Journal of Development Economics 126 (2017) 1–18

Contents lists available at ScienceDirect

Journal of Development Economics journal homepage: www.elsevier.com/locate/jdeveco

Seeing is believing- can increasing the number of female leaders reduce sex selection in rural India?

MARK

Priti Kalsi Department of Economics, Rochester Institute of Technology, 92 Lomb Memorial Drive, Rochester, NY 14623, USA

A R T I C L E I N F O

A BS T RAC T

JEL classification: J13 J16 I18 O20

Cultural values regarding gender roles encourage gender discrimination and the practice of sex selection. Increasing political and work force participation of women challenges such norms. Exploiting the implementation of an Indian law that required one-third of local political seats to be reserved for women, I investigate the impact of female leadership on sex selection in rural India. I find an increase in the survival of higher birth order girls if political seats at the local level have been reserved for women. I argue that the likely underlying mechanism is a change in beliefs due to exposure to female leaders.

Keywords: Sex selection Female politicians Gender-discrimination

1. Introduction Sex selection has been linked to skewed sex ratios in several Asian countries.1 Although the normal child sex ratio is considered to be 105 boys per 100 girls, the 2011 Indian census revealed that this ratio was 109 boys per 100 girls in India. Methods to control the sex of a child range from sex-selective abortions to infanticide and neglect which results in death of a female child (Das Gupta et al., 2003).2 The Indian government's legislative efforts to curb the practice of sex selection have been largely ineffective. Despite a 1996 ban on prenatal sex detection by ultrasound, the sex ratio at birth in India has tended to skew more male since 1980 (Arnold et al., 2002). While improved enforcement of the ban might succeed in reducing sex ratios, such a policy could have unwanted consequences. Research suggests that in countries with a preference for boys, prenatal sex selection may be a substitute for forms of postnatal discrimination such as infanticide, malnourishment, and an overall lower level of investment in female children (Kalsi, 2015; Lin et al., 2014). Given the issue of substitution across forms of discrimination, policies aimed at reducing the demand for boys by targeting the underlying son preference are preferable to policies that restrict the supply of sex selection technologies. In this paper, I explore whether improving the status of women within society could reduce son preference. I take advantage of a 1993 amendment to India's constitution, which established a three-tier system of rural governing local bodies

1 2

(Panchayats) within each state. States were required to develop Panchayats at the village, the block, and the district level. The law also required that these newly developed Panchayats reserve one-third of all political positions, including those of chairpersons, for women. As a result, a state's adoption of the law led to a drastic increase in female political representation at local levels within the state. Increased female political representation has been shown to improve the status of female constituents. For instance, female political reservations have been linked to an increased likelihood that a woman will be elected in the future (Beaman et al., 2009) and that a woman will speak at a political meeting (Beaman et al., 2010). Moreover, parents have been shown to report higher aspirations for their daughters after being exposed to female leaders (Beaman et al., 2012). Recent work has also linked reservations of political seats for women to increased reporting of crime against women (Iyer et al., 2012) and increased establishment of female-owned businesses in the informal sector (Ghani et al., 2014). I extend the literature by investigating if increased female representation improves the desirability of a daughter and thus reduces sex selection in rural India. However, improved status of females is not the only mechanism through which female leadership could alter sex ratios. Research has shown that female political involvement shifts policymaking towards increased investment in goods preferred by women (Beaman et al., 2010; Chattopadhyay and Duflo, 2004); although, this finding does not extend to South Indian states, where gender disparity is of lesser

E-mail address: [email protected]. See Anderson and Ray (2010), Das Gupta (1987), Goodkind (1996), Lin et al. (2014), and Qian (2008). Appendix B summarizes the historical context of the availability of sex-selective abortions in India.

http://dx.doi.org/10.1016/j.jdeveco.2016.12.002 Received 12 August 2014; Received in revised form 15 December 2016; Accepted 16 December 2016 Available online 21 December 2016 0304-3878/ © 2016 Elsevier B.V. All rights reserved.

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

level Panchayat decreases the high birth order ratio of boys in areas where exposure to district Panchayat leaders is arguably high, but not in the areas where the Panchayats target their goods and services. This paper begins with an overview of the policy change in Section 2.1. The timing of when states in the sample have an election is explicitly discussed, and the choice of states in the most preferred sample is explained in Section 2.2. The data sets are explained in Section 3, and Section 4 provides summary statistics. The methodology relies on variation in the timing around when a state has a relevant election reserving seats for women and the birth order of children; the corresponding estimating equation is introduced in Section 5. Section 6 presents the main results. Various robustness checks are reviewed in Section 7. Section 8 explores possible mechanisms consistent with the results. Section 9 concludes.

concern (Ban and Rao, 2008). Additional evidence indicates that health is also an area where female leaders invest more than their male peers (Bhalotra and Clots-Figueras, 2014). Taken together, these studies suggest that female leaders may invest differently in fertility and health services, which in turn could affect sex ratios. In this paper, I establish a link between female leadership and reduced sex selection in rural India. I argue that the mechanism is increased status of women and not differential investments made by women politicians. Similar to Iyer et al. (2012) and Ghani et al. (2014), the methodology used in this paper also relies on the timing of state-level implementation of female political seat reservations. Exploiting variation both in the timing of when a state had its first Panchayat elections (effectively the first time seats for women were reserved) and in the birth order of children, I estimate a difference-in-differences (DD) model. The model investigates birth order-specific effects because sex selection is known to increase disproportionately at higher birth orders.3 Prior to female seat reservations, my sample consists of 117 boys per 100 girls at birth orders 3 and greater, 107 boys per 100 girls at the second birth order, and 106 boys per 100 girls at the first birth order. I find that higher birth order children born in rural India following a Panchayat election that reserved seats for women are less likely to be male, while there is no evidence that sex ratios changed for second and first-born children. The estimates imply that post-reservations, there are between 6 to 12 fewer boys per 100 girls at birth orders 3 and greater in rural India. These effects are large and imply an increased survival of approximately 900,000 to 1,800,000 high birth order girls over the course of the study. By investigating child death rates, I test whether the underlying mechanism is that changes in health services disproportionately favored highest birth order girls. I do not find that high birth order girls, amongst children who have survived infancy, have different survival rates during early childhood than girls in general, suggesting that health services remained constant across birth orders. I also test the claim that female leaders directly reduced sex selection fertility services by investigating the impact of female reservations on sex ratios across different socioeconomic statuses. Studies have shown that the use of sex selective abortions is most common amongst more educated women, and less educated women do not appear to rely on abortions as a method of sex selection (Jha et al., 2011; Pörtner, 2016). Thus, restriction of fertility services should more directly impact more educated women. I find that the effects in this paper are strongest for poor and illiterate women, suggesting that the restriction of fertility services likely does not explain the results. Rather, because the least educated women do not rely on abortion as a means to sex select, the reduction in sex ratios documented in this paper is likely driven by increased survival of high birth order girls. While the main analysis in the paper exploits state-level variation in the timing associated with an increased total level of female leadership within a state, I also investigate if female headship at the district-level Panchayat explains the reduction in high birth order sex ratios in rural areas. I find that the effect is not explained by female headship at the district-level Panchayat, suggesting that female leadership at lowerlevel Panchayats (the village or the block level) is likely driving the results. Surprisingly, however, I also find that female headship at the district level reduces high birth order sex ratios in the urban population within the district. This is unexpected as the amendment affects locallevel government for rural, and not urban, India. I argue that there are reasons to believe that exposure to district-level women leaders is more prominent in urban towns due to the location of district-level offices and media exposure. Consistent with exposure to women leaders being the mechanism, I find that seat reservations for women at the district-

2. Background 2.1. Historical context of the 73rd amendment Prior to the adoption of the 73rd and the 74th Amendments, states were the smallest units of government recognized by the Indian constitution.4 The 73rd Amendment decentralized government in rural India, while the 74th Amendment did so in urban India. The amendments were a response to a public debate over the national government's failure to deliver public services, infrastructure, and alleviate poverty. There was a general consensus among politicians that devolving powers to the local level was the solution (Chaudhuri, 2003). Earlier incarnations of the legislation were defeated because the states did not have enough discretion in the implementation of the bill (Chaudhuri, 2003). Allowing states flexibility in design and implementation, the 73rd and the 74th Amendments were re-introduced in the Parliament and were eventually passed in December of 1992. The 73rd Amendment went into effect in April of 1993, whereas the 74th Amendment went into effect in June of 1993 (Chaudhuri, 2003). This paper exploits the implementation of the 73rd Amendment, and hence focuses on rural India. The 73rd Amendment established a pyramid structure for a threetiered local rural government, with the village-level Panchayat at the base. The Gram Sabha, or the people, elect members of the village Panchayat and also help to hold elected members accountable and ensure funds are being used properly. The block-level Panchayats are next up in the hierarchy of local Panchayats, and they provide the link between the village and the district-level Panchayats. The district Panchayat provides a direct link between the state and local governments. A graphic representation of the structure of local government in rural India is shown in Fig. A1 of the Appendix. The task of the new local government bodies was to implement development plans based on local needs of rural areas. Responsibilities included land improvement, infrastructural and ecological development, poverty alleviation, and development of women, children, and individuals of historically disadvantaged castes. The 73rd Amendment included both mandatory and discretionary provisions. The mandatory provisions called for the establishment of local Panchayats at the village, the block, and the district levels. Further requirements were direct elections, mandatory every 5 years, for Panchayats at all three levels. One-third of total seats, including those of chairpersons, were to be reserved for women at Panchayats at all three levels. Note that the law did not require reservations at any state level branch of government, but only at the three tiers of the Panchayat Raj Institutions within the state. However, states had elections that reserved seats for women at all three levels at the same time. Thus, timing of state-level election refers to the timing of increased female political representation at the three Panchayat levels within the state.

3 Chu (2001), Ebenstein (2007), Lin et al. (2014), and Kalsi (2015) also report that sex selection is greatest for higher birth orders. Das Gupta (1987) provides evidence that higher birth order girls are more discriminated against in rural Punjab, India.

4

2

This section draws heavily from Chaudhuri (2003) and Vyasulu and Vyasulu (1999).

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

case the timing of elections will be correlated with male preference. Since the law required that states pass conformity laws in line with the 73rd Amendment within a year and also required that the states complete decisions on their new Panchayats within two years, states that had an election by 1995 were abiding by the law. Because the issue of endogenous timing of election is a smaller concern for states that complied with the law, the preferred sample consists of states that had their first election by 1995, or approximately within two years following the law change. This effectively reduces the analysis to 12 states in total. A map of these states is shown in Fig. A2 of the Appendix. A few states were quite progressive in their adoption of the 73rd Amendment and reserved seats prior to its coming into force. The state of Orissa held village and block-level Panchayat elections in line with the 73rd Amendment between May and June of 1992, and the state of Maharashtra did so at the district and the village level in November of 1992. The eagerness of Orissa and Maharashtra to adopt the law prior to its coming into force could be correlated with state unobservables which are related to the states' male preference. For example, one could make the case that the states moving towards reservations of seats for women prior to the amendment is an indication that these states were becoming more accepting of females before the law change. Thus, Orissa and Maharashtra should arguably be excluded from the preferred sample. Additionally, the state of Haryana endeavored to reduce sex selection by offering a financial incentive to eligible parents with daughters in 1994.6 This policy was introduced in the state in October of 1994, while the state had its first election which reserved seats for women in June of 1994. With the coinciding timing of these two events, and with treatment determined by timing, it is difficult to distinguish which policy underlies the effect for the state of Haryana. Thus, I also exclude Haryana from the preferred sample. Limiting the states to those that had their first election by 1995, while excluding Orissa, Maharashtra, and Haryana reduces the analysis to 9 Indian states. Although the preferred sample consists of only 9 states, according to the 2011 Indian census, nearly 56 percent of the nation's population resided in these states. The location of these 9 states within India is shown in Fig. A3 of the Appendix. As male preference in India varies by region (e.g. South Indian states do not exhibit a strong male preference), it is reassuring to find that the preffered sample has states from the northern, eastern, and southern regions of India. Because of the spatial variation in the location of states in the preferred sample, the results for the preferred sample are arguably representative of India as a whole. The results in this paper are shown for both a sample of all Indian states and for a sample of the 9 law-abiding states in Fig. A3 of the Appendix. The findings for both samples are analogous, with results in law-abiding states being generally larger and more robust.

Table 1 Panchayat election dates. Post-Reservation

State Orissa Maharashtra West Bengal Karnataka Haryana Madhya Pradesh Tripura Rajasthan Andhra Pradesh Uttar Pradesh Gujarat Kerala Tamil Nadu Goa Manipur Punjab Assam Bihar

Pre-Amendment

Share of Boys 1992

Election 1

Election 2

Election

Month May Nov May Dec June June August March March April June Sept Oct Jan Jan June Nov April

Year 1997 1997 1998 2000 2000 2000 1999 2000 2001 2000 2002 2000 2001 2002 2002 2003 2007 2006

Year 1992 1992 1988

Year 1992 1992 1993 1993 1994 1994 1994 1995 1995 1995 1995 1995 1996 1997 1997 1998 2000 2001

1991 1982 1988 1970 1975 1986 1991 1978 1993 1992 1978

0.486 0.509 0.544 0.518 0.519 0.521 0.547 0.537 0.504 0.541 0.532 0.525 0.520 0.477 0.515 0.537 0.581 0.523

Share of boys in 1992 calculated from DLHS II data.

The states were given one year to pass conformity acts by either amending existing laws or by passing entirely new laws in line with the 73rd Amendment (Chaudhuri, 2003). Additionally, states were given two years to complete decisions on the new Panchayats and failing to do so posed the risk of losing the central government's assistance (Jain, 1996). Following the passage of the Amendments, a large number of women were suddenly brought into politics, and most women were somewhat unprepared for leadership (Vyasulu and Vyasulu, 1999). The first round of elections between 1993 and 1994 brought in about 800,000 women to work for local governments in a nation where there has initially been very little female involvement (Vyasulu and Vyasulu, 1999). Through their interactions with the local people as leaders, women serving as Panchayat members have reported enhancement in their status amongst the community (Jayal, 2006). Even in states with high levels of patriarchy, women reported status gains. For example, 72 percent of female members of Panchayats in Madhya Pradesh and 90 percent of female members in Rajasthan reported an increase in status (Jayal, 2006). 2.2. State elections India currently has 29 states, of which four were carved out from existing states after the year 2000. Table 1 provides a list of relevant elections for the 18 Indian states for which I was able to collect data. According to the 2011 census, 90.42 percent of the nation's population lives in one of these 18 states. For each state, I report the month and year of the first election reserving seats for women. Also reported is the year the state had a second election. When possible, I report the last Panchayat election the state had prior to the constitutional establishment of the Panchayats.5 Additionally, I report the state's share of boys born in 1992, a year prior to the law change. Table 1 shows that there is a great degree of variation in the timing of first elections. This could present potential issues of endogeneity in an analysis which relies on time variation in state elections. For example, it is possible that states with delayed elections were simply buying time before they had to adapt to female leadership, in which

3. Data The main analysis uses data from the women's questionnaire from the second round of the District-Level Health and Facility Survey (DLHS). The DLHS data are available for purchase from the International Institute for Population Sciences (IIPS) in Mumbai, India. The survey was completed between 2002 and 2004 and covers all of the 593 districts covered in the 2001 Indian census. One thousand representative households in each Indian district were surveyed. In total, there are 620,107 households in the data set, of which 415,135 are in rural areas. The women's questionnaire surveyed married women between the ages 14 and 44 years. These data include detailed information on a woman's fertility history. For each preg6 Haryana is one of the wealthiest states in India and has one of the most distorted sex ratios in the nation. To reduce male preference, the state introduced the Apni Beti Apna Dhan (ABAD) program in October of 1994. The program provided families a monetary award within 15 days of a birth of a daughter, and each daughter was also endowed with an additional reward redeemable at the age of 18 (Sinha and Yoong, 2009).

5 Local Panchayats operated prior to the establishment of the 73rd Amendment in some states, although enshrining them in the constitution required a three tier system of Panchayats, regularity in elections, and reservation of seats for women.

3

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

reservations. It is useful to compare these changes in highest birth order sex ratios with changes that occur for first-borns. Since sex selection in India does not occur for first births (Jha et al., 2011; Pörtner, 2016), a comparison of the changes in sex ratios between first birth and higher birth orders can be used to control for changes in gender-specific survival rates which are not explained by sex selection. For both samples, there is no statistically significant change in sex ratios amongst first or second-born children. Comparing the changes in sex ratios at the highest parity to the changes in sex ratios at the first parity after the policy change yields a difference-in-differences estimate of the policy effect. The difference-in-differences estimate of means reveals that high birth order children are 1.1 percentage points less likely to be male in all states and 2.4 percentage points less likely to be male in law-abiding states after political reservations for women are made. These estimates are statistically significant at the one percent level. Table 3 also shows similar difference-in-differences estimates for birth order 2 children. The difference-in-differences estimates for second-borns are small, negative, and statistically indistinguishable from zero for both samples. Finally, Table 3 shows the total change in sex ratios for all children over this time period. For all states, the total ratio of boys fell from 0.527 to 0.521 after the law change and this decline is statistically significant at the one percent level. For lawabiding states, the total ratio of boys fell from 0.528 to 0.521 and this decline is also statistically significant at the one percent level. Table 3 provides evidence that the share of boys declined for highest birth order children born after reservations for women were made. The findings of Table 3 are provided in the spirit of the main methodologies, which I discuss next.

nancy, the data provide information on the child's date of birth, birth order, sex, and whether the child is still alive. The data also provides information on household and mother's characteristics such as the type of house, religion, education level of both the husband and the wife, and the mother's age at the time of child's birth. The finest level of location identified in the DLHS is the district, and household's village or block information is not included. In addition to the fertility survey, I collected data on the time the new policy of reserving political seats for women became effective in each state. These data are mostly collected from a textbook titled Status of Panchayati Raj in the States and the Union Territories of India 2000 (Mathew, 2000), but I also relied on states' Panchayat websites to provide additional information. These dates are shown in Table 1. A key feature of the 73rd Amendment is that female seat reservations were assigned at random in most states. Since fertility data from the DLHS provide district identifiers, I also make use of data on district Panchayat reservations of chairpersons to study whether reserving a seat for a woman alters sex selection in that particular district. Data on district-level chairperson reservation are obtained from Iyer et al. (2012). 4. Summary statistics Table 2 provides summary statistics for mothers in both the sample of all states and the sample of law-abiding states using data from DLHS II. I split the table between mothers with at least one child who is born after their state reserved seats for women and mothers with no children born following reservations. The summary statistics across both samples of states are similar. In general, mothers with at least one child born after their state reserved female political seats are younger, less likely to be able to read, and less likely to come from houses made of “strong construction”. They also have given birth to more children and have had fewer children die. For the complete sample, mothers with at least one child born after reservations have fewer boys and more girls ever born to them. For the sample of law-abiding states, mothers with at least one child born after reservations have no difference in the number of boys born, but they have significantly more girls born to them. Table 3 provides summary statistics in line with the main estimating equation. It shows changes in birth order-specific mean ratio of boys amongst surviving children after the law change for all of India and for the preferred sample of states that adopted the policy in a timely manner. The table shows that mean sex ratios decreased by 1.2 percentage points for highest birth order children in all states and by 1.9 percentage points in law-abiding states after female political seat

5. Methodologies The main specification is a difference-in-differences (DD) model described in Eq. (1). The sample consists of surviving children from rural India. Because the sample is restricted to children who have survived, changes in high birth order sex ratios captures both prenatal and postnatal forms of sex selection. Focusing on surviving children is also useful because underreporting of dead children is a known issue in survey data. “Forgetting” of dead children is particularly worse for girls in sex-selective cultures (Rose, 1999). To study if sex-selective abortions, specifically, are important in explaining the results, the sample should include all children to observe changes in sex ratios at birth. However due to systematic “forgetting” of dead children, a sample of all reported children is biased. I include results with the inclusion of all children, while attempting to correct for recall error, in Appendix B of the paper. It does not appear that a reduction in sex-selective abortions is responsible for the results in this paper. Studying changes in sex ratios amongst surviving children allows for the effects to be driven by postnatal forms of sex-selection. In the model, the first difference is across time, looking before and after states have elections that reserve seats for women, and the second difference is across birth order of child.

Table 2 Summary statistics for mothers in the sample. All States

Law-Abiding States

At least 1 child born after reservations?

No

Yes

Diff

No

Yes

Diff

Age Mother is Literate Number of Children Dead Number of Children Born Total Boys Born Total Girls Born Have a Strong House Age at First Birth Observations

31.32 0.46 1.84

25.95 0.44 1.79

−5.36*** −0.03*** −0.05***

32.98 0.48 1.87

25.95 0.40 1.78

−7.03*** −0.08*** −0.09***

2.60 1.48 1.12 0.23 18.91 29,888

2.67 1.36 1.31 0.19 18.50 146,384

0.06*** −0.12*** 0.18*** −0.04*** −0.41***

2.34 1.39 0.96 0.33 18.81 9497

2.70 1.39 1.32 0.21 18.37 89,530

Boyics = β1 I (Order ≥ 3)i × Post Reservecs + β2 I (Order = 2)i × Post Reservecs + β3 Post Reservecs + β4 I (Order ≥ 3)i + β5 I (Order = 2)i + γs + θc + ΓXics + ϵics

0.36***

(1)

The dependent variable is an indicator variable for whether child i, of birth cohort c, born in state s, is a boy. Post Reservecs is an indicator variable for whether the child is born after his/her state had its first Panchayat election following the Amendment, or equivalently after the first time the child's state reserved seats for female leaders. To increase the precision of how the treatment is defined, I use both month and year variation to code birth and post-reservation dates. Thus, all children born after the month their state reserved seats are coded as being born after reservations. This definition of the treatment variable

0.00 0.36*** −0.13*** −0.44***

Sample weights used. Sample restricted to mothers with children born between 1987 and 2004 in rural areas. Law-abiding states are those that reserved seats by 1995 excluding Haryana, Maharashtra, and Orissa. ** p < 0.05, * p < 0.1. *** p < 0.01.

4

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

Table 3 Birth Order-Specific Differences in Mean Ratio of Boys. All States

Order ≥ 3 Order=1 Difference Order=2 Order=1 Difference All Children

Law-Abiding States

Pre-Reserve

Post-Reserve

Difference

Pre-Reserve

Post-Reserve

Difference

0.538 0.514 0.023*** 0.523 0.514 0.009** 0.527

0.526 0.513 0.012*** 0.520 0.513 0.007** 0.521

−0.012*** −0.001 −0.011** −0.003 −0.001 −0.002 −0.006***

0.544 0.511 0.033*** 0.523 0.511 0.012** 0.528

0.525 0.517 0.008** 0.521 0.517 0.004 0.521

−0.019*** 0.005 −0.024*** −0.002 0.005 −0.008 −0.006***

Sample weights used. Sample restricted to surviving children born between 1987–2004 in rural areas. Law-abiding states are those that reserved seats by 1995 excluding Haryana, Maharashtra, and Orissa. * p < 0.1. *** p < 0.01. ** p < 0.05.

is more relevant for postnatal forms of sex selection, which is more likely the form of sex selection for the rural population investigated in this paper.7 As families tend to sex select the most at the highest birth orders, I interact the effect of being born after reservations were adopted, Post Reservecs, with indicators for both birth order 2 and birth order 3 or greater separately. Although my results indicate that most sex selection in India occurs at birth orders 3 or greater, by including separate effects for birth order 2, I am able to explore any changes that may occur at the second birth order as well. Children of birth order 1 are the omitted category, and β1 and β2 are the parameters of interest. They capture the additional change in the likelihood that a child born at a high birth order after political seats are reserved for women is a boy with a change in the likelihood that a first-born child is a boy after increased female political involvement.8 Also included are fixed effects for birth orders 3 or greater and birth order 2. The regression includes state fixed effects, γs, to control for state-specific differences in the ratio of boys. Fixed effects for birth year of the child, θc, are included to help control for annual trends in child sex ratios in rural India.9,10 I also control for factors that could affect a mother's fertility and her son preference, such as age at the time of the child's birth, literacy, and religion in Xics. The type of house the family resides in (whether the construction of the house is considered weak, semi-strong, or strong) is included as a proxy for household income. Since female political involvement at the state level outside the 73rd Amendment could also impact sex selection, I control for whether the child was born at a time when his/her state's head of government, the chief minister, was a female. Standard errors are clustered at the state. The number of clusters in the preferred sample of law-abiding states is small (9 states) and clustering underestimates standard errors when there are such few groups (Cameron et al., 2008). To deal with this issue, I also present p-values testing the null hypothesis that the true effect is zero using wild bootstrap-t methods discussed in Cameron et al. (2008). The validity of a DD design assumes that once all other variables are controlled for, trends in high and low birth order sex ratios are identical prior to the law change and that they would remain identical in absence of the law change. Table 4 presents evidence for the equal trends assumption for a sample of children born prior to the law

Table 4 Test for pre-trends: birth order-specific differences in mean ratio of boys prior to reservations. All States

Order ≥ 3 Order=1 Diff Order=2 Order=1 Diff All Children

Law Abiding States

Early-Pre

Late-Pre

Diff

Early-Pre

Late-Pre

Diff

0.543 0.515 0.029*** 0.523 0.515 0.008 0.527

0.539 0.515 0.025*** 0.525 0.515 0.010* 0.527

−0.004 0.000 −0.004 0.002 0.000 0.002 0.00

0.545 0.509 0.036*** 0.523 0.509 0.014* 0.526

0.548 0.514 0.034*** 0.527 0.514 0.013* 0.531

0.003 0.005 −0.002 0.004 0.005 −0.001 0.005

Sample weights used. Early-pre-period is cohorts born between 1987-1989 and late-preperiod is cohorts born between 1990–1992. Law-abiding states are those that reserved seats by 1995 excluding Haryana, Maharashtra, and Orissa. ** p < 0.05. *** p < 0.01. * p < 0.1.

change. I split the pre-reservations period used in this paper into two separate time periods, and I refer to the earlier half of this period consisting of years 1987 till 1989 as the “early-pre” period and the later half of this period from 1990 till 1992 as the “late-pre” period. The table provides results in a similar fashion as Table 3, but investigates whether child sex ratios followed a similar trajectory across birth orders before the law change. It can be seen that prior to reservations, child sex ratios remained unchanged across all birth orders in both the full sample of all states and the sample of law-abiding states. All differences in sex ratios in the “late-pre” and the “early-pre” period are small and statistically insignificant. It is thus not surprising that total sex ratios also remained unchanged. Overall, Table 4 supports the assumption that sex ratios followed a similar trend across all birth orders prior to political reservations for women. Figs. 1 and 2 further support the equal trends assumption for a sample of all states and a sample of law-abiding states respectively. These figures graph the mean ratio of boys by birth order within a 16 year window around the timing of reservations of seats for women. They also present visual evidence of high birth order sex selection in rural India as higher birth order children have the higher ratio of boys prior to reservations. It can be seen that birth order-specific trends in the ratio of boys remained relatively steady across each birth order before states began reserving seats. This provides visual evidence for the equal trends assumption necessary for the validity of a DD model. The figures also present visual evidence for the effects reported in this paper: The ratio of boys fell for later-born children once political seats for women were reserved, while sex ratios did not change for first and second-born children. The trends in sex ratios also show a relatively persistent effect as the ratio of highest birth order boys declined a year after reservations for women were made, and that decline in sex ratios

7 Since the first group of women to change their decision to abort would have to be in the first trimester of their pregnancy when female leadership increases, treatment should be defined as 5 to 6 months after reservations if sex-selective abortions are important. 8 As sex selection does not generally occur at birth order 1, an argument for comparing birth order 2 to birth order 3 and greater can be made to better capture trends in sex selection. Results in this analysis are robust to comparing birth order 2 children to birth order 3 or greater children. 9 Results are also robust to including state-specific linear time trends. 10 An argument can be made to also include birth order-specific time, or birth ordertime-state trends. Only results for law-abiding states are robust to adding a birth order×birth year or a birth order×year×state time trend in Eq. (1).

5

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

Fig. 1. Figure shows birth order-specific ratio of boys for cohorts born different # of years since first reservation of seats for women. Sample restricted to live children in rural areas in a sample of all states.

Fig. 2. Figure shows birth order-specific ratio of boys for cohorts born different # of years since first reservation of seats for women. Sample restricted to live children in rural areas of law-abiding states that reserved seats by 1995, excluding Haryana, Orissa, and Maharashtra.

6. Results

remained relatively steady at least up until 8 years following the time seats for women were first reserved. Figs. 1 and 2 are reassuring and show that the timing of states' decision to reserve seats for women is associated with a decline in ratio of boys at birth orders at which sex selection is most prevalent. However these mean trends in sex ratios fail to account for many different individual characteristics that could also affect sex-specific child survival. The next section shows that the findings are robust to controlling for key variables described in Eq. (1) such as birth year trends and other variables that might determine an individual's propensity to sex select.

Table 5 presents the results from estimating Eq. (1). Column 1 reports the results from estimating the equation for all states. For all states, a child born at the highest birth order after a state reserved seats for women is about 1.28 percentage points less likely to be a boy, while highest birth order children born before the law change were 1.85 percentage points more likely to be a boy. This decline in high birth order sex ratio for the entire sample is statistically significant at the five percent level. The DD estimate of Eq. (1) for high birth order children is comparable to the DD estimate of a 1.1 percentage points decline using differences in raw means shown in Table 3. Column 2 limits the 6

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

Table 5 Birth order-specific change in share of boys.

Variables

I (Order ≥ 3) × Post Reserve I (Order = 2) × Post Reserve

Post Reserve

I (Order ≥ 3) I (Order = 2) Net Change in Ratio of Boys Reservations by 1995? Orissa, Maharashtra & Haryana Included? West Bengal included? Birth Years N States Wild Bootstrap-t p-value on I (Order ≥ 3) × Post Observations

(1) boy

(2) boy

(3) boy

(4) boy

(5) boy

−0.0128** (0.00589) −0.00183 (0.00556) 0.00810 (0.00787) 0.0185*** (0.00406) 0.00709* (0.00389) 0.00164 (0.00561) N Y Y 1987–2004 18 0.066

−0.0188** (0.00648) −0.00267 (0.00627) 0.0109 (0.0124) 0.0222*** (0.00441) 0.00660 (0.00459) 0.00171 (0.00916) Y Y Y 1987–2004 12 0.046

−0.0263*** (0.00499) −0.00783 (0.00699) 0.0218 (0.0160) 0.0257*** (0.00359) 0.00951* (0.00445) 0.00790 (0.0128) Y N Y 1987–2004 9 0.0

−0.0383*** (0.00962) −0.0189 (0.0199) 0.0304 (0.0197) 0.0254** (0.00779) 0.0111 (0.00682) 0.00842 (0.0125) Y N Y 1991–1995 9 .016

−0.0223*** (0.00333) −0.00352 (0.00592) 0.00945 (0.0161) 0.0234*** (0.00330) 0.00959 (0.00507) −0.00152 (0.0141) Y N N 1987–2004 8 0.014

531,849

391,686

302,626

89,772

284,112

Sample weights used. State clustered standard errors reported. Sample restricted to surviving children of rural areas. All specifications include birth year, state, mother's age at time of birth, mother's literacy, mother's religion, gender of the state's chief minister at the time of child's birth, and type of house fixed effects. Net Effect estimated using identical specifications of Eq. (1), while dropping birth order interaction terms. P-value reported for I (Order ≥ 3) × Post Reserve using wild bootstrap-t methods as discussed in Cameron et al. (2008) with 1000 repetitions. *** p < 0.01. ** p < 0.05. * p < 0.1.

statistically insignificant, is large and positive. The interpretation of the results changes if families are increasing sex selection for first-borns and thus can afford to reduce sex selection for later births. While Pörtner (2016) has shown that sex selection in India has been moving to lower birth orders as household's desire lower fertility, Pörtner (2016) also shows that this behavior has not moved to the first birth order.12 Nonetheless, the large and positive coefficient estimates of Post Reserve imply that the total effect of the policy is a small positive and statistically insignificant effect. Table 5 also shows the total effect for each respective column's sample. The total effect of the policy is estimated using an equation similar to Eq. (1), but dropping the birth order interaction terms with Post Reserve. The positive estimate seen for first-borns counters the decline seen at the highest birth order and as a result, the policy has a zero change in the total ratio of boys. Even though I am unable to reject the claim that total sex ratios remained unchanged after the policy, negative and statistically significant effects seen for high birth orders suggest, that without the policy, sex ratios for the later-born children, and thus also total sex ratios, would have been higher. There is also some indication that the single state of West Bengal is determining the large coefficient magnitudes of the effect for firstborns. The state of West Bengal has the fourth largest population in India and is one of the earliest states, and the first in the preferred sample, to reserve seats for women. Hence, the state is important in the estimation of the main effects. When investigated alone, the state shows a large (3.12 percentage points), statistically significant increase in sex ratios at the first parity and a decrease in sex ratios at the highest parity. Table 8 shows these results. Within West Bengal, first parity sex ratio pre-reservation is relatively low at 0.50, which is below the

sample to states that had their first election by 1995 as the law required. For these states, children born at the highest birth order prior to the law change are 2.22 percentage points more likely to be a boy in comparison to first birth order children. The likelihood that the child is a boy at the highest birth order declines by 1.88 percentage points if the child is born after the state had elections that reserved seats for women in compliance with the 73rd Amendment. As mentioned above, Orissa and Maharashtra's adopting the law prior to its coming into force could be an indication that the states were already on a trend towards becoming more acceptable towards female leaders, or females in general. Also the state of Haryana executed a different policy targeting sex ratios around the same time as its Panchayat election, and thus should also be excluded from the preferred sample. Column 3 presents the results from the preferred sample of law-abiding states while excluding Orissa, Haryana, and Maharashtra. This leads to a larger reduction in high birth order sex selection (2.63 percentage points).11 These results are also comparable to the DD estimate of a 2.4 percentage points decline in high birth order sex ratios estimated using raw means for law-abiding states shown in Table 3. Column 4 further restricts the sample and compares changes in sex ratios for children born right around the law change, or those born between 1991 and 1995. When looking for an effect right around the law change, I find that high birth order children are 3.91 percentage points less likely to be a boy. Finding a similar effect in a such a narrow window is reassuring as it can be more plausibly argued that changes in unobservables following reservations for women are not explaining the effect. Although previous literature finds that the Indian ban on ultrasound did not deter prenatal sex selection (Arnold et al., 2002; Visaria, 2007), finding an effect within the birth year window of 1991 and 1995 is also evidence that the 1996 ban on ultrasound does not explain the effect. It is worth noting that the coefficient on Post Reserve, although

12 I do not find evidence that the timing of elections is associated with reduced fertility in states that reserved seats for women. To check this, I created a woman-year panel dataset and estimated if the timing of reservations is associated with a decrease in the likelihood that a women observed a birth, while controlling for all of the additional controls of specification (1). My estimates find statistically insignificant effects of state election timing on the likelihood of a birth for both sample of all states and a sample of law-abiding states.

11 Since the preferred sample consists of only 9 states, it is important to check that a single state is not driving the results. Omitting any one of the 9 states from the preferred sample yields similar results.

7

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

not been implemented. This produces an estimate of between 906,508 and 1,813,015 girls who otherwise would have been sex selected over this time period, but were not. In the next section, I show that the results of this section are robust to several tests.

normal sex ratios of approximately 0.512. Although the state shows a steep increase in sex ratios at the first parity, it is possible that first birth order sex ratios in the state are reverting back to a normal range following reservations in the state, and this gets picked up as a positive effect. On the other hand, it is also possible that first-born sex selection increased in the state. Again, this seems unlikely since earlier birth order sex selection is more consistent with urban and not rural populations studied in this paper. For comparison, I include the estimation result for the preferred sample with the state of West Bengal excluded in Column 5 of Table 5. When the state of West Bengal is excluded in Column 5, the coefficient estimate on Post Reserve is small (0.009) and statistically insignificant while the main effect for high birth orders is largely unchanged.13 In Columns 3 and 4, the number of states is reduced to 9 states, and this number is further reduced to just 8 states in Column 5. Bertrand et al. (2004) shows that clustering yields over-rejection of the null hypothesis of no effect when the number of clusters falls below 10. Cameron et al. (2008) proposes a solution and presents wild bootstrapt techniques for when the number of clusters is small. I present the pvalue for the coefficient of interest using the wild bootstrap-t techniques of Cameron et al. (2008). This method rejects the null of a zero effect at the one, the five, and the five percent level in Columns 3, 4, and 5 respectively, when the number of clusters are greatly reduced.14

7. Robustness checks 7.1. Sex-selective deaths Sex selection in rural India often occurs after birth and prenatal choices such as sex-selective abortions are less common (Das Gupta et al., 2003). Additionally, I fail to find direct evidence of a reduction in sex-selective abortions in results presented in Appendix B of the paper. Here, I investigate whether changes in sex ratios are explained by a reduction in female infant mortality rates. Since sex selection is worse at higher birth orders, I expect that high birth order female infant mortality rates would decrease if postnatal sex selection decreased after states reserve political seats for women. As mentioned earlier, a mother's memory of a child's death may be biased if she values male children more than female children, which would result in underreporting of female deaths. Recall bias, though, is known to be greater for events that happened a long time ago. In the results presented in Appendix B, I find that mothers who have been married for 23 or fewer years do not exhibit strong recall bias. Thus, I limit the sample in this section to mothers who have been married for at most 23 years.16 Eq. (2) studies the impact of Panchayat elections following the 73rd Amendment on sex and birth order-specific reported death by age one rates. I estimate a difference-in-difference-in-differences (DDD) model with the three differences being across birth order, whether the child is born before or after reservations are made, and the child's sex. Diedisc is a dummy variable indicating that child i, of birth cohort c, in state s, has died before age one. It is regressed on the interaction of whether the child is born after the state had a relevant election with a dummy variable for whether the child is a girl and also interacted with whether the child is born at a high birth order. Additional controls are as described in Eq. (1). The omitted category in this specification is firstborn boys.

6.1. Number of girls saved I find that high birth order sex ratios in rural India improved by between 1.3 and 2.6 percentage points after reservations of political seats for women. These estimates are large and indicate that between 906,508 and 1,813,015 additional girls would not have survived over the course of the study, had female leadership not increased in rural India. To calculate these values, I first estimate the number of rural children born at birth orders 3 or greater after female political seats for women are reserved in each state. Using rural India's birth rates from year 1992 to 2004 and each state's rural population in 1991, I estimate the number of children born in each year a state reserved seats for the full year.15 Indian census data provide state-wise rural population estimates for the three census years: 1991, 2001, and 2011. Children born after reservations in my sample are born between 1992 and 2004, and the use of population estimates from either the 1991 census or the 2001 census could be justified. I use data from the 1991 census because population is lower in 1991, and thus, these estimates provide a lower bound on the true number of girls saved. I find, that during the time period of this study, a total of 162,165,943 children are born in rural areas after states had reserved seats for women. According to the Indian census, rural fertility rate over this time period is approximately 3.5 children per woman. This suggests that approximately 43% of the children born in rural India are of birth orders 3 or greater. My sample of rural children born after reservations also finds that 44% of them are born at third or later birth orders. Thus, one can infer that 69,731,355 of the 162,165,943 children born after states reserved seats for women over the time period of the study are high birth order children. My estimates suggest that the ratio of high birth order girls is between 1.3 to 2.6 percentage points higher than it would have been, had the policy

Diedics = β1 I (Order ≥ 3)i × Postcs × Girli + β2 I (Order = 2)i × Postcs × Girli + β3 Postcs × Girli + β4 I (Order ≥ 3)i × Postcs + β5 I (Order = 2)i × Postcs + β6 I (Order ≥ 3)i × Girli + β7 I (Order = 2)i × Girli + β8 Postcs + β9 Girli + β10 I (Order ≥ 3)i + β11 I (Order = 2)i + γs + θc + ΓXics + ϵics (2) Table 6 presents the results from estimating Eq. (2). I find that female political seat reservations reduce high birth order female infant mortality rates by 0.71 percentage points and 1.2 percentage points in samples of all states and law-abiding states respectively. Results in Table 6 also show that female infants in general are less likely to be reported as dead, as the coefficient on Girl is negative and statistically significant. High birth order girls, however, are significantly more likely to have died by age one. While one may suspect that this is just a fertility effect and that high birth order children come from larger families so they are more likely to die, the same is not true for boys. Since first birth order boys are the omitted category, the estimated coefficient on I (Order ≥ 3) and I (Order = 2) provide the differential in death by age one rates for high birth order boys in comparison to boys born at birth order 1. Negative and statistically significant coefficients on high birth order fixed effects indicate that high birth order boys are less likely to be reported dead by age one than first-born boys. Both

13 When West Bengal is excluded from the samples in Columns 1, 2, 3, and 4, the estimate on Post Reserve is 0.003, 0.003, 0.009, and 0.014 respectively. The main effects of female political reservations on changes in high birth order sex ratios are similar to those with the inclusion of West Bengal, except that the effect for a sample of all states is no longer statistically significant. 14 All results in Table 5 are statistically significant at the one percent level if clustering is done at the district or state-birth-year level. Smaller standard errors are expected, since both ways of clustering are known to over-reject the null hypothesis of a zero effect in a DD model in which treatment varies at the state level (Bertrand et al., 2004). 15 Data on rural birth rates from 1992–2003 are from Thukral (2006). Since Thukral (2006) does not include rural birth rate for 2004, I assign 2004 the birth rate of 2003.

16 The results in Appendix B suggest that in the full sample of states, the cut-off point to correct for recall error is women married for 18 or fewer years. The results in Table 6 are robust to using the more restrictive cut-off point.

8

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

Table 6 Gender-specific change in death by age one rates.

Variables

I (Order ≥ 3) × Post × Girl I (Order = 2) × Post × Girl Post×Girl Post

I (Order ≥ 3) × Post I (Order = 2) × Post I (Order ≥ 3) × Girl I (Order = 2) × Girl Girl

I (Order ≥ 3) I (Order = 2) Wild Bootstrap-t p-value on I (Order ≥ 3) × Post × Girl Observations

Table 7 Differential effect after ultrasound ban in 1996?.

All States died by age ≤1

Law-Abiding States died by age ≤1

−0.00711* (0.00340) −6.21e−05 (0.00255) 0.00152 (0.00211) −0.00308 (0.00196) 0.00632** (0.00260) 0.00256 (0.00230) 0.0156*** (0.00224) 0.00605*** (0.00167) −0.0120*** (0.00146) −0.0134*** (0.00244) −0.0105*** (0.00237) 0.088

−0.0120** (0.00391) 0.000346 (0.00409) 0.00258 (0.00325) −0.00538* (0.00246) 0.00963*** (0.00253) 0.00371 (0.00288) 0.0184*** (0.00289) 0.00360 (0.00246) −0.0119*** (0.00234) −0.0163*** (0.00250) −0.0103*** (0.00295) 0.03

552,843

316,593

Variables

I (Order ≥ 3) × Post Reserve I (Order = 2) × Post Reserve Post Reserve

I (Order ≥ 3) × Post 1996 I (Order = 2) × Post 1996

I (Order ≥ 3) I (Order = 2) Wild Bootstrap-t p-value on I (Order ≥ 3) × Post Reserve Observations

All States boy

Law-Abiding States boy

−0.00740 (0.0107) −0.00169 (0.0101) 0.00558 (0.0104) −0.00752 (0.00871) −0.000184 (0.00793) 0.0196*** (0.00366) 0.00712* (0.00380) 0.516

−0.0387*** (0.0105) −0.0181 (0.0192) 0.0299 (0.0198) 0.0142 (0.00785) 0.0117 (0.0175) 0.0257*** (0.00359) 0.00947* (0.00444) .018

531,849

302,626

Sample weights used. State clustered standard errors reported. Sample of surviving children born in rural areas between 1987 and 2004. All specifications include birth year, state, mother's age at time of birth, mother's literacy, mother's religion, gender of the state's chief minister at the time of child's birth, and type of house fixed effects. Lawabiding states are those that have elections within 2 years of the 73rd Amendment, excluding Haryana, Maharashtra, and Orissa. P-value reported for I (Order ≥ 3) × Post Reserve using methods of wild bootstrap-t, as discussed in Cameron et al. (2008), with 1,000 repetitions. ** p < 0.05. *** p < 0.01. * p < 0.1.

Sample weights used. State clustered standard errors reported. Sample restricted to birth cohorts 1987-2004 in rural areas and to women married for at most 23 years. Lawabiding states are those that reserve seats for women by 1995, excluding Haryana, Maharashtra, and Orissa. All specifications include birth year, state, mother's age at time of birth, mother's literacy, mother's religion, gender of the state's chief minister at the time of child's birth, and type of house fixed effects. P-value reported for I (Order ≥ 3) × Post × Girl using methods of wild bootstrap-t, as discussed in Cameron et al. (2008), with 1000 repetitions. *** p < 0.01. ** p < 0.05. * p < 0.1.

order sex selection in the main analysis is not likely driven by the enforcement of the ban on ultrasound. I estimate the differential in high birth order sex ratios for children born in or after 1996, while also controlling for the time when reservations for women are made. These results are shown in Table 7. For a sample of all states, I find a negative and statistically insignificant effect of reservations on the ratio of boys at high birth orders. I also find a negative and statistically insignificant effect of the ultrasound ban on the ratio of high birth order boys. However there are 6 states in the full sample that reserve seats for women after 1996 and the timing of the ultrasound ban coincides with these states reserving political seats. If the sample is limited to law-abiding states, or those that reserved seats by 1995, the timing of seat reservations for women does not coincide with the ultrasound ban. For these states, the effect of seat reservations for women is a decline in highest birth order ratio of boys by 3.87 percentage points, and a statistically insignificant effect of the ultrasound ban on high birth order ratio of boys. Overall, the Indian ultrasound ban does not seem to explain the findings.

high infant mortality for high birth order girls and low infant mortality for high birth order boys are consistent with high birth order postnatal sex selection. Also, note that coefficients on I (Order ≥ 3) × Post and I (Order = 2) × Post are not negative. This suggests that the reduction in sex ratios is not likely explained by generally improved infant care provided by female leaders. If this were the case, then health improvements should also be observed for higher birth order male infants. Instead, I find that reported deaths for higher birth order male infants increased after female leaders are brought into power. This may appear somewhat shocking, but since higher birth order boys were 1.63 percentage points less likely to die by age one, the increase in death rates following reservations by 0.96 percentage points implies that the disproportion that existed in the death rates for high birth order boys declines. Overall these results are consistent with a reduction in postnatal sex selection and show that fewer high birth order girls died during infancy following political seat reservations for women.

7.3. The West Bengal case An additional robustness check considers the impact of reservations in West Bengal, the single state in which Panchayat elections have operated in a regular fashion. The state has held a Panchayat election every 5 years since 1978 and its election following the 73rd Amendment was also 5 years after its previous election. Thus, timing of the state of West Bengal's election can be more confidently argued to be exogenous. Table 8 presents the estimation of Eq. (1) with the inclusion of district fixed effects for the state of West Bengal alone. Standard errors are clustered at the district level. With a model of a single state, birth year fixed effects and timing of reservation are perfectly correlated, and a birth year trend common across all birth orders picks up some of the effect of reservations as a declining trend in sex ratios. Thus the estimate of Post Reserve, relative to a time trend of declining sex ratios, is positively biased. For a better understanding of how sex ratios across birth orders changed after the state reserved its

7.2. Ultrasound ban in 1996 The Indian government passed the Pre-Natal Diagnostic Techniques (PNDT) Act, which banned sex detection by ultrasound in January of 1996. Since the passage of the law coincides with the time period of the study, it is important to make sure that the PDNT Act is not explaining the effects presented in this study. By finding an effect of reduced high birth order sex ratios for children born prior to the enforcement of the ban, results in column 4 of Table 5 already suggest that the ban does not explain the findings reported in this analysis. Here, I provide further evidence that the finding of reduced high birth 9

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

Amendment, however reservations for women were not made until after the passage of the Amendment. Finding a similar effect for a state that had already set up a form of local governments suggests that the mechanism is not solely explained by devolving government power to the local level. The mechanism could also be more complex, and it is possible that exposure to female leaders changes parents' beliefs and thus reduces sex selection. In this section, I explore these possible mechanisms that could explain the reported results.

Table 8 Birth order-specific change in ratio of boys, West Bengal only.

Variables

I (Order ≥ 3) × Post Reserve I (Order = 2) × Post Reserve Post Reserve

I (Order ≥ 3) I (Order = 2) Observations

(1) boy −0.0530*** (0.0153) −0.0356** (0.0165) 0.0312*** (0.00902) 0.0387** (0.0133) 0.0109 (0.0140) 18,514

8.1. Health infrastructure: childhood mortality rates

Sample weights used. District clustered standard errors reported. Sample of surviving children born between 1987–2004 in rural areas of West Bengal. Specification includes district, mother's age at time of birth, mother's literacy, mother's religion, gender of the state's chief minister at the time of child's birth, and type of house fixed effects. * p < 0.1. *** p < 0.01. ** p < 0.05.

Previous sections present results showing that the timing of increased female leadership within a state decreased sex ratios at the higher birth orders, or at birth orders that had the most skewed sex ratios, in rural India. However there are many different channels through which these effects could operate. For example, it is possible that female leaders invest in health infrastructure, which disproportionately helps the survival of the vulnerable high birth order girl. Similarly, female leaders could also invest in infrastructure that helps the return from a female child. Women leaders could also directly impact fertility services available to their female constituents. If female leaders improve access to fertility services, then sex selection could even increase. Alternatively, the reported results could be explained by decreased access to sex-detection technology. It is also possible that the establishment of local bodies, which is also marked by the timing of elections, somehow alters sex ratios and that the involvement of female leaders is not important in explaining the findings. However, the claim that devolution of powers, and not leadership by women, explains the effect is refuted by studying the case of West Bengal. While reservations for women and the devolution of powers coincide in most states, the West Bengal case establishes that the latter is not driving the change in sex selection. West Bengal devolved powers and established a functioning Panchayat system of government at the local level well before the passage of the 73rd

While one could investigate health statistics of children in the sample to infer the availability and the success of the health facilities, the data only provide information on health investments such as vaccination status for children born between 1999 and 2004. Since information for child's birth and death is provided for a complete history of births, I am able to use death rates in early childhood as an estimate of overall investments in health. Bhalotra and Clots-Figueras (2014) also uses mortality rates as a measure of the effectiveness of health interventions. According to Million Death Study Collaborators and others (2010), pneumonia, diarrhea, and measles are the primary reasons for childhood deaths, conditional on neonatal survival, in India. Since these diseases can be treated and prevented, improving access to health services is expected to reduce childhood mortality. Investigating the impact of female political seat reservations on child mortality rates can provide an insight on whether the underlying mechanism is improved health service access, which disproportionately helps the high birth order girl. Table 6 has already shown that female political involvement is associated with a reduction in high birth order infant mortality rates for girls. This reduction in death rates before age one is consistent with a reduction in sex-selective infanticide or better care of high birth order female infants. Better care for infants could be provided without improved access to health care facilities if mothers improve care at home or increase breastfeeding. However investigating childhood death rates, conditional on survival until age one, can provide evidence on continued investments in childhood health which would be consistent with an improvement in access to health facilities. To test the claim that improved access to health facilities disproportionately helped high birth order girls during early childhood, I investigate whether they have a relatively lower childhood mortality risk following reservations. Table 9 presents results from estimating Eq. (2) but changes the dependent variable to an indicator for whether a child has died before turning 5, given that the child survived infancy. As previously discussed, a mother's recall of dead children might depend on the sex of the child. As before, I limit the sample to mothers who have been married for at most 23 years to reduce this bias.18 Conditional on surviving to age 1, girls in general, relative to boys in general, are less likely to die during childhood following female political seat reservations. Interestingly, the estimate of Post is positive and statistically significant, indicating that death rates in early childhood increased for first-born male children. There is a statistically insignificant effect of reservations for high birth order girls in both samples, implying that high birth order girls did not gain in childhood survival rates relative to first-born girls. Overall, these result suggest that the main effect is not likely explained by female leaders improving access to health services that disproportionately helped the survival of high birth order girls.

17 Inclusion of birth year fixed effects does not change the effect of reservations on high birth order sex ratios, but produces a larger coefficient, 0.09, on post-reservations sex ratios at the first birth order.

18 Results are robust to the stricter sample restriction of mothers married for at most 18 years.

seats in May 1993, I provide these results without the inclusion of birth year fixed effects.17 Within the state of West Bengal, children born at birth orders 3 or greater after reservations are 5.3 percentage points less likely to be a boy. There is also a reduction in second order sex ratios following reservations by 3.56 percentage points. As discussed earlier, West Bengal also has an increase in sex ratios at the first birth order after reservations. I find that the first birth ratio of boys increased by 3.12 percentage points, but given that the state had low first birth order sex ratios pre-reservation (0.50 ratio of boys for first births), it is possible that this increase captures mean reversion for the state. Nonetheless, the total effect for changes in sex ratios at birth orders 3 and greater is a decline in high birth order sex ratios by 2.18 percentage points and this total effect is statistically significant at the ten percent level. 8. Possible mechanisms

10

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

increases at the first parity for illiterate women. As before, the increase in first birth order sex ratios in West Bengal appears to be an important determinant of the large and positive effect. Omitting West Bengal from the sample of illiterate women in all states (Column 2) yields a small, statistically insignificant effect of Post Reserve equal to 0.009, while the main effect for high birth order children is a decrease in high birth order sex ratios by 1.48 percentage points and it is statistically significant at the one percent level.20 Overall, results in Tables 9 and 10 suggest that female leaders changing health or fertility services likely does not explain the finding of reduced ratio of high birth order boys in rural India. The poorer and the less-educated groups explaining the results also implies that the results are driven by a reduction in postnatal, as opposed to prenatal, sex selection.

Table 9 Gender-specific change in death by age 5 rates conditional of survival till age 1.

Variables

I (Order ≥ 3) × Post × Girl I (Order = 2) × Post × Girl

Post × Girl Post

I (Order ≥ 3) × Post I (Order = 2) × Post I (Order ≥ 3) × Girl I (Order = 2) × Girl Girl

I (Order ≥ 3) I (Order = 2) Wild Bootstrap-t p-value on I (Order ≥ 3) × Post × Girl Observations

All States died by age ≤5

Law-Abiding States died by age ≤5

−0.00306 (0.00320) −0.00347 (0.00366) −0.00500*** (0.00136) 0.00697*** (0.00163) −0.00609*** (0.00111) −0.00140 (0.00193) 0.0175*** (0.00322) 0.0108*** (0.00358) −0.00331 (0.00208) 0.00192 (0.00160) −0.00253** (0.00111) 0.434

−0.00398 (0.00294) −0.000235 (0.00319) −0.00591** (0.00218) 0.0100*** (0.00228) −0.00495** (0.00163) 0.000305 (0.00252) 0.0179*** (0.00242) 0.00798** (0.00299) −0.00191 (0.00301) 0.00108 (0.00254) −0.00293* (0.00140) 0.85

537,547

307,025

8.3. Exposure to female leaders To better understand the mechanism, I investigate if the districtlevel leadership is important in driving the results. Results thus far have not explored which level of female leadership (district, block, or the village) is important in driving the results as all three Panchayat levels saw an increase in female representation at the same time. Exploiting data on reservations for female chairpersons at the districtlevel Panchayats, I show that a female chairperson reservation at the district level had no impact on sex selection in rural areas. Since female leaders at the district level do not drive the results, it is suggested that reservations for women at either the village or the block level are likely driving the results for rural areas. This is consistent with literature finding that lower-level female empowerment is most important in explaining the effects (Iyer et al., 2012). Perhaps more surprising is my finding that a female chairperson seat reservation at the district-level decreases sex ratios in urban areas. Given that the responsibilities of district-level Panchayat leaders lie in rural areas, it is not expected that urban areas exhibit a decrease in sex ratios. I argue that there are reasons to believe that urban populations have greater exposure to female district-level chairpersons even though they operate under a local body that focuses on the development of rural areas. For one, the district Panchayat offices are located in the district headquarters, which are urban towns. Additionally, local newspapers often report developments occurring at the district Panchayat. Given that readership of newspapers is much greater in urban areas, access to newspapers also increases exposure to district Panchayat leaders in urban versus rural areas.21 To show this, I employ data on random assignment of female chairpersons at the district level and estimate a simple model for both urban and rural populations separately. Note that unlike all of the other estimation results in this paper thus far that only study rural areas (where the 73rd Amendment applied), when studying the direct effect of district Panchayat chairperson seat reservations, I present results for both urban and rural communities. Using district-level data, I estimate a DD model described in Eq. (3). I estimate the equation for a sample of surviving children born after 1990 and before the second round of

Sample weights used. State clustered standard errors. Sample restricted to children who survived until age one, born between 1987–2004 in rural areas, and to women married for at most 23 years. Law-abiding states are those that reserve seats for women by 1995, excluding Haryana, Maharashtra, and Orissa. All specifications include birth year, state, mother's age at time of birth, mother's literacy, mother's religion, gender of the state's chief minister at the time of child's birth, and type of house fixed effects. P-value reported for I (Order ≥ 3) × Post × Girl using methods of wild bootstrap-t, as discussed in Cameron et al. (2008), with 1000 repetitions. *** p < 0.01. ** p < 0.05. * p < 0.1.

8.2. Fertility infrastructure Another possible mechanism through which increasing the number of female local leaders could reduce sex ratios is that women leaders restrict access to sex-detection technology. Although, I do not directly observe access to fertility services for each birth, it is possible to infer access by studying the effect across different socioeconomic statuses of mothers. Studies have shown that sex-selective abortion in India is most common amongst the most educated (Jha et al., 2011; Pörtner, 2016) and the wealthiest groups (Jha et al., 2011). These studies also show that sex selection does not appear to occur for firstborns across all socioeconomic statuses in India. It is useful to split the main analysis by education and wealth level to infer whether reduced access to sexselective abortion is explaining the results. Table 10 presents the results by literacy status of the mother. Given that more educated women are more likely to use sex-selective abortions, the impact of reduced access to sex-detection technology is expected to be stronger for them. However, results in Table 10 show that the effect of declined high birth order sex selection is driven by less educated women.19 Analogous results by household wealth levels are also shown in Table A1 of the Appendix. Those results are similar to the findings in Table 10 and show that female reservations only reduced high birth order sex ratios in the poorest households. It is worth noting that within the sample of all states, ratio of boys

20 Omitting West Bengal from Columns 1, 2, 3, and 4 respectively provides a statistically insignificant effect of reservations on the ratio of first-born boys equal to -0.008, 0.009, 0.0107, and 0.0084 respectively. The exclusion of West Bengal from Columns 1, 2, 3, and 4 does not change the main effect of reservations on the ratio of boys at the highest birth order by very much, and the estimates are similar to those in Table 10. 21 The 2005–2006 National Family Health Survey (NFHS) data reveal that 70 percent of the men in rural areas who were 20 years of age or older in 1995 read a newspaper with less frequency than once a week. The analogous ratio for men of the same age distribution was only 30 percent in urban areas. Similar estimates for women age 20 or older in 1995 reveals that 85 percent of them in rural areas read a newspaper with less frequency than once a week. However, amongst women in urban areas that were of age 20 or older in 1995, 57 percent of them read a newspaper with less frequency than once a week. This suggests that readership of newspapers around the law change was much more prevalent in urban areas making urban areas more aware of the female leaders at the district-level Panchayat.

19 The data also include a variable for the mother's years of schooling, but it is extremely incomplete for women in rural areas. Nonetheless, defining sample cut-off points based on years of schooling show that women with the most schooling (8 or more years of schooling) are not driving the results.

11

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

Table 10 Heterogeneous Effects by Mother's Literacy. All States

Law-Abiding States

Can Read and Write?

Yes

No

Yes

No

I (Order ≥ 3) × Post Reserve

0.00156 (0.0106) 0.00680 (0.00842) −0.00241 (0.0113) 0.00999 (0.00828) 0.000531 (0.00758) 0.926

−0.0176*** (0.00550) −0.00748 (0.00544) 0.0146* (0.00805) 0.0232*** (0.00426) 0.0111*** (0.00257) 0.01

−0.00979 (0.00826) 0.00379 (0.0103) 0.0160 (0.0171) 0.0107 (0.00681) 0.00165 (0.0107) 0.298

−0.0303*** (0.00650) −0.0140 (0.00762) 0.0249 (0.0179) 0.0323*** (0.00376) 0.0143*** (0.00326) 0.008

191,652

340,122

92,727

209,859

I (Order = 2) × Post Reserve Post Reserve

I (Order ≥ 3) I (Order = 2) Wild Bootstrap-t p-value on I (Order ≥ 3) × Post Reserve Observations

Sample weights used. State clustered standard errors reported. Sample restricted to surviving children born between 1987–2004 in rural areas. Law-abiding states are those that reserve seats for women by 1995, excluding Haryana, Maharashtra, and Orissa. All specifications include birth year, state, mother's age at time of birth, mother's literacy, mother's religion, gender of the state's chief minister at the time of child's birth, and type of house fixed effects. P-value reported for I (Order ≥ 3) × Post Reserve using methods of wild bootstrap-t, as discussed in Cameron et al. (2008), with 1000 repetitions. ** p < 0.05. *** p < 0.01. * p < 0.1.

points decline in the share of boys for children born at the highest birth order (Column 2).23 Columns 1 and 2 do not include district fixed effects and the effect of whether a district eventually reserved a seat for a woman, Reservedd, is identified. It is important that the estimate of Reservedd is statistically indistinguishable from zero, as it shows that, prior to reservations, high birth order sex ratios were similar in districts that eventually reserved seats for women to those that did not. I find this to be the case. Columns 3 and 4 present results from rural and urban areas respectively, with the inclusion of district fixed effects. The estimates are very similar with and without district fixed effects. These results indicate that areas where exposure to female district Panchayat chairperson is arguably greater exhibit a decline in high birth order sex ratios, whereas areas where goods and services are provided do not. Additionally, finding that a reservation for a female chairperson at the district Panchayat does not alter sex ratios in rural areas implies that female leaders at lower levels are likely driving the results shown in this paper for rural areas. Given that the village Panchayat interacts with the village population frequently (2–3 times a year in most village Panchayats), exposure to female leaders serving under the 73rd Amendment is also likely greatest at lower levels of local bodies.

elections that reserved seats for women. This allows me to cleanly identify the effect of the district chairperson seat reservation for a woman on the likelihood that the child is male, as the control group has never had seats reserved for women previously. To simplify the model, the sample is further restricted to children of birth orders 3 or greater. This is based on previous sections' findings that sex selection in India is largely observed at high birth orders. Nevertheless, the results from estimating a full model with children of all birth orders in the sample yields analogous results.22 The analysis is limited to the states Andhra Pradesh, Gujarat, Kerala, Rajasthan, and West Bengal. These are the states in the preferred sample for which I have district Panchayat reservation status data on the first election.

Boyidc = β1 Reservedd × Postdc + β2 Postdc + γc + ρd + ΓXidc + ϵidc

(3)

In Eq. (3), the dependent variable is an indicator for whether child i, of district d, and of birth year cohort c is a boy. It is regressed on the interaction of whether the chairperson seat at the district Panchayat was reserved for a woman, Reservedd, with whether the child is born after the district reservations were made, Postdc. Also included in the model is a fixed effect for Postdc. District and birth cohort fixed effects are included to control for district-specific and cohort-specific effects respectively. Additional controls in Xidc are identical to those in specification (1). Note that the main effect of reservation, Reservedd, drops out because the model includes district fixed effects. To estimate the effect of district reservation status, Reservedd, I also estimate Eq. (3) without district fixed effects. Table 11 shows the estimates of Eq. (3) that study the impact of chairperson seat reservations at the district Panchayat for urban and rural areas. Column 1 presents the estimation results for a sample of high birth order children from rural areas, and I do not find that district chairperson reservations for women had a statistically significant impact on the likelihood that a child is a boy. The sign of the effect of female district Panchayat chairperson in rural areas is even positive, suggesting district-level female chairpersons might have even eased families' ability to sex select in rural areas. In urban areas, however, female district chairperson reservations lead to a 7.43 percentage

9. Conclusion This paper shows that the share of boys at high birth orders declines for children born following the implementation of local political seat reservations for women in rural India. I also find that infant mortality rates for higher birth order girls decline. My findings are consistent with the prior that female political empowerment can reduce gender bias, and hence sex selection. The results in this paper are also robust to several tests. I argue that the results are driven by exposure to female leaders, and not through a change in provision of goods and services that benefit high birth order girls. I find that increased female leadership is not associated with improved high birth order female child mortality 23 A valid concern is that women in urban areas reduce their overall fertility, and thus move sex-selective behavior to earlier-born children. However, I do not find a statistically significant effect of timing of district-level chair reservation on the likelihood a mother observes a birth in either rural or urban areas.

22

Estimating a difference-in-difference-in-differences version of Eq. (3) with all birth order-specific effects for a sample of children of all birth orders yields similar results. These are presented in Table A2 of the Appendix.

12

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

with the hypothesis that prenatal and postnatal discriminations are substitutes (Goodkind, 1996; Lin et al., 2014) and implies that bans on prenatal sex selection could lead to increased postnatal sex selection, postnatal female discrimination, and the prevalence of dangerous illegal abortions. A better way to address the issue would be to target the underlying son preference that leads to sex selection by addressing the status of women directly. This research provides an example of one such policy, the reservation of local seats for women, which is shown to reduce the prevalence of sex selection.

rate, suggesting that female leaders did not improve health infrastructure in a manner that benefited high birth order girls the most. Moreover, I find that women from worse socioeconomic backgrounds drive the results. This suggests that women leaders did not restrict fertility services, since this group is not likely to use sex-selective abortions. The source of the effect in rural areas appears to be reservations for women closest to home, at the village-level or the block-level Panchayat. This is indicative that exposure to female leaders is important. In addition, I find that reservations at the district-level decreased sex ratios in urban and not in rural India. These results further validate the idea that visibility of female leaders is of greatest importance in the relationship between sex selection and female reservations, as the female leaders of the district-level Panchayat are likely more visible to the urban population due to newspaper readership and the location of the district Panchayat buildings. However, this paper is unable to investigate the channel through which exposure to female leaders may reduce sex selection. It is possible that the conceived return from a daughter is increased. Future studies are encouraged to investigate changes in return from a daughter following the 73rd Amendment. These results also shed light on previous findings in the literature on sex selection. Kalsi (2015) shows that sex selection in Taiwan is more prevalent following the legalization of abortion, and that girls born at high birth orders are more likely to attend a university if they are born following the legalization of abortion. This result is consistent

Acknowledgements I thank Francisca Antman for providing incredible feedback, help, and support throughout this project. I also thank Tania Barham, Terra McKinnish, and Murat Iyigun for all of their feedback that improved this work greatly. My peers and good friends Nathan Adkins, Zachary Feldman, Dustin Frye, Xavier Gitaux, Gisella Kagy, Edward Kosack, Steven Smith, and Zachary Ward provided especially important conversations and feedback that helped significantly improve this work. I also thank Yuya Kuda, participants at the Northeast Universities Development Consortium 2014, and participants at presentations given at the University of Colorado. I thank Lakshmi Iyer for sharing data on district-level reservations. I also thank the Department of Economics at the University of Colorado for assisting to pay for the DLHS data sets. Any remaining errors are mine alone.

Appendix A Figs. A1–A3. Tables A1 and A2.

Fig. A1. Panchayat Raj Institutions: Three-tier structure of local rural government. Average number of elected members given. Average population served per elected official is reported in parenthesis. Source: (Alok, 2011).

13

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

Fig. A2. States with elections within 2 years following the implementation of 73rd Amendment.

Fig. A3. States with elections within 2 years following the implementation of 73rd Amendment: excluding states that had elections prior to the reform and Haryana.

14

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

Table A1 Heterogeneous effects by household wealth. All States

Law-Abiding States

Strength of House Construction

Weak

Medium

Strong

Weak

Medium

Strong

I (Order ≥ 3) × Post Reserve

−0.0160* (0.00772) −0.0117* (0.00631) 0.0105 (0.00925) 0.0183** (0.00690) 0.0118** (0.00522) 0.054

−0.00762 (0.0114) 0.00685 (0.00715) 0.00391 (0.00990) 0.0171** (0.00769) 0.00244 (0.00506) 0.52

−0.0102 (0.0126) 0.00315 (0.0121) 0.00638 (0.0153) 0.0223** (0.00818) 0.00573 (0.0111) 0.454

−0.0285** (0.0107) −0.0137 (0.00858) 0.0170 (0.0196) 0.0288** (0.00930) 0.0124 (0.00669) 0.004

−0.0252*** (0.00718) −0.000971 (0.00896) 0.0294 (0.0195) 0.0231** (0.00855) 0.00394 (0.00581) 0.048

−0.0191 (0.0170) −0.00674 (0.0140) 0.0237 (0.0216) 0.0251* (0.0109) 0.0136 (0.0148) 0.31

234,992

203,049

93,808

124,646

118,155

59,825

I (Order = 2) × Post Reserve

Post Reserve

I (Order ≥ 3) I (Order = 2) Wild Bootstrap-t p-value on I (Order ≥ 3) × Post Observations

Sample weights used. State clustered standard errors reported. Sample restricted to surviving children born between 1987-2004 in rural areas. Law-abiding states are those that reserve seats for women by 1995, excluding Haryana, Maharashtra, and Orissa. All specifications include birth year, state, mother's age at time of birth, mother's literacy, mother's religion, gender of the state's chief minister at the time of child's birth, and type of house fixed effects. P-value reported for I (Order ≥ 3) × Post Reserve using methods of wild bootstrap-t, as discussed in Cameron et al. (2008), with 1000 repetitions. *** p < 0.01. ** p < 0.05. * p < 0.1. Table A2 The effect of first district-level Panchayat election.

Variables

I (Order ≥ 3) × Reserved × Post I (Order = 2) × Reserved × Post

Reserved × Post

I (Order ≥ 3) × Post I (Order = 2) × Post I (Order ≥ 3) × Reservedd I (Order = 2) × Reservedd

I (Order ≥ 3) I (Order = 2) Post N Districts Observations

(1) boy Rural

(2) boy Urban

0.0440 (0.0272) 0.0299 (0.0264) −0.0200 (0.0195) −0.0549*** (0.0159) −0.0275* (0.0163) −0.00722 (0.0187) −0.0177 (0.0195) 0.0308** (0.0136) 0.0211* (0.0108) 0.0370** (0.0147) 102 64,014

−0.0647** (0.0314) −0.00144 (0.0490) −0.0181 (0.0431) 0.0454** (0.0184) 0.0297 (0.0207) 0.0144 (0.0244) 0.0207 (0.0248) 0.0172 (0.0213) −0.0121 (0.0164) −0.0151 (0.0287) 102 26,483

District clustered standard errors. All specifications include district, birth year, mother's age at time of birth, mother's literacy, mother's religion, gender of the state's chief minister at the time of child's birth, and type of house fixed effects. Sample consists the states of Rajasthan, West Bengal, Gujarat, Andhra Pradesh, and Kerala. Sample restricted to surviving children born after 1990 until first district reservations were in effect. *** p < 0.01. ** p < 0.05. * p < 0.1.

Appendix B. Sex-selective abortions B.1. Historical context of sex-selective abortion availability in India Abortion in India was made legal in 1971 under the Medical Termination of Pregnancy Act (Visaria, 2007). Under the act, abortion is only allowed when pregnancy carried health risks to a woman, endangered her mental health, resulted from rape, or resulted from contraceptive failure (Visaria, 2007). Due to the many restrictions under which abortion is allowed to be carried out legally and also because of the limited availability of clinics that offered legal abortions, illegal abortions have been estimated to be between 8 to 11 times more common than legal abortions (Chhabra 15

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

Fig. B1. Mean ratio of boys and 95% confidence intervals for first-born children by years of marriage.

Fig. B2. Mean ratio of boys and 95% confidence intervals for first-born children by years of marriage. Sample restricted to law-abiding states that reserved seats by 1995, excluding Haryana, Orissa, and Maharashtra.

Table B1 Effect on sex-selective abortions. Samples restricted by mother's years of marriage.

Variables

I (Order ≥ 3) × Post Reserve I (Order = 2) × Post Reserve Post Reserve

I (Order ≥ 3) I (Order = 2) Years married? Law-abiding states only? Wild-boot p-value on I (Order ≥ 3) × Post Observations

(1) boy

(2) boy

(3) boy

(4) boy

0.00398 (0.00604) 0.00104 (0.00595) 0.000378 (0.00738) −0.00139 (0.00471) 0.000844 (0.00325) ≤18 N 0.538

−0.00137 (0.0107) −0.00364 (0.00918) 0.00567 (0.0130) −0.000753 (0.00803) 0.00226 (0.00507) ≤18 Y 0.924

−0.00207 (0.00468) 0.00125 (0.00555) 0.00291 (0.00638) 0.00319 (0.00310) 0.000568 (0.00309) ≤23 N 0.684

−0.00934 (0.00689) −0.00365 (0.00823) 0.00638 (0.0113) 0.00561 (0.00453) 0.00209 (0.00461) ≤23 Y 0.20

459,571

260,979

552,847

316,596

Sample weights used. State clustered standard errors reported. Sample restricted to all children born between 1987–2004 in rural areas. Law-abiding states are those that reserve seats for women by 1995, excluding Haryana, Maharashtra, and Orissa. All specifications include birth year, state, mother's age at time of birth, mother's literacy, mother's religion, gender of the state's chief minister at the time of child's birth, and type of house fixed effects. *** p < 0.01, ** p < 0.05, * p < 0.1.

16

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

Table B2 Effect on postnatal sex selection. Samples restricted by mother's years of marriage.

Variables

I (Order ≥ 3) × Post Reserve I (Order = 2) × Post Reserve

Post Reserve

I (Order ≥ 3) I (Order = 2) Years married? Law-abiding states only? Wild-boot p-value on I (Order ≥ 3) × Post Observations

(1) boy

(2) boy

(3) boy

(4) boy

−0.00433 (0.00672) −0.00205 (0.00601) 0.00440 (0.00980) 0.0125** (0.00516) 0.00741* (0.00384) ≤18 N 0.508

−0.0170 (0.0114) −0.00797 (0.00802) 0.0249 (0.0182) 0.0192* (0.00923) 0.00976** (0.00414) ≤18 Y 0.142

−0.0102* (0.00572) −0.00161 (0.00545) 0.00712 (0.00846) 0.0171*** (0.00372) 0.00703* (0.00362) ≤23 N 0.084

−0.0231*** (0.00607) −0.00781 (0.00675) 0.0233 (0.0164) 0.0236*** (0.00393) 0.00946** (0.00388) ≤23 Y 0.0

415,958

233,287

499,337

282,491

Sample weights used. State clustered standard errors reported. Sample restricted to surviving children born between 1987-2004 in rural areas. Law-abiding states are those that reserve seats for women by 1995, excluding Haryana, Maharashtra, and Orissa. All specifications include birth year, state, mother's age at time of birth, mother's literacy, mother's religion, gender of the state's chief minister at the time of child's birth, and type of house fixed effects. *** p < 0.01. ** p < 0.05. * p < 0.1.

and Sheel, 1993). While illegal providers include non-medical personnel, medical professionals and gynecologists are also known to provide illegal abortions (Arnold et al., 2002). According to government statistics, approximately 2.7 in 100 known pregnancies result in an abortion in India (Arnold et al., 2002). Ultrasound became common in India only after the early 1990s (Clark, 2000). The use of ultrasound and the subsequent increase in sex-selective abortions led to a feminist movement that demanded the practice of sex detection to be banned. The Government of India responded with the passage of the Pre-Natal Diagnostic Techniques (PNDT) Act of 1994, which came into force in January of 1996 (Government of India, 2006). The PNDT Act did not ban the use of ultrasound for prenatal care, but prohibited the sex of the child to be revealed. Since the Act, there has been no indication that the PNDT reduced sex selection (Visaria, 2007), and the enforcement of the law has been practically nonexistent (Arnold et al., 2002). Retherford and Roy (2003) highlights the many loopholes in the law. For example, government clinics are monitored much more closely than private clinics (Retherford and Roy, 2003). Moreover, doctors now reveal the sex of the child verbally instead of in writing (Retherford and Roy, 2003). Due to an increase in sex selection, child sex ratios in India have worsened from 105.8 in 1991, to 107.8 in 2001, and more recently to 109 in 2011 (The Economist, 2011). While the ability to sex select has risen with the introduction of ultrasound, its use is most common amongst more educated women, and less educated women do not appear to rely on abortions as a means to sex select (Pörtner, 2016). In what follows, I study if sex-selective abortions, specifically, are affected in rural India after increased female representation. B.2. The effect of female political leaders on sex-selective abortions Here, I test if the use of sex-selective abortions declined after increased female political representation. Since sex-selective abortions change sex ratios at birth, I rely on the woman's entire history of birth in this section. This is different from the results in the paper which study the ratio of boys amongst children who are alive at the time of the survey. One must employ caution when dealing with birth data, as they are known to suffer from recall bias with mothers failing to report dead children. Moreover, in a society with male preference, underreporting is a greater concern for dead female children, resulting in higher reported sex ratios at birth than in actuality. Therefore, it is important to correct for recall bias when studying self-reported sex ratios at birth. Recall error, however, has been consistently shown to increase with the length of the recall period (Beckett et al., 2001; Malthiowetz et al., 2001). One way to reduce recall error is to focus on women who were married relatively recently (Pörtner, 2016). I first show how recall error varies overtime in my sample. Figs. B1 and B2 show the ratio of first-born boys by the number of years the mother has been married. Since sex selection at first parity is uncommon, reported first birth order sex ratios at birth which are significantly higher than the natural ratio (0.512) suggest systematic recall error is a problem. Fig. B1 suggests that recall starts to become a problem for women in all states after 18 years of marriage, as the ratio of boys is significantly higher than the natural ratio of boys. For the preferred sample of states, Fig. B2 suggests that recall is more problematic after 23 years of marriage. One way to study if sex-selective abortions became less common following increased female political representation is to study sex ratios at birth for women less likely to suffer from recall bias. Table B1 presents the results of estimating Eq. (1) on a sample of all reported births, while focusing on a sample of women for whom recall is likely less problematic. I present the results by both cut-off points of women who have been married for 18 or fewer years and women married for 23 or fewer years. Since the first group of women to change their decision to abort would have to be in the first trimester of their pregnancy when female leadership increases, treatment (Post Reservecs) is redefined as 6 months after reservations. For a sample in which recall is less problematic, the estimated effects of female political seat reservations on the highest birth order sex ratios at birth are both small and statistically indistinguishable from zero. While the purpose of this exercise is to study if the use of sex-selective abortions changed after increased female political representation, reducing the sample size in such manner may alter the main results as well. For comparison, I also present the main results which focus on surviving children (and thus also study postnatal sex selection) with the same sample restrictions as in Table B1. Since Table B2 focuses on the effects on 17

Journal of Development Economics 126 (2017) 1–18

P. Kalsi

postnatal sex selection, treatment is defined such that all children born after the month their state reserved seats are coded as being born after reservations. The results show that the sample cut-off point of women married for 18 or fewer years does not yield statistically significant results. However, the main effect is robust to the more generous sample cut-off point of women married for at most 23 years. It could be that limiting the sample to women who have been married for 18 or fewer years is too restrictive. The median years of marriage is 11 for this sample. Given that the survey was completed between 2002 and 2004, a large proportion of the sample only married after political seats for women were already reserved. The sample of women married for at most 23 years is less restrictive with the median years of marriage being 13 years, allowing for a larger sample of women who were married at the time of first female seat reservations. Overall, Table B1 suggests that the findings of the paper are likely not explained by a reduction in sex-selective abortions. Furthermore, it does not appear that the lack of an effect is purely due to new sample restrictions. Appendix C. Supplementary data Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.jdeveco.2016.12.002. entrepreneurship in india. J. Dev. Econ. 108, 138–153. Goodkind, D., 1996. On substituting sex preference strategies in East Asia does prenatal sex selection reduce posnatal discrimination. Popul. Dev. Rev. 22 (March (1)), 111–125. Government of India, 2006. Handbook on: Pre-conception and Pre-natal Diagnostic Techniques Act, 1994 and rules with amendments. Technical report, Ministry of Health and Family Welfare- Government of India. Iyer, L., Mani, A., Mishra, P., Topalova, P., 2012. The power of political voice women's political representation and crime in India. Am. Econ. J.: Appl. Econ. 4 (4), 93–165. Jain, D., 1996. Panchayat raj: Women changing governance. Technical report, Gender if development monograph seried number 5, New York: UNDP. Jayal, N.G., 2006. Engendering local democracy the impact of quotas for women in India's panchayats. Democratization 13 (1), 15–35. Jha, P., Kesler, M.A., Kumar, R., Ram, F., Ram, U., Aleksandrowicz, L., Bassani, D.G., Chandra, S., Banthia, J.K., 2011. Trends in selective abortions of girls in India analysis of nationally representative birth histories from 1990 to 2005 and census data from 1991 to 2011. Lancet 377 (9781), 1921–1928. Kalsi, P., 2015. Abortion legalization, sex selection, and female university enrollment in taiwan. Econ. Dev. Cult. Change 64 (1), 163–185. Lin, M.-J., Qian, N., Liu, J.-T., 2014. More missing women, fewer dying girls The impact of sex- selective abortion on sex at birth and relative female mortality in Taiwan. J. Eur. Econ. Assoc. 12 (4), 899–926. Malthiowetz, N., Brown, C., Bould, J., 2001. Measurement error in surveys of the lowincome populations. Stud. Welf. Popul.: Data Collect. Res. Issues, 157–194. Mathew, G., 2000. The Status of Panchayati Raj in the States and Union Territories of India 2000. Concept Publishing Company. Million Death Study Collaborators and others, 2010. Causes of neonatal and child mortality in India: a nationally representative mortality survey. The Lancet 376(9755), pp. 1853–1860. Pörtner, C.C., 2016. Sex selective abortions, fertility, and birth spacing. Working Paper. Qian, N., 2008. Missing women and the price of tea in China the effect of sex-specific earnings on sex imbalance. Q. J. Econ. 123 (August (3)), 1251–1285. Retherford, R.D., Roy, T., 2003. Factors affecting sex-selective abortion in India and 17 major states. Technical Report 21, International Institute for Population Services.? Rose, E., 1999. Consumption smoothing and excess female mortality in rural India. Rev. Econ. Stat. 81 (February (1)), 41–49. Sinha, N., Yoong, J., 2009. Long-term financial incentives and investment in daughters: Evidence from conditional cash transfers in North India. Gender and Development. The Economist (2011, April 9,). Seven brothers: An aversion to having daughters is leading to millions of missing girls. Thukral, R., 2006. Indiastat E-yearbook 2006: Socio-Economic Reference Database of the Post Reform Period in India, 1991 to 2005. Publication Cell, Datanet India Pvt. Limited, New Delhi. Visaria, L., 2007. Deficit of girls. In: Patel, T. (Ed.), Sex-selective Abortion in India: Gender, Society and new Reproductive Technologies. Sage Publications. Vyasulu, P., Vyasulu, V., 1999. Women in panchayat raj. Grass roots democracy in Malgudi. Economic and Political Weekly.

References Alok, V., 2011. Role of panchayat bodies in rural development since 1959. Technical report, Indian Institute of Public Administration New Delhi. Anderson, S., Ray, D., 2010. Missing women age and disease. Rev. Econ. Stud. 77 (October), 1262–1300. Arnold, F., Kishor, S., Roy, T.K., 2002. Sex-selective abortions in India. Popul. Dev. Rev. 28 (4), 759–785. Ban, R., Rao, V., 2008. Tokenism or agency? The impact of women's reservations on village democracies in South India. Econ. Dev. Cult. Change 56 (3), 501–530. Beaman, L., Chattopadhyay, R., Duflo, E., Pande, R., Topalova, P., 2009. Powerful women does exposure reduce bias? Q. J. Econ. 124 (4), 1497–1540. Beaman, L., Duflo, E., Pande, R., Topalova, P., 2010, February. Political reservation and substantive representation: evidence from Indian village councils. Technical report, Brookings Institution. Beaman, L., Duflo, E., Pande, R., Topalova, P., 2012. Female leadership raises aspirations and educational attainment for girls a policy experiment in India. Sci. Mag. 335 (February (6068)), 582–586. Beckett, M., Da vanzo, J., Sastry, N., Panis, C., Peterson, C., 2001. The quality of retrospective data: an examination of long-term recall in a developing country. J. Human. Resour., 593–625. Bertrand, M., Duflo, E., Mullainathan, S., 2004. How much should we trust differencesin-differences estimates? Q. J. Econ. 119 (1), 249–275. Bhalotra, S., Clots-Figueras, I., 2014. Health and the political agency of women. Am. Econ. J.: Econ. Policy 6 (2), 164–197. Cameron, A.C., Gelbach, J.B., Miller, D.L., 2008. Bootstrap-based improvements for inference with clustered errors. Rev. Econ. Stat. 90 (3), 414–427. Chattopadhyay, R., Duflo, E., 2004. Women as policy makers evidence from a randomized policy experiment in India. Econometrica 72 (5), 1409–1443. Chaudhuri, S., 2003, May. What difference does a constitutional amendment make? The 1994 panchayati raj act and the attempt to revitalize rural local government in India. Chhabra, R., Sheel, C.N., 1993. Abortion in India-an Overview. Veerendra Printers, Delhi. Chu, J., 2001. Prenatal sex determination and sex-selective abortion in rural central China. Popul. Dev. Rev. 27, 259–281. Clark, S., 2000. Son preference and sex composition of children evidence from India. Demography 37 (1), 95–108. Das Gupta, M., 1987. Selective discrimination against female children in rural Punjab, India. Popul. Dev. Rev. 12 (March (1)), 77–100. Das Gupta, M., Zhenghua, J., Zhenming, X., Bohua, L., Chung, W., Hwa-Ok, B., 2003. Why is son preference so persistent in East and South Asia? A cross-country study of China, India and the republic of Korea. J. Dev. Stud. 40 (2), 153–187. Ebenstein, A.Y., 2007, December. Sex preference, sex selection, and women’s labor supply. University of California, Berkeley, (Ph.D. thesis). (14541). Ghani, E., Kerr, W.R., O'Connell, S.D., 2014. Political reservations and women's

18