In low-fertility populations, childlessness has increased in recent decades. Not having children is no longer only a physiological constraint, but also a choice. In countries where offspring are more numerous and considered integral to unions, childlessness often carries a stigma and is more likely to affect less advantaged populations. This situation could change with socioeconomic transformations and the emergence of educated population groups freeing themselves from traditional family values. India is an example of this change, as described in this article. The country is characterized by rare but polarized childlessness at the two extremes of the social scale, which for the authors draws a line between poverty-related childlessness among the least educated women and opportunity-related childlessness among the most educated.
1We propose a new interpretation of the dynamics of childlessness in India over recent decades. Based on the recent decomposition of childlessness by Baudin et al. (2015), we use micro-level data to show that a significant part of childlessness among Indian couples can be explained by the emergence of better educational and economic opportunities for women. This phenomenon exhibits clear geographical heterogeneity, without contradicting the fact that childlessness among many Indian women is a result of primary infertility due to sterility, sexually transmitted diseases, and poverty.
2Given the strong demographic pressure faced by India (United Nations, 2019), the attention of demographers has been monopolized by the question of when and how the Indian fertility transition will end, as well as questions about family planning programs. Consequently, little room remains for studying another important dimension of fertility: childlessness. We contribute to filling in this literature gap by focusing on definitive childlessness among married Indian women. Childlessness is defined by the absence of any living birth in a woman’s life; and definitive childlessness usually refers to a woman remaining childless after age 40.  The latest census, conducted in 2011, recorded the highest ever definitive childlessness rate in India: 7.89% of women over age 40. Although the rate is only around half that of richer countries like the United States (measured for the year 2014; Livingston, 2015), it is nevertheless far from the approximately 1% natural sterility rates among 25-year-old women estimated by Leridon (2008). Childlessness in India is clearly not a marginal issue.
I – Theoretical hypothesis
3We begin by discussing why we rely on the decomposition proposed by Baudin et al. (2015) rather than on the more usual distinction between voluntary and involuntary childlessness. Then, we develop a theoretical framework adapted to the specific context of India. From this emerges a set of hypotheses to be tested in subsequent sections.
1 – Decomposing childlessness
4The literature on childlessness has extensively discussed the issue of voluntary versus involuntary. For example, Poston and Cruz (2016) explored the National Survey of Family Growth data on American women to discuss alternative methods for separating voluntarily from involuntarily childless women. They explain why none of these methods are perfect and point to the need for intensive personal interviews to identify the underlying factors of childlessness related to women’s reproductive health, personal aspirations, and entire marital history. To our knowledge, large datasets built on such interviews do not exist for India or any other country.
5The absence of accurate data for India makes it impossible not only to break down childlessness into its voluntary and involuntary components, but also—and more importantly—to evaluate their socioeconomic gradients. For this reason, we use the decomposition proposed by Baudin et al. (2015), who distinguish three kinds of childlessness: natural sterility, poverty-driven childlessness, and opportunity-driven childlessness.
6While natural sterility refers to the innate inability to give birth and is uniformly distributed in the population (Leridon, 2008), poverty-driven childlessness refers to women who fail to have children because of their poverty. One main cause of definitive childlessness among the poor in less developed countries lies in their higher degree of exposure to venereal diseases and malnutrition (Retel-Laurentin, 1974; McFalls, 1979; Romaniuk, 1980; Frank, 1983; Poston et al., 1985). Weiss et al. (2008) and Solomon et al. (2009) confirmed that married women belonging to socially disadvantaged groups in India are at higher risk of infection by chlamydia and gonorrhea (STDs that potentially cause infertility) and HIV. In addition to poverty, Madhivanan et al. (2009) documented an increased risk of adverse pregnancy outcomes caused by Trichomonas vaginalis among women exposed to sexual intercourse at an early age through marriage. Even if recent evidence regarding malnutrition and fecundity may seem less salient than in the past, Nanda (2009) at least documented a positive association between low fertility and stunting among women in parts of India. Notably inegalitarian in our current times, poor societies lack universal access to assisted reproduction techniques (ART), which reinforces the socioeconomic gradient of poverty-driven childlessness: the rich can afford expensive ART, whereas the poor are excluded from it. Allahbadia (2013) and Rasool and Akhtar (2018) and have documented poor people having low access to ART in India.
7Opportunity-driven childlessness refers to women who have delayed motherhood to the point where having children has become unfeasible or undesirable.  One main reason behind this postponement is the time cost of having children. India is similar to most societies for whom having children requires investing time in child-rearing instead of labor market activities—for women much more than for men. Renouncing labor market activities forms part of the opportunity cost of having children (Becker, 1981). Another dimension of this opportunity cost is added by the need to abandon other personal aspirations whose importance is assumed to be positively associated with education. This fact was documented by Ghosh (2015) in the case of Kolkata’s second demographic transition, while Surkyn and Lesthaeghe (2004) covered this in a more general discussion.
8Our decomposition ascribes no reasons for childlessness to each childless woman, an approach that fares no better than the classical decomposition into voluntary and involuntary childlessness. Nevertheless, we depart from the classical decomposition (for instance, into wanting or never wanting to have children) by abandoning attitudinal concepts toward childlessness and instead rely on the measurable socioeconomic and biological factors of childlessness.  This conceptual framework allows us to formulate a theory of childlessness in India, which we describe in the next section.
2 – A theory of childlessness for India
Poverty, opportunities, and education
9Finding reliable and precise data on poverty and economic opportunities is generally difficult when exploring these as causes of childlessness. Baudin et al. (2019) argued that these two phenomena are advantageously proxied by women’s educational attainment. In this section, we analyze the implications of this argument and how to transpose it to the Indian context. Appendix Figure A.1 summarizes our entire theory in a causal diagram.
10Regarding economic opportunities, most datasets cover women’s professions—information with severe limitations. Indeed, a woman’s job at age 40+ measures the opportunities she has taken advantage of at the time of the interview but not those she was offered during her fertile years. This becomes an issue in cases of women having chosen not to work or ended up in that status. Not working does not necessarily mean that a woman did not enjoy good economic opportunities; it may also mean she is not enjoying them at the time of the interview. Low female labor force participation rates in India (Chatterjee et al., 2018) make current occupational status a weak proxy for economic opportunities, which is why education level may be better for measuring the economic opportunities offered to a woman in her youth.
11As for poverty, both absolute and relative measures are reliable for India; for instance, wealth percentiles. However, both types are limited because they are measured at the time of the survey. Temporality remains a major issue, just as it does for economic opportunities: poverty at age 40+ is only a proxy for poverty suffered during reproductive life. Again, educational attainment, decided before or just at the beginning of fertile life, circumvents this issue.
12If education is in general a good proxy for economic opportunities as well as poverty, the specific case of India should be no different. Tilak (2002) showed how educational poverty, measured as low levels of educational attainment, is one of the main causes (if not the sole cause) of income poverty. Even if income poverty amplifies educational poverty, the positive association between them is indisputable. Educational poverty is also a primary cause of capability poverty in Sen’s (1993) sense of the term. For India, Duraisamy (2002) showed that the educational premium is rather weak for low levels of education (primary), whereas it is significantly strong for higher levels (secondary and tertiary). They documented a decrease in the wage premium for primary education and an increase for higher levels during the period 1983–1994. This trend intensified after the 1991 economic reforms, and income/wage inequality has skyrocketed ever since. Kijima (2006) showed that this trend results mainly from rising skills returns and the associated increase in demand for skilled labor. Chakraborty and Bakshi (2016) confirmed these findings by showing that, on average, not learning English in primary school reduces weekly wages by 68%. Using different datasets and alternative measurements, Tilak (2007) found the same results. Based on this rich literature, we assume that, in the Indian context, lacking education is associated with poverty and that higher education opens up economic opportunities.
13Education is the main driver of poverty and economic opportunities, but their intensity evolves in opposite directions as educational attainment increases, meaning that poverty remains severe at low educational levels, whereas increasing education strongly reduces the burden of poverty. However, this does not greatly expand economic opportunities, which require increased education if they are to increase significantly. For sufficiently educated individuals, the burden of poverty is weak. Our argument finds strong support in Chatterjee et al. (2018), who documented a U-shaped relationship between women’s education and labor force participation in India. One main reason for this empirical regularity is that increasing education among less educated women empowers them to marry into wealthier families, thus allowing them to leave the labor force. Above a certain educational threshold, this effect becomes weak and is dominated by an increase in economic opportunities.
14Consequently, increasing educational attainment among women with low levels of education tends to reduce the probability of poverty-driven childlessness without significantly increasing the probability of opportunity-driven childlessness. On the other hand, the risk of deprivations leading to infertility is already minimal among women with more education, such that increased education translates only into better economic opportunities and thus a higher probability of childlessness. Finally, women with an intermediate education are likely to face the lowest probability of ending their reproductive lives childless, as they are protected against extreme poverty but do not enjoy the greatest number of economic opportunities. This part of our theory assumes that women’s education has a U-shaped effect on the probability of childlessness.
Complementary mechanisms: male education, caste, religion, and state
15Our main prediction may be mitigated not only by male education and the Indian caste system but also by India’s geographical, institutional, and religious diversity. The husband’s education goes beyond reducing poverty by altering the couple’s appreciation of economic opportunities offered to the wife. Indeed, the husband’s income reduces the wife’s relative opportunity costs of child-rearing, as the couple’s relative loss is less when the husband also has a high salary.  We therefore assume that a husband’s education reduces a highly educated woman’s incentives to postpone births, which in turn reduces the probability of remaining childless. Accordingly, we assume that, overall, men’s education reduces women’s probability of being childless.
16The husband’s and wife’s relative education levels may also be determinants, given that higher education levels are linked to higher aspirations, as explained by Surkyn and Lesthaeghe (2004). As women obtain more education and economic opportunities, their status within the couple improves, thus granting them stronger negotiation power. Following Chiappori (1988), this means that the family’s objectives are more aligned with the wife’s aspirations, and it relates to childlessness as the wife’s fertility preferences take on greater importance when deciding whether to have children. Nevertheless, the direction of the effect remains ambiguous because we cannot assume either that an Indian woman invariably wants fewer children than her husband or that she more often wants to postpone her first birth. Symmetrically, if a man is less inclined than his wife to adopt a western lifestyle, having more education reinforces his negotiation power and thus reduces the probability of the wife not having children. We know that an effect transiting through negotiation power exists, but we cannot formally identify its sign (positive or negative). Nevertheless, these are always identified as second-order effects in the economic literature on negotiation power within couples (Chiappori and Donni, 2011).
17India is a large, culturally and institutionally heterogeneous country made diverse by its 29 official languages,  the caste system, and the coexistence of all major world religions. Studies have documented the country’s North–South divide regarding: (a) openness to fertility change (Dommaraju and Agadjanian, 2009); (b) different fertility outcomes based on education level and religious affiliation (Kulkarni and Alagarajan, 2005); and (c) differentials in maternal health care use based on caste differences (Kumar and Gupta, 2015). Even the implementation of national policies, such as education policies (Chakraborty and Bakshi, 2016), differs by state. State diversity is also driven by biogeographic factors like climate, pollution intensity, types of natural resources, etc.
18Based on these empirical regularities, we identify three ways in which state-level diversity may influence the probability of being childless. First, state specificities contribute to the formation of reproductive norms regarding ideal family size, childlessness, and the status of childless women.  Second, state specificities and institutions directly influence the reproductive conditions women face, such as malnutrition, sanitation, protection against venereal diseases, childbirth conditions, access to modern ART, etc. Third, states have the power to change some aspects of educational policies.
19Despite its economic development and improved economic opportunities for women, India is a patrilineal society in which women still experience tremendous pressure to bear a child soon after marriage. In some states, cultural barriers may annihilate the positive impact that better economic opportunities have on the probability of remaining childless.
20Religious affiliation is another potential factor influencing the probability of remaining childless at the end of reproductive life. While Koropeckyj-Cox and Pendell (2007) found that religious affiliation has an effect on attitudes and intentions toward childlessness, Ram (2005) identified its impact on definitive childlessness rates. We therefore hypothesize that religious groups may have specific attitudes toward childlessness. Using the theoretical framework of Goldscheider and Uhlenberg (1969), we assume that religious differentials in fertility and childlessness may be driven by the minority status and pronatalist values of religious groups like Catholics and Muslims.
21Another important dimension linking Indian culture with social structures is the division of society into castes. Post-independence India aimed to recognize and elevate the marginalized sections of society by establishing in the Constitution scheduled castes (SC), scheduled tribes (ST), other backward classes (OBC), and others. The SCs and the STs (lowest in the caste hierarchy) were historically the most deprived of certain basic human rights, lived in extreme poverty, suffered malnutrition, and were socially excluded. The OBCs are also historically deprived castes, while the general castes comprise all the other upper castes. Several studies have shown that caste remains a strong factor in Indian society, and women from the SC and ST groups often bear the greatest burden of social exclusion in terms of educational exclusion, poverty, the lowest use of maternal health care (Kumar and Gupta, 2015), lack of occupational mobility across generations (Banerji, 2012), and high fertility outcomes (Ramesh, 2014). We thus assume that belonging to a caste directly impacts male and female educational attainment and poverty status because the burden of poverty is heavier on the lowest castes with low education and disappears with higher levels of education. We then assume that women belonging to SC and ST groups may suffer higher probabilities of childlessness than others.
II – Data and descriptive statistics
1 – DLHS data
22We use secondary data from all three rounds of the District-Level Household and Facility Survey ([DLHS] 1998–1999, 2004–2005, and 2007–2008). The DLHS was conducted by Mumbai’s International Institute for Population Studies, funded by the Ministry of Health and Family Welfare, and it provides cross-sectional micro-level data that covers all districts and states in the country. DLHS data guarantee a large sample size and wide coverage of the Indian population over a long time span, while additionally providing sufficient information on childlessness, marriage, fertility, and other sociodemographic characteristics for all respondent categories.  Because the profession has raised some issues about DLHS data quality and suggests that Demographic and Health Survey (DHS) data are superior to DLHS, we also use the DHS 4 data (2015–2016) to test the external validity of our findings in Section V.
23After deleting missing values and input errors, then grouping women into birth cohorts and selecting only age groups of 40 to 49 years old,  we arrive at a final sample size of 158,112 currently married women born between 1953 and 1968, among whom 4,725 are childless.  We then obtain a childlessness rate equal to 2.9% against the 6.69% measured by the Census of India. A large portion of this difference arises from the alternative samples considered because our subsample includes only currently married women, while the census considers all women, including those who have never married or are divorced, widowed, or separated. Childlessness rates rise to over 90% among never-married women in India; among ever-married women, they are much higher for non-currently than currently married women. Reassuringly, DLHS and DHS estimates deliver very comparable childlessness rates among women by marital status. Asymmetries in birth omissions may explain the remaining differences between census and survey data. These two issues are discussed in detail in Online Appendix A: https://doi.org/10.34847/nkl.ab72r44w.
2 – Fertility and childlessness patterns in India
24The literature has extensively documented a negative relationship between a mother’s fertility and income, and thus a positive relationship between the degree of poverty and fertility (Birdsall et al., 2001). Therefore, if childlessness results mainly from poverty or infertility, either no correlation or a positive one should exist between childlessness rates and the average completed fertility of women in Indian states. Only sterility matters if no correlation exists, while, if positive, poorer states should have both higher childlessness and higher fertility. Using state-level data (Figure 1A), we show that the correlation between childlessness rates and the average completed fertility of women is negative. This contradicts the idea that childlessness in India is due only to infertility or poverty, and it indicates the possible existence of opportunity-driven childlessness. However, this does not mean that childlessness is only opportunity driven in India; outliers like the states of Haryana and Nagaland indicate that both childlessness and the average completed fertility of women can be high, which is also indicative of a potentially high prevalence of poverty-driven childlessness. 
Figure 1 - Correlations between childlessness rates and (a) average completed fertility and (b) educational attainment
Figure 1 - Correlations between childlessness rates and (a) average completed fertility and (b) educational attainmentNotes: Women born between 1939 and 1968. All states are included, but union territories are excluded.
3 – The education gradient
25In our subsample, 29% of women never went to school, 24.8% received primary education, 38.6% secondary education,  and less than 10% have at least some years of postsecondary education. Figure 1B indicates that childlessness at the country level exhibits a J-shaped relationship with years of education, as childlessness increases after 9–11 years of schooling. This becomes more salient when focusing on the youngest cohorts.
26In a cross-state perspective, childlessness rates exhibit only a weak linear correlation of 0.09 with average education when considering all states. However, this correlation becomes clearly positive and equals 0.35 when we exclude states representing less than 1% of the DLHS sample (Chandigarh, Uttarakhand, Delhi, Sikkim, Tripura, and Goa), which we do because they exhibit relatively extreme percentages and create outliers because of their small size. This suggests that states with high levels of education (presumably the most developed ones) also have higher childlessness, while those with low education levels have lower childlessness rates. A similar pattern can be seen at the district level, where childlessness is higher among highly educated (graduate and above) than less educated women in all districts of India (see Online Appendix Figure C1). Though the existence of opportunity-driven childlessness can be expected from the above descriptive findings, it remains unclear whether the effect of education is robust to other socioeconomic effects among childless women. The next section dispels this doubt using multivariate regression models.
III – Education and childlessness at the individual level
27In this section, we go beyond descriptive statistics and identify the main individual-level determinants of a woman’s probability of ending her reproductive life childless.
1 – Methodology
28We use logistic regression models to study the determinants of a woman’s probability of ending her reproductive life childless, for which we use information about completed fertility to build the dichotomous variable “childlessness.” It takes a value of 1 if the respondent has no children and 0 otherwise. From our theoretical hypothesis (illustrated in Appendix Figure A.1), it appears that the relationship between education and causes of childlessness may be confounded by at least the caste system, spatial diversity, and cohort. Following the methodology of Wunsch (2007), we must control for these elements when estimating the relationship between education and childlessness.
29We consider two kinds of fixed effects: cohort and state. Eight cohorts were compiled for the study; the oldest women were born in 1953–1954 and the youngest in 1967–1968. The state fixed effects in the models consider all 35 states and union territories in India. 
30To take into account potential confounding factors and control variables, we have introduced independent variables in a stepwise manner. In the first step, we group the woman’s education level into four categories: no education, primary, secondary, and higher. In the second step, we do the same for the husband’s education. Although not made explicit in our causal diagram, we control for the variable “minor marriage,” which takes a value of 1 if the woman married before age 18 (the legal marriageable age in India). This allows us to control for prolonged exposure to the risk of pregnancy without introducing confounding variables (see Section III.2 for details). In the third step, we add cultural variables for religious affiliation (Hindu, Muslim, Christian, Sikh, Buddhist, or other) and caste category (Scheduled Caste, Scheduled Tribe, Other Backward Class, General, or Other). Instead of using state fixed effects in the fourth step, we add the variable “state development level” to categorize states into the most, least, and intermediately developed, based on their average years of schooling among women (averages are computed from the 2011 census). States where women average more than 7.4 years are categorized as developed, those where they average less than 5.1 years as least developed, and all others as intermediate states.
31Importantly, observations in our main regressions are not weighted with the sample weights offered by DLHS due to the difficulty of gathering information on how weights were computed in the first two waves. For this reason, we suspect that the comparability of data between waves cannot be guaranteed using weights. 
2 – The U-shaped effect of female education on childlessness
32Model 1 (Table 1) exhibits a U-shaped relationship between educational attainment and the probability of being childless at age 40+ after controlling for cohort and state fixed effects. This result still holds in Models 2 to 4 and confirms the main prediction of our theory. 
33Model 2 shows that the higher the husband’s education, the lower the probability of the woman remaining childless, with a significantly and strongly negative gradient. This means that the husband’s education clearly plays the role of insurance against poverty-driven childlessness among low-educated women, while it reduces the opportunity cost of having children for highly educated women.
34As discussed in Section I.2, fertility decisions are negotiated by both the wife and the husband. The negative effect of male education on female childlessness may be due to negotiation power reallocated in favor of men. Generally, men bear much lower child-rearing costs than women, making them less prone to not having children and so making them negotiate in favor of parenthood.
Table 1 - Determinants of the probability of ending reproductive life childless
Table 1 - Determinants of the probability of ending reproductive life childless***p < .01. **p < .05. *p < .10.
Note: Odds ratios reported.
35For a childhood or teenage marriage, the greater exposure time to the risk of conception multiplies by 2 the chance of not being childless compared to a woman who marries later. Furthermore, being in a traditional arranged marriage may also increase the incentive or family pressure to conform to the traditional family system, which views having children as compulsory. In the DLHS data, the correlation between the respondent’s education and age at marriage is positive, but this does not prevent education from exerting a U-shaped influence on the probability of remaining childless.
36One may wonder why we have not divided age at marriage into more groups to capture late entry into marriage, as this could be important, considering that women who do not want to have children may also delay marriage as much as possible. We test this alternative in Online Appendix Table D1 and show that our results improve in terms of goodness of fit. Indeed, women who marry later are likelier to remain childless. That said, including this finer categorization for age of entry into marriage erases the U-shaped relationship between education and childlessness, only to replace it with a decreasing relationship, as having children and marrying late are both influenced by the economic opportunities offered to women. Thus, endogeneity issues arise when using age at marriage as a determinant of childlessness, making any result potentially spurious. This last criticism does not apply when considering only minor marriage because this kind of marriage is decided by the family and not the respondent herself. 
37The possibility remains that reverse causality exists between education and childlessness, as the women in these cohorts who did not want children may have focused on studying. However, even if the causality is reversed, both mechanisms nevertheless refer to opportunity-driven childlessness because not wanting children may lead a woman to avoid having them by focusing on education and job opportunities.
38In Model 3, caste has only a limited effect, as the probability of remaining childless is higher only among the STs when comparing them and all others to the general caste. This result may be initially surprising, but less so upon recalling that we control for state fixed effects. Indeed, discrimination against the lower castes has decreased over time and differs in space, which our fixed effects capture. 
39We find that both Christians and Muslims are less likely to remain childless than Hindus. As explained in Section I.2, which describes our theory, this could be due to pronatalist aspects of these religions and to their minority status in the country. The fourth model adds the state development level, which describes the state-specific average years of schooling among women. To avoid introducing multicollinearity, we eliminated the state fixed effect from that model. The finding is in line with the macro model of Poston and Trent (1982). At the micro level, education itself has an impact on the probability of being childless, and this effect is reinforced by a macro effect of development, proxied by average education at the state level.
40Compared to a woman with some years of secondary education, a woman having never gone to school is 1.21 times more likely to finish her fertile life childless, and a woman with some years of postsecondary education is 1.38 times more likely (Model 3). 
IV – Identification checks
41Following Wunsch (2007) and Heckman (2008), we know that the nonlinear effect of education on the probability of remaining childless does not constitute irrefutable proof that poverty explains the decreasing part of this relationship, nor is the increasing part explained by better economic opportunities. Indeed, some underlying factors may influence both education and childlessness in a way our theory fails to consider. To explore this possibility and progress toward a causality analysis, we use some unique features of the third wave of the DLHS to conduct identification checks.
42The ideal dataset would offer a precise measure of both poverty and economic opportunities offered to women for each wave of observations. This is not the case for the DLHS, as only Wave 3 offers these two measures. We therefore must focus on this wave and restrict our analysis to the 51,709 women who answered the question about their economic activity during the year preceding the survey.  We start with Model A (Table 2), where education is absent, and we introduce a relative measure of wealth/poverty that recodes the quintiles of Bassani et al.’s (2014) wealth index into three groups. 
Table 2 - Determinants of the probability of ending reproductive life childless using direct measures of poverty and economic opportunities
Table 2 - Determinants of the probability of ending reproductive life childless using direct measures of poverty and economic opportunities***p < .01, **p < .05, *p < .10.
Note: Odds ratios reported.
43We obtain that poor women have a much higher probability of being childless than others. Interestingly, poverty continues to have a positive effect after reintroducing the respondent’s education into this model, but low education fails to recover its negative effect, thus indicating that it is low educational attainment that captures poverty’s effect on childlessness in our main regressions.
44Model C inspects the effect of the respondent’s economic opportunities on the probability of remaining childless. Wave 3 of the DLHS offers a unique variable called “occupation,” which accounts for about 97 occupation categories. However, we divide this variable into five categories: no occupation, laborers, low skilled, medium skilled, and high skilled.  Before commenting on our results, let us recall that occupation is measured at the time of the survey and is thus only a proxy for the economic opportunities offered to a woman throughout her life. Therefore, birth history may have influenced the occupational history of these women, even though present economic occupation is strongly linked to past economic opportunities. This is also true for our poverty measure.
45We obtain that women who enjoyed very good opportunities are more childless than others, a first result that also confirms our theoretical model. Model D directly compares the effects of poverty and of economic opportunities, with the result that both variables have a positive effect on childlessness, as predicted by our theory. The poorer the person, the more chances she has of remaining childless; while the more favorable economic opportunities she enjoyed, also the more chances of remaining childless.
46Model E reintroduces the respondent’s educational attainment, showing that poverty continues to exert a negative effect on the probability of remaining childless, while education has a purely positive effect on it. Inversely, this purely positive effect of education eliminates the positive effects of economic opportunities. Stated differently, high education levels and highly skilled occupations capture the same effect.
V – Confirmation of results from DHS 4
47This section confirms the previous results obtained in our identification checks using the fourth wave of the Indian DHS, produced by the International Institute for Population Sciences. This wave was the first time the survey covered all districts in India, and it provides information on education, marriage, family, and health, among others.
48The information we extracted is comparable to what we used in previous sections, thus allowing us to measure childlessness and state of residence in the same way. Education is categorized into six levels: no education, incomplete primary, complete primary, incomplete secondary, complete secondary, and higher. We built a cohort variable that assigns respondents to 2-year birth cohorts from 1967 to 1975. DHS 4 proposes five wealth quintiles, while we divide women’s occupations the same way as in Section IV, but with an additional unemployed category for women who declared being new job seekers. Our results again control for caste, minor marriage, and religion.
49Models A and B (Table 3) both test our theory’s main prediction that education exerts a U-shaped influence on Indian women’s probability of ending reproductive life childless. Model A selects observations just as our main models in Table 1 do. Remarkably, the U-shaped relationship appears among the most recent generations when increasing the number of educational groups to five instead of three. Model B limits the analysis to women for whom occupational information is available, which severely reduces the number of observations but allows us to perform identification checks like those in Section IV. The results from Model A and from Model B are qualitatively the same, although only the no education and higher education categories become significant after reducing the number of observations.
Table 3 - Determinants of the probability of ending reproductive life childless, DHS 4
Table 3 - Determinants of the probability of ending reproductive life childless, DHS 4***p < .01. **p < .05. *p < .10.
Notes: Odds ratios reported. Model B, C, and D are limited to women for whom occupational information is available.
50Interestingly, introducing the DHS relative wealth index leads to a loss of all significance for education’s negative impact on less educated women’s probability of remaining childless. This result faithfully reflects the one obtained with the DLHS data in the previous section. Furthermore, introducing the variable for the respondent’s occupational status has no effect on the probability, while the positive effect of education persists among highly educated persons. Again, this result faithfully reflects the one obtained with DLHS data.
51Although the Indian childlessness rate is not the highest in the world, childlessness nevertheless concerns more than 12 million women over age 40—clearly an issue in India, yet it remains underrated and insufficiently explored. We have extended Baudin et al.’s (2015) theoretical framework to the Indian context under the hypothesis that the dynamics of childlessness in India is governed by the opposition between women’s poverty and their economic opportunities. In proxying for these two drivers with educational attainment, we expected the relationship between education and childlessness at the individual level to be U-shaped. This result has been confirmed by every robustness check we performed, by our identification strategy, and by the variables we controlled for. We have provided evidence that the husbands’ education protects against poverty-driven childlessness and that assortative mating does not confound the U-shape. This is a key insight from our paper. We have also confirmed that religious affiliation and caste continue to matter for childlessness.
52India is a highly complex country enriched by many cultures, institutions, and biogeographic diversity, which translates into a strong geographical gradient of childlessness and its uncertain future. The U-shaped relationship found between education and childlessness can be viewed as a clear sign that modern forms of marital arrangements are emerging and female empowerment has become rooted in education. Some might claim that very few highly educated women exist in India, but this is inaccurate. In highly developed states like Kerala, over 60% of women aged 40–50 have at least some years of high school; and 7.5% had some years of postsecondary education.  In Goa, we find that over 75% of women between 40 and 50 have at least completed high school, while over 15% have either a university degree or some years of postsecondary education.  These data may prefigure the future of education for Indian women, while the part of our U-shape that increases may prefigure the future of childlessness in India. That said, our results also suggest that the persistence of poverty, traditional values, and cultural habits (minor marriage, caste division, persistence of very high marriage rates, etc.) are major obstacles to the emancipation of women and the associated increase in childlessness. India’s characteristically consistent opposition between tradition and modernization will continue to play a key role in the dynamics of childlessness.
53Some aspects of our approach call for future research and caution. First of all, we obtained our results from a subset of currently married women, which entails an important assumption for India, as a non-negligible part of marriages are abandoned by husbands (Dommaraju, 2016). These marriages are more likely than others to be less fertile and, by extension, more childless. We discarded those because the accessible survey data indicated no date for when the husband left the household, thus making it difficult—if not impossible—to control for this event in a robust way. Keeping these women in our sample would have posed the risk of confounding the results, since the sooner a husband abandons the marriage, the greater the chances his wife will face poverty. Another specificity of our data is that they are provided by surveys, namely all DLHS rounds and the DHS. As we have discussed, Indian survey data tend to underestimate childlessness rates among women, while the country’s census data may overestimate them. Thus, further research is clearly needed.
AcknowledgmentsThis research received scientific and financial support from the ARC Project 15/19-063 on Family Transformations: Incentives and Norms. Thomas Baudin benefited from financial support from the French National Research Agency through the project MALYNES (no. ANR-18-CE26-0002). Koyel Sarkar also benefited financially from the Fonds de la Recherche Scientifique, Belgium, as an aspirant (2017–2021).We would like to thank Dudley Poston, Ester Rizzi, David de la Croix, Philippe Bocquier, Mikko Myrskyla, Akansha Singh, Li Ma, Malgorzata Mikckuka, and Robert Stelter for their helpful remarks. We are also grateful to the editorial board of Population and three anonymous referees for their constructive critiques, which have led to substantial improvements in our paper. In addition, we have benefited significantly from seminar and conference presentations in Rostock, Louvain-la-Neuve, Prague, Durbuy, and Cape Town. Any remaining errors and imprecisions are entirely our responsibility.
Figure A.1 - Causal diagram of childlessness
Figure A.1 - Causal diagram of childlessness
Data from the fourth wave of the Demographic and Health Survey in India (2015–2016) indicate that, among women who have children, 99.98% of them had their first child by age 40 (authors’ calculations).
Opportunity-driven childlessness cannot be identified as voluntary childlessness because trying but failing to have a first child at 38 is certainly not voluntarily choosing to be childless so much as it is to enter late in order to enjoy economic and other kinds of opportunities.
Attitudinal signs of modernization, compatible with modern forms of childlessness, can be detected in Indian society. Recent works point out the practice of modern, similarly British forms of marriage among ethnic groups in the Darjeeling hills of India as examples of developmental idealism (Allendorf, 2013; Allendorf and Pandian, 2016). Other works also show single-child families to be an emerging fertility trend in the country (Basu and Desai, 2016). Recent qualitative projects describe how certain metropolitan (Mumbai, Chennai, Vadodara, and Pune) working women prioritize their careers and personal aspirations over merely being mothers (Bhambhani and Inbanathan, 2018).
Stated differently, let us assume, first, that raising a child obligates a woman to spend 1 year out of the labor force and, second, that she earns 100,000 rupees per year. Having a child would cost 100,000 rupees, whatever the husband’s wages; but it would represent 10% of household income if the husband earns 900,000 rupees per year, while it would be 90% if the husband earns 11,111 rupees.
Our dataset provides no variables for the language spoken at home.
This is also true for districts, villages, and cities, although we lack data at these geographical scales.
Single women were not included in all rounds, and those that included them did not ask about their number of children.
This age group selection prevents selection bias due to cohort-based mortality after 50.
The totals of household and currently married women covered by the DLHS are, respectively, 529,817 and 474,463 in the first round, 620,107 and 507,622 in the second, and 720,320 and 548,780 in the third.
Even if we dispense with attitudinal concepts in our theory and analysis, attitudinal signs within DLHS data confirm the existence of childlessness not due to medical reasons. See Online Appendix B.
Having primary or secondary education does not mean here that a woman completed these, only that she had at least some years of the educational stage in question.
The term fixed effect refers here to dummy variables that control for birth cohort and respondent’s state of residence.
We have nevertheless used weighted rather than unweighted data to test all our regression models, finding a significant change only for the impact of being Christian compared to Hindu. This has no significance when using weighted data, but it does with unweighted data. Thus, we can reasonably declare that weighting issues are minor in our study. All results are available upon request.
Goodness of fit measured by adjusted count R2 is low at first sight because being childless is very rare in India and this limits the logistic model’s performance. This issue is discussed and a solution for fixing it is proposed in Online Appendix D.
We do not assume that all marriages over age 18 are not arranged, but they definitely are for under 18.
Suppressing state and cohort fixed effects reveals that women in the general caste are significantly less childless than women in any of the other castes.
As an alternative to state fixed effects, we tested models with state-level ecological variables, namely the average childlessness rate and average number of children per women over age 40. We find that higher state childlessness rates correlate with a greater individual probability of remaining childless. The opposite is true for average number of children among women over 40. These results show that state fixed effects may effectively control for cultural norms on reproduction and for other ecological differences like the prevalence of venereal diseases or other factors leading to subfecundity. Results are available upon request.
In every model in Table 2, we use all of our main model’s control variables from the previous section, except for husband’s education, to avoid strong collinearity with the other variables. All results remain valid when introducing husband’s education, but significance may change (available upon request).
We recoded Bassani et al.’s (2014) wealth index into three categories, combining poorest and second quintiles into poor, fourth and richest into rich, and third corresponds to middle. The original is a composite index that corresponds to a weighted mean of 14 elements. First, 12 binary items take a value of 1 for an affirmative answer and 0 otherwise: having a refrigerator, electric fan, pressure cooker, chairs, table, sewing machine, mobile phone, mattress, electricity, television, radio, and bed. Second, two scores: number of bedrooms (four modalities from 1 to 4+) and head of household’s highest education level, from 0 (no school) to 3 (higher than secondary). These result in a poverty score for each household, transposed into quintiles.
Classification details available upon request.
Data come from our sample.
The democratization of education can be assessed by comparing the educational attainment of our sample’s alternative age groups. For all India, 6.35% of women aged 40–50 attended a university, while this is true for 7.83% aged 30–40 and 8.97% aged 25–30. With some variance, this trend is observed in all Indian states.