of 16-25 Year Olds(a)(b)Notes: Panel (a) of this Figure presents the occupation distribution of female workers in the16-25 age cohort across the six survey rounds. Panel (b) shows the corresponding figuresfor urban male workers aged 16-25.106Chapter 5ConclusionThis thesis attempts to understand the factors that influence two major problems inIndia; high infant mortality and fertility. It also explores the distributional impactsof the rapid rate of growth that India has faced in the last three decades on thelabor market characteristics and outcomes of women vis-a-vis men.In the economics literature, there is substantial evidence that politicians havestrong incentives to signal their ability to improve the economic conditions of vot-ers in election years. However, there has been very little research on the welfareimplication of these election cycles. In Chapter 2, I attempt to test whether thetiming of elections have any impact on child survival in India. In particular, I haveanalyzed whether children born before scheduled state assembly election yearsare more likely to survive their infancy as compared to children born in the off-election years and if this impact is higher for districts where the previous electionwas particularly close. My work shows that children born before the scheduledstate assembly election years are less likely to die in infancy as compared to chil-dren born in off-election years. Moreover the decline in infant mortality rate is107higher in areas where the state ruling party had a narrow margin of victory in theprevious election. The results also show that children born before scheduled elec-tion years are less likely to suffer from low birth weight and the mothers are likelyto have more regular antenatal checkups. The probability that mothers receivestetanus injection during pregnancy is significantly higher for children born in theelection years.It is generally held that there exists a negative relationship between fertilityand female employment (Cramer, 1980; Felmlee, 1993; Smith-Lovin and Tick-amyer, 1978). The association between fertility and female employment reflectsthe incompatibility between child care and work force participation. Since child-rearing is intensive in mother’s time, female employment decreases the number ofchildren. However evidence on the relationship between female employment andfertility in developing countries is limited. The chapter 3 of my thesis empiricallyexamines the causal impact of female labor force participation on the whole andfemale employment in different sectors on the fertility rate. The results show thatwhile female labor force participation and female employment in agriculture hasno statistically significant effect on fertility, female employment in manufacturingsignificantly reduces fertility.Chapter 4 looks at the evolution of male-female gap in India in the last threedecades. This is a joint work with Dr. Viktoria Hnatkovska and Dr. AmartyaLahiri. Using data from six consecutive rounds of the employment-unemploymentschedule of National Sample Survey covering the period 1983-2010, we havestudied the evolution of gaps in education attainment rates, occupation choices,108and wages in India across male and female workers. Our primary finding isthat there has been significant convergence in these indicators over the last threedecades.109BibliographyAgrawal, Tushar. 2013. “Gender and caste-based wage discrimination in India:some recent evidence.” Journal for Labour Market Research 47 (4): 1–12.Ahn, Namkee, and Pedro Mira. 2002. “A note on the changing relationship be-tween fertility and female employment rates in developed countries.” Journalof population Economics 15 (4): 667–682.Akhmedov, Akhmed, and Ekaterina Zhuravskaya. 2004. “Opportunistic PoliticalCycles: Test in a Young Democracy Setting.” The Quarterly journal of eco-nomics 119 (4): 1301–1338.Alesina, Alberto. 1988. “Macroeconomics and Politics.” In NBER Macroeco-nomics Annual 1988, 3:13–62. Cambridge, Mass.: MIT Press.Alesina, Alberto, and Nouriel Roubini. 1992. “Political Cycles in OECD Economies.”The Review of Economic Studies 59 (4): 663–688.Anukriti, S, and Todd J Kumler. 2015. “Women's Worth: Trade, Female Income,and Fertility in India.”Arriaga, Eduardo E. 1983. “Estimating fertility from data on children ever bornby age of mother.” International Research Document No. 11.Bardhan, Pranab. 1984. Land, Labor and Rural Poverty. Oxford University Press,New York.Becker, Gary S. 1960. “An economic analysis of fertility.” In Demographic andeconomic change in developed countries, 209–240. Columbia University Press.110Berger, Helge, and Ulrich Woitek. 1997. “How opportunistic are partisan Germancentral bankers: Evidence on the Vaubel hypothesis.” European Journal ofPolitical Economy 13 (4): 807–821.Berman, 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. 2001. “Political Agency, Government Re-sponsiveness and the Role of the Media.” European Economic Review 45(4): 629–640.Bhaumik, Sumon Kumar, and Manisha Chakrabarty. 2008. “Does move to markethave an impact on earnings gap across gender? Some evidence from India.”Applied Economics Letters 15 (8): 601–605.Black, Sandra E, Paul J Devereux, and Kjell G Salvanes. 2007. “From the Cra-dle to the Labor Market? The Effect of Birth Weight on Adult Outcomes.”Quarterly Journal of Economics 122 (1): 409–439.Blau, Francine D., and Lawrence M. Kahn. 2007. “The Gender Pay Gap: HaveWomen Gone as Far as They Can?” The Academy of Management Perspec-tives 21 (1): 7–23.Buekens, Pierre, and Allen Wilcox. 1993. “Why do small twins have a lower mor-tality rate than small singletons?” American journal of obstetrics and gyne-cology 168 (3): 937–941.C¸ag˘atay, Nilu¨fer, and S¸ule O¨zler. 1995. “Feminization of the labor force: the ef-fects of long-term development and structural adjustment.” World develop-ment 23 (11): 1883–1894.Caldwell, JC, and P Caldwell. 1987. “The cultural context of high fertility in sub-Saharan Africa.” Population and Development Review 13 (3): 409–37.Chattopadhyay, Raghabendra, and Esther Duflo. 2004. “Women as Policy Makers:Evidence from a Randomized Policy Experiment in India.” Econometrica 72(5): 1409–1443.111Clark, Shelley. 2000. “Son preference and sex composition of children: Evidencefrom India.” Demography 37 (1): 95–108.Cole, Shawn. 2009. “Fixing Market Failures or Fixing Elections? AgriculturalCredit in India.” American Economic Journal: Applied Economics 1 (1):219–250.Crafts, Nicholas FR. 1989. “Duration of marriage, fertility and women’s employ-ment opportunities in England and Wales in 1911.” Population Studies 43(2): 325–335.Cramer, JC. 1980. “Fertility rate and female employment: problems of causal di-rection.” American Sociological Review 45 (2): 167–90.Cropper, M, and C Griffiths. 1994. “The interaction of population growth andenvironmental quality.” American Economic Review 84 (2): 250–4.Cutler, David, Angus Deaton, and Adriana Lleras-Muney. 2006. ““The Determi-nants of Mortality.” Journal of Economic Perspectives 20 (3): 97–120.Cutler, David, Winnie Fung, Michael Kremer, Monica Singhal, and Tom Vogl.2010. “Early-life Malaria Exposure and Adult Outcomes: Evidence fromMalaria Eradication in India.” American Economic Journal: Applied Eco-nomics 2 (2): 72–94.Dahlberg, Matz, and Eva Johansson. 2002. “On the Vote-Purchasing Behavior ofIncumbent Governments.” American Political Science Review 96 (01): 27–40.Debroy, Bibek, and A.T. Santhanam. 2001. “Matching Trade Codes with Indus-trial Codes.” Population and Development Review 27 (1): 33–63.Deshingkar, Priya, and Shaheen Akter. 2009. Migration and Human Develop-ment in India. Technical report. Human Development Report Office (HDRO),United Nations Development Programme (UNDP).Deshingkar, Priya, and Edward Anderson. 2004. “People on the Move: New Pol-icy Challenges for Increasingly Mobile Populations.” Natural Resource Per-spectives, no. 92.112Drazen, Allan, and Marcela Eslava. 2010. “Electoral manipulation via voter-friendlyspending: Theory and evidence.” Journal of Development Economics 92 (1):39–52.Dre`ze, Jean, and Mamta Murthi. 2001. “Fertility, education, and development:evidence from India.” Population and Development Review 27 (1): 33–63.Duraisamy, Palanigounder. 2002. “Changes in returns to education in India, 1983–94: by gender, age-cohort and location.” Economics of Education Review 21(6): 609–622.Eswaran, Mukesh. 2006. “Fertility in Developing Countries.” In UnderstandingPoverty, edited by Abhijit Banerjee, Roland Benabou, and Dilip Mookherjee,143–160. Oxford University Press, New York.Fang, Hai, Karen N Eggleston, John A Rizzo, and Richard J Zeckhauser. 2010.“Employment and Fertility in Rural China: Implications for the “One–Child”Policy.” NBER Working Paper.Fatima, Ambreen, and Humera Sultana. 2009. “Tracing out the U-shape relation-ship between female labor force participation rate and economic develop-ment for Pakistan.” International Journal of Social Economics 36 (1/2): 182–198.Felmlee, Diane H. 1993. “The Dynamic Interdependence of Women's Employ-ment and Fertility.” Social Science Research 22 (4): 333–360.Firpo, Sergio, Nicole M. Fortin, and Thomas Lemieux. 2009. “UnconditionalQuantile Regressions.” Econometrica 77, no. 3 (May): 953–973.Fortin, N. M. 2006. Greed, Altruism, and the Gender Wage Gap. Working Papers.University of British Columbia.Fujiwara, Thomas. 2014. “Voting Technology, Political Responsiveness, and In-fant Health: Evidence from Brazil.” Princeton University, mimeo.Gaddis, Isis, and Stephan Klasen. 2014. “Economic development, structural change,and women's labor force participation.” Journal of Population Economics 27(3): 639–681.113Galor, Oded, and David N Weil. 1996. “The Gender Gap, Fertility, and Growth.”American Economic Review 86 (3): 374–87.Goh, Aiting. 1999. “Trade, employment and fertility transition.” Journal of Inter-national Trade & Economic Development 8 (2): 143–184.Goldin, Claudia. 1995. “The U-Shaped Female Labor Force Function in Eco-nomic Development and Economic History.” In Investment in Women’s Hu-man Capital and Economic Development, edited by T. P. Schultz, 61–90.University of Chicago Press.Gonzalez, Maria de los Angeles. 2002. “Do Changes in Democracy Affect the Po-litical Budget Cycle? Evidence from Mexico.” Review of Development Eco-nomics 6 (2): 204–224.Grier, Kevin. 2008. “US presidential elections and real GDP growth, 1961-2004”.”Public Choice 135 (3-4): 337–352.Gupta, Monica Das, and Manju Rani. 2004. India’s Public Health System: HowWell Does it Function at the National Level? Vol. 3447. Washington, DC:World Bank Publications.Hnatkovska, Viktoria, and Amartya Lahiri. 2012. “Structural Transformation andthe Rural-Urban Divide.”Hnatkovska, Viktoria, Amartya Lahiri, and Sourabh B. Paul. 2012. “Castes andLabor Mobility.” American Economic Journal: Applied Economics 4, no. 2(April): 274–307.. 2013. “Breaking the Caste Barrier: Intergenerational Mobility in India.”Journal of Human Resources 48, no. 2 (Spring): 435–473.Jensen, Robert. 2012. “Do labor market opportunities affect young women’s workand family decisions? Experimental evidence from India.” The QuarterlyJournal of Economics 127 (2): 753–792.Jones, Gareth, Richard W Steketee, Robert E Black, Zulfiqar A Bhutta, and Saul SMorris. 2003. “How many child deaths can we prevent this year?” The lancet362 (9377): 65–71.114Khanna, Shantanu. 2012. “Gender wage discrimination in India–Glass ceiling orsticky floor?” Centre for Development Economics (CDE)Working Paper 214.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.Kingdon, Geeta Gandhi. 1998. “Does the labour market explain lower femaleschooling in India?” The Journal of Development Studies 35 (1): 39–65.Kingdon, Geeta Gandhi, and Jeemol Unni. 2001. “Education and women’s labourmarket outcomes in India.” Education Economics 9 (2): 173–195.Kitzman, Harriet, David L Olds, Charles R Henderson, Carole Hanks, RobertCole, Robert Tatelbaum, Kenneth M McConnochie, Kimberly Sidora, DennisW Luckey, David Shaver, et al. 1997. “Effect of Prenatal and Infancy HomeVisitation by Nurses on Pregnancy Outcomes, Childhood Injuries, and Re-peated Childbearing: A Randomized Controlled Trial.” Jama 278 (8): 644–652.Klein, Michael W. 1996. “Timing is All: Elections and the Duration of UnitedStates Business Cycles.” Journal of Money, Credit and Banking 28 (1): 84–101.Kolesnikova, Natalia, and Yang Liu. 2011. “The Gender Pay Gap May Be MuchSmaller Than Most Think.” The Regional Economist 19 (4): 14–15.Kottis, Athena Petraki. 1990. “Shifts over time and regional variation in women’slabor force participation rates in a developing economy: The case of Greece.”Journal of Development Economics 33 (1): 117–132.Kreyenfeld, Michaela. 2004. “Fertility Decisions in the FRG and GDR: An Anal-ysis with Data from the German Fertility and Family Survey.” DemographicResearch Special Collections 3 (11): 275–318.Lahoti, Rahul, and Hema Swaminathan. 2013. “Economic Growth and FemaleLabour Force Participation in India.” IIM Bangalore Research Paper, no.414.115Lindbeck, Assar. 1976. “Stabilization Policy in Open Economies with Endoge-nous Politicians.” American Economic Review 66 (2): 1–19.Luci, Angela. 2009. “Female labour market participation and economic growth.”International Journal of Innovation and Sustainable Development 4 (2): 97–108.Mammen, Kristin, and Christina Paxson. 2000. “Women’s Work and EconomicDevelopment.” Journal of Economic Perspectives 14 (4): 141–164.McCallum, Bennett T. 1978. “The Political Business Cycle: An Empirical Test.”Southern Economic Journal 44 (3): 504–515.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.Nordhaus, William D. 1975. “The Political Business Cycle.” The Review of Eco-nomic Studies 42 (2): 169–190.Olsen, Wendy, and Smita Mehta. 2006. “A Pluralist Account of Labour Participa-tion in India.”Pande, Rohini P, and Nan Marie Astone. 2007. “Explaining son preference in ruralIndia: the independent role of structural versus individual factors.” Popula-tion Research and Policy Review 26 (1): 1–29.Persson, Torsten, and Guido Enrico Tabellini. 1990. Macroeconomic Policy, Cred-ibility and Politics. Vol. 38. Chur, Switzerland; New York, N.Y.: HarwodAcademic Publishers.Po¨rtner, Claus Chr. 2001. “Children as insurance.” Journal of Population Eco-nomics 14 (1): 119–136.Rindfuss, Ronald R, Karen Benjamin Guzzo, and S Philip Morgan. 2003. “Thechanging institutional context of low fertility.” Population Research and Pol-icy Review 22 (5-6): 411–438.116Rogoff, Kenneth. 1990. “Equilibrium Political Budget Cycles.” American Eco-nomic Review 80 (1): 21–36.Rogoff, Kenneth, and Anne Sibert. 1988. “Elections and Macroeconomic PolicyCycles.” The Review of Economic Studies 55 (1): 1–16.Rosenzweig, Mark R, and Oded Stark. 1989. “Consumption Smoothing, Migra-tion, and Marriage: Evidence from Rural India.” Journal of Political Econ-omy 97 (4): 905–926.Ross, Michael. 2006. “Does Democracy Reduce Infant Mortality.” American Jour-nal of Political Science 50:860–874.Roy Chaudhuri, Arka. 2012. “Trade Reforms and Gender Wage Differentials inIndia.” In Conference volume of Gender Equality Rights and Trade Regimes:Coordinating Compliance conference held in Ottawa, Ontario on December2, 2010. Edited by B. Potter Pitman, Heather Gibb, and Erika Cedillo, 47–72.The North South Institute; Asia Pacific Dispute Resolution Research, Facultyof Law, University of British Columbia.Santow, Gigi, and Michael Bracher. 2001. “Deferment of the first birth and fluctu-ating fertility in Sweden.” European Journal of Population/Revue europe´ennede De´mographie 17 (4): 343–363.Schuknecht, Ludger. 1996. “Political Business Cycles and Fiscal Policies in De-veloping Countries.” kyklos 49 (2): 155–170.Schultz, TP. 1985. “Changing world prices, women’s wages, and the fertility tran-sition: Sweden, 1860-1910.” The journal of political economy 93 (6): 1126.Singh, Nirvikar. 2008. “Decentralization And Public Delivery Of Health Care Ser-vices In India.” Health Affairs 27 (4): 991–1001.Smith, James P, and Finis R Welch. 1989. “Black Economic Progress after Myrdal.”Journal of Economic Literature 27, no. 2 (June): 519–64.Smith-Lovin, Lynn, and Ann R Tickamyer. 1981. “Fertility and patterns of la-bor force participation among married women.” Biodemography and SocialBiology 28 (1-2): 81–95.117Tam, Henry. 2011. “U-shaped female labor participation with economic develop-ment: Some panel data evidence.” Economics Letters 110 (2): 140–142.Tilak, JBG. 1980. “Education and labour market discrimination.” Indian Journalof Industrial Relations 16 (1): 95–114.Topalova, Petia. 2007. “Trade liberalization, poverty and inequality: Evidencefrom Indian districts.” In Globalization and Poverty, 291–336. Universityof Chicago Press.Topalova, Petia, and Amit Khandelwal. 2011. “Trade liberalization and firm pro-ductivity: The case of india.” Review of economics and statistics 93 (3): 995–1009.Veiga, Linda G, and Francisco Jose´ Veiga. 2007. “Does opportunism pay off?”Economics Letters 96 (2): 177–182.118Appendix AAppendix to Chapter 2Data SourcesThe data used in this chapter come from a number of sources. They cover the 14main politically competitive states in India and span the period 1975-1998.The unit of observation in this chapter is an individual child. Since twinshave a significantly higher mortality rate as compared to singletons (Buekens andWilcox (1993)), I have considered only the singleton births for this chapter. Theresults are however robust to the inclusion of twins as shown in Section A.3.The infant mortality variable is defined as a dummy variable indicating whethera child died before the age of 12 months. The variable is defined for children bornbetween 1975 and 1998. The data is derived from the second round of NationalFamily Health survey (NFHS II) conducted in 1998-99.The neonatal mortality variable is defined as a dummy variable indicatingwhether the child died before the age of 1 month. The variable comes from theNFHS II and is defined for children born during 1975-1998.119The 2-12 months mortality variable is given by a dummy variable indicatingwhether the child died between 2 and 12 months of their life. This variable alsocomes from the NFHS II and is defined for children born 1975-1998.Child gender, birth order and month of birth dummies also come from NFHSII and are defined for the period 1975-1998. These variables are used as controlsin sections 2.5.1-2.5.5 and section 2.6.The number of antenatal visits gives the total number of antenatal visits madeby a mother during her pregnancy. The incidence of low birth weight is givenby a dummy variable equal to 1 if the mother reported that the child was of lowbirth weight. Whether the mother had a tetanus injection during pregnancy is alsogiven by a dummy variable indicating whether the mother had at least one tetanusinjection during pregnancy. These variables are defined for children born during1987-1992 whose mothers were surveyed in NFHS I and children born during1995-1998 whose mothers were surveyed in NFHS II.The child level and mother level controls used in the mechanisms regressionscome from NFHS I and NFHS II. The controls include child gender, birth orderdummies, month of birth dummies, a dummy for whether the mother belongs toScheduled Castes (historically disadvantaged social group in India), a dummy forwhether the mother belongs to Scheduled Tribes, a dummy indicating whether themother is Muslim and a dummy for urban area of residence. The regressions alsoinclude mothers’ and fathers’ years of education. Although the information oncontrols is available for all mothers, the sample used in this survey is restricted tomothers of children born in the period 1987-1992 (data derived from NFHS I) and1201995-1998 (data derived from NFHS II). This is because the dependent variablesin the mechanism regressions are available only for those years.The dummy variables for being born 0-1 months, 0-2 months, 0-6 months, 0-12 months and 13-24 months before and after elections are derived from officialdata of the election commission of India and the first and second rounds of NFHS.The election commission of India website gives the month and year of all electionswhich took place during the sample period. The NFHS data has information onthe month and year of birth of a child. Using these two datasets I have created thedummies for being born before and after elections. These variables are definedfor the period 1975-1998.Average district voter turnout is constructed from the constituency level elec-toral data obtained from the official website of the election commission of India.Since a constituency is smaller than a district, turnout has been aggregated to thedistrict level.Real GDP per capita and real state GDP per capita lagged by two years areobtained from the publicly available EOPP (Economic Organization and PublicPolicy Programme) database.121Appendix TablesTable A.1: Elections and Infant Mortality (Changing Omitted Category)(1) (2) (3) (4)Born 0-12 Months before Scheduled Election -1.137∗∗∗ -1.313∗∗∗ -1.606∗∗∗ -1.059∗∗(0.398) (0.432) (0.467) (0.501)Born more than 24 Months before Scheduled Election -0.220 -0.520 0.0395(0.376) (0.424) (0.456)Born 12-24 Months before Scheduled Election 0.0866 -0.390 0.160(0.390) (0.424) (0.520)Born 1-12 Months after Scheduled Election 0.371 0.198 0.469(0.436) (0.422) (0.488)Born 12-24 Months after Scheduled Election -0.203 -0.376 -0.638(0.469) (0.521) (0.494)Observations 175804 175804 175804 175804r2 0.341 0.341 0.341 0.341Notes: Standard errors in parentheses. Each column represents a separate regression. Infant mortality isthe dependant variable. The coefficients reported correspond to dummies for children born before and afterscheduled elections. The omitted category in the regression corresponding to Column 1 is a dummy indicatingchildren born more than 24 months before scheduled elections. The omitted category in the regression corre-sponding to Column 2 is a dummy indicating whether a child is born 12-24 months before scheduled election.The omitted category in the regression corresponding to Column 3 is a dummy indicating whether a child isborn 1-12 months after scheduled elections. The omitted category in the regression corresponding to Column4 is a dummy indicating whether the child is born 12-24 months after scheduled election. Apart from thereported coefficients, all regressions include mother’s fixed effect, dummies for year of birth, month of birth,order of birth, sex of the child. Other controls include average district turnout, real state domestic product percapita and month of polling dummies. Sample includes children born in the period 1975-1998 from 14 majorstates in India. Errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.01122Table A.2: Elections and Infant Mortality (With Current GDP as Control)(1) (2) (3) (4)Lagged GDP Lagged GDP Current GDP Current GDPBorn 0-12 Months before Scheduled Election -1.194∗∗∗ -1.137∗∗∗ -1.136∗∗∗ -1.034∗∗∗(0.351) (0.398) (0.349) (0.397)Born 12-24 Months before Scheduled Election 0.0866 0.227(0.390) (0.394)Born 0-12 Months after Scheduled Election 0.371 0.429(0.436) (0.434)Born 12-24 Months after Scheduled Election -0.203 -0.158(0.469) (0.467)Observations 175804 175804 175804 175804r2 0.341 0.341 0.341 0.341Notes: Standard errors in parentheses. Each column represents a separate regression. Infant mortality is the dependantvariable. While the regressions presented in columns 1 and 2 include real state GDP per capita lagged by two years as acontrol; the regressions presented in columns 3 and 4 include contemporaneous real state GDP per capita as a control.The coefficients reported correspond to dummies for children born before and after scheduled elections. In additionto the reported coefficients, all regressions include mother’s fixed effect, dummies for year of birth, month of birth,order of birth, sex of the child. Other controls include average district turnout, and month of polling dummies. Sampleincludes children born in the period 1975-1998 from 14 major states in India. Errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.01123Table A.3: Elections and Infant Mortality (Including Multiple Births)(1) (2) (3) (4)Both single andOnly single births multiple birthsBorn 0-12 Months before Scheduled Election -1.194∗∗∗ -1.137∗∗∗ -1.311∗∗∗ -1.226∗∗∗(0.351) (0.398) (0.354) (0.404)Born 12-24 Months before Scheduled Election 0.0866 0.177(0.390) (0.402)Born 1-12 Months after Scheduled Election 0.371 0.314(0.436) (0.439)Born 12-24 Months after Scheduled Election -0.203 -0.0585(0.469) (0.481)Observations 175804 175804 178169 178169r2 0.341 0.341 0.339 0.339Notes: Standard errors in parentheses. Each column represents a separate regression. Infant mortality isthe dependant variable. The sample used in regressions corresponding to columns 1 and 2 include onlythe single births. Columns 3 and 4 include multiple births in addition to single births. The coefficientsreported correspond to dummies for children born before and after scheduled elections. In addition tothe reported coefficients, regressions include mother’s fixed effect, dummies for year of birth, month ofbirth, order of birth, sex of the child. Other controls include average district turnout, real state domesticproduct per capita lagged by two years and dummies for month of polling. Errors are clustered at districtlevel.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.01124Table A.4: Elections and Infant Mortality (Including NFHS I)(1) (2) (3) (4)0-12 months 0-6 months 0-12 months 0-6 monthsbefore before before beforeBorn 0-12 Months before Scheduled Election -1.194∗∗∗ -0.873∗∗∗(0.351) (0.277)Born 0-6 Months before Scheduled Election -1.035∗∗ -0.866∗∗(0.419) (0.338)Observations 175804 175804 318535 318535r2 0.341 0.341 0.355 0.355Notes: Standard errors in parentheses. Each column represents a separate regression. Infant mortality is thedependant variable. The individual level data used for estimating regressions corresponding to columns 1 and 2come from NFHS II. The sample used in columns 2 and 4 report come from both NFHS I and NFHS II. Columns 1and 3 report coefficients on dummies indicating whether the child is born 0-12 months before scheduled election.Columns 2 and 4 report coefficients on dummies for children born 0-6 months before scheduled election. Inaddition, all regressions include mother’s fixed effect, dummies for year of birth, month of birth, order of birth,sex of the child. Other controls include average district turnout, real state domestic product per capita and monthof polling dummies. Sample includes children born in the period 1975-1998 from 14 major states. Errors areclustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.01125Table A.5: Presence of Doctor During Delivery(1) (2) (3) (4)0-6 months before 0-12 months beforeBorn 0-6 Months before Scheduled Election 0.00190 0.103∗∗(0.0299) (0.0471)Absolute Margin x Born 0-6 Months before Scheduled Election -0.203∗∗(0.0935)Born 0-12 Months before Scheduled Election -0.0194 -0.0372(0.0190) (0.0295)Absolute Margin x Born 0-12 Months before Scheduled Election 0.0461(0.0542)Absolute Margin -0.0341 -0.0358(0.0268) (0.0268)Observations 13977 13977 13977 13977r2 0.366 0.366 0.366 0.366Notes: Standard errors in parentheses. Each column represents a separate regression. The dependant variable isa dummy indicating whether a child is of low birth weight is the dependant variable. Columns 1 and 3 report thecoefficients on dummies for children born 0-6 months and 0-12 months before scheduled elections. Columns 2 and4 report the coefficients on dummy variables indicating whether a child is born 0-6 or 0-12 months before scheduledelections and interactions between the election dummy and the absolute margin of victory of the ruling party/coalitionin the previous election. Columns 2 and 4 also report the coefficient on the absolute margin of victory variable. Inaddition to the reported coefficients, all regressions include district fixed effect, dummies for year of birth, monthof birth, order of birth, sex of the child, mother’s and father’s year of schooling and dummies for rural residence,membership in SC/ST. Other controls include average district turnout, real state domestic product per capita lagged bytwo years and month of polling dummies. Sample includes children born in the period 1995-1998 from 14 major statesin India. Errors are clustered at district level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.01126Table A.6: Mechanisms (All states)(1) (2) (3) (4) (5) (6)Tetanus during Low BirthAntenatal Visits Pregnancy WeightBorn 0-6 Months before Scheduled Election 0.294∗∗∗ 0.0489∗∗∗ -0.0280(0.0843) (0.0150) (0.0181)Absolute Margin x Born 0-6 Months before Scheduled Election -0.318∗∗ -0.0746∗∗ 0.0120(0.150) (0.0292) (0.0307)Born 0-12 Months before Scheduled Election 0.176∗∗ 0.0256∗∗ -0.00733(0.0748) (0.0114) (0.0149)Absolute Margin x Born 0-12 Months before Scheduled Election -0.204 -0.0511∗∗ 0.00352(0.138) (0.0197) (0.0247)Absolute Margin -0.0464 -0.0441 0.00229 0.00314 -0.00269 -0.00201(0.0647) (0.0663) (0.0147) (0.0149) (0.0114) (0.0118)Observations 59337 59337 65660 65660 56013 56013r2 0.487 0.487 0.265 0.252 0.0413 0.0412Notes: Standard errors in parentheses. Each column represents a separate regression. The dependant variable in columns 1 and 2 is the number ofantenatal visits made by the mother during pregnancy. The dependant variable in columns 3 and 4 is a dummy indicating whether the mother hadat least one tetanus injection during pregnancy. The dependant variable in columns 5 and 6 is a dummy indicating whether a child is of low birthweight. The table reports coefficients on dummy variables indicating whether a child is born 0-6 or 0-12 months before scheduled elections andinteractions between the election dummy and the absolute margin of victory of the ruling party/coalition in the previous election. The coefficienton the absolute margin of victory variable is also reported. In addition to the reported coefficients, all regressions include district fixed effect,dummies for year of birth, month of birth, order of birth, sex of the child, mother’s and father’s year of schooling and dummies for rural residence,membership in SC/ST. Other controls include average district turnout, real state domestic product per capita lagged by two years and month ofpolling dummies. Sample includes children born in the period 1987-1992 and 1995-1998 from 14 major states in India. Errors are clustered atdistrict level.∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.01127Appendix BAppendix to Chapter 3Arriaga Method for calculating the Total FertilityRateThe data for the total fertility rate for the year 1991 has been collected from thegovernment publication on fertility and child mortality for the year 1991 (Gov-ernment of India, 1997). This report used the Arriaga method (Arriaga (1983))of fertility estimation for computing the total fertility rate. The census data pro-vides information regarding the average number of children ever born (P) in adistrict tabulated by the age-group of the mother. Arriaga suggested transformingthe recorded average number of children ever born data into estimates of age spe-cific fertility. The report used data on the average number of children ever bornby mothers 5-year age group from the census 1991 and census 1981. Next theyobtained the average number of children ever born for women at exact age x atthe time of the two censuses by fitting a ninth degree polynomial on the recordeddata. Then by linear interpolation, the information on the number of children ever128born by the exact age of the mother for one year after the earlier date (1982) andone year before the latter date (1990) is found. Next the single-year age-specificfertility rates for 1991 are calculated asfx=Px+1(1991)-Px(1990)where fx denotes the age-specific fertility rates of mothers at the age x. Theage specific fertility rates in the conventional five years groups is calculated bytaking arithmetic average of the single year age specific fertility rates within eachfive year group. The total fertility rate is then calculated by summing these agespecific fertility rates by 5 year intervals and multiplying this sum by the numberof age-groups.However for calculating the total fertility rate of the year 2001, I have usedthe data on the average children born from the 2001 census only. Since the datafor only one period is used, after fitting the polynomial to the average number ofchildren ever born, the difference between two consecutive exact ages are taken asthe single year specific fertility rates. I have next constructed the total fertility ratein the same way as discussed above. I have used the software PASEX obtainedfrom the US Census Bureau for this calculation.129

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.24.1-0300647/manifest

Comment

Related Items