In 1871, when India held its first census, there were 5.5 million fewer females than males.1 One hundred and twenty years later, in 1992, Amartya Sen estimated a deficit of 37 million females in India, drawing attention to the “missing women” of South Asia.2 India continues to be noticeably masculine. According to the most recent census, for every 1000 males, there were only 933 females,3 and the corresponding ratio for ages 0–6 years was 927 girls for every 1000 boys.4 The disproportionate distribution of sexes, at least in more recent years, has been surmised to be driven largely through the use of medical technologies by physicians and prospective parents to determine the sex of fetuses followed by selective abortion of female fetuses. This explanation was initially suggested in the 1980s,5^–7 and has gained considerable acceptance since then.8^–13 Some 10 million female fetuses are estimated to have been aborted over the last two decades in India.14

In response to this disconcerting trend, and after much public discussion, the Indian government enacted legislation in 1994 entitled the “Pre-Natal Diagnostic Techniques (PNDT) Act” to regulate and prevent the misuse of technologies for sex determination.15 The PNDT Act was implemented in 1996. Its scope was further expanded in 2003 with the prohibition of a whole range of activities that might facilitate deliberate sex selection (see box 1).15

While there has been substantial research on the issue of sex imbalance in India, systematic investigation of social patterning in the proportion of sexes is scarce. Typically, higher levels of socioeconomic status (SES) have been associated with improved health outcomes.16 However, within a context of substantial preference for sons17 and the presence of affordable and accessible technologies for fetal sex determination, one might posit that distribution of sexes would be relatively more disproportionate in higher SES groups compared with lower SES groups. With overall improvements in economic well-being that have occurred in recent years, it is also likely that SES differentials might be considerably weaker in recent times compared with earlier time periods. It is also not clear whether the PNDT Act achieved its desired goal. One would anticipate a more equal distribution of the sexes among newborns following the implementation of the Act if it were successful. Consequently, in this study, we examine three questions: (1) to what extent the odds of having a male infant vary by SES, measured through household income, parental education and social caste; (2) did the odds of having a male infant decrease during the period following the implementation of the PNDT Act?; and (3) has the association between SES and the odds of having a male infant changed in more recent years compared with earlier years?

METHODS

We used a nationally representative, population-based sample with time-series household survey data provided by the Indian National Sample Survey Organization (INSSO) for five most recent years: 2004/05, 1999/00, 1993/94, 1987/88 and 1983. INSSO uses a stratified multistage sampling design for selection of the units in urban as well as rural areas. The first-stage units were the census village for rural areas and the INSSO urban frame survey blocks for urban areas. Households were then randomly sampled from these first-stage units, within rural and urban areas.18 Heads of household were interviewed; data collected included counts of the children under age 1 year and their sex who were alive and living in the household at the time of each survey, as well as information on various socioeconomic and demographic variables. The availability of information on household socioeconomic status, especially on income, was an important strength of the INSSO data. For analyses pertaining to assessing the influence of the PNDT Act, data from 1983, 1987/88 and 1993/94 represented the “pre-PNDT” implementation period, while data from 1999/00 and 2004/05 represented the “post-PNDT” implementation period.

The outcome was a binary variable indicating the sex of the infant (1 = male, 0 = female). With the absence of reliable, national and public data on proportion of sexes at birth, proportion of sexes among infants was the only available reasonable proxy variable reflecting the sex differentials in births. We also calculated sex ratio, defined as the number of male infants divided by the number of female infants.

The predictor variables of interest were household income, parental education and social caste, all measuring aspects of SES, and a variable representing time periods before or after the implementation of the PNDT Act. We also included state of residence in our models to account for potential state variation in sex differentials among infants. Consumption-based expenditure data were used as a proxy for household income.19 20 Given the predominance of informal as well as a non-monetary or in-kind wage economy, household consumption expenditure is considered an appropriate proxy for income in developing economies.19 20 We categorised income in quartiles for analysis. Education was measured in terms of years of education for the head of the household (typically male) as well as the spouse of the head of the household (typically female). We grouped years of education using the following Indian educational benchmarks: no formal education; primary (⩽5 years); secondary (6–12 years); and post-secondary (⩾13 years). Social caste was based on the head of the household’s self-identification as belonging to one of the following groups: scheduled caste, scheduled tribe or other caste. “Scheduled caste” consists of groups that are lowest in the traditional Hindu caste hierarchy21 and, historically, have experienced significant social and economic exclusion and disadvantage. “Scheduled tribe” comprises approximately 700 tribes who tend to be geographically and socioeconomically isolated, with extreme levels of deprivation.22 “Other caste” is a residual category representing a higher status in the caste hierarchy. The “other backward class”, which comprises a diverse collection of intermediate castes, considered low in the caste hierarchy, but above scheduled castes, were also part of the “other caste” grouping. As “other backward class” was introduced as a category starting only in the early 1990s, in these analyses we consider “other caste” and “other backward class” as one group.

Other covariates included religion, place of residence and number of surviving children. Religious affiliation was categorised as Hindu, Muslim, Christian, Sikh and others. Place of residence was defined as urban or rural based on the household’s location in a census-defined urban or rural area. Finally, we included the number of children in the household as a covariate in adjusted analyses.

We used a binary logistic regression model, estimating the log odds of having a male infant. Specifically, multivariable models were used to assess the independent effects of the SES variables and the PNDT Act on the log odds of having a male infant, conditional on other variables. All the analyses were repeated for pre- and post-PNDT time periods.

RESULTS

The infant male:female ratio in the pooled sample was 1.08 (95% confidence interval (CI) 1.06 to 1.10) (table 1). Male:female ratio increased with income; it was 1.14 (95% CI 1.09 to 1.18) in the richest income quartile, whereas it was 1.04 (95% CI 1.01 to 1.08) in the poorest income quartile. Heads of household with post-secondary education had the highest male:female ratio (1.25, 95% CI 1.15 to 1.35) among the educational groups. Male:female ratios were not different across the social caste groups. Male:female ratio was higher in urban areas (1.11, 95% CI 1.08 to 1.14) than in rural areas (1.07, 95% CI 1.05 to 1.09). Male:female ratio was higher during the period following the implementation of the PNDT Act (1.10, 95% 1.07 to 1.13) compared with the period prior to the implementation of the PNDT Act (1.07, 95% 1.05 to 1.09). There were notable variations in male:female ratios by states, such that Punjab had the highest male:female ratio (1.37, 95% CI 1.26 to 1.50) while Karnataka had the lowest male:female ratio (0.95, 95% CI 0.8 to 1.03). Kerala, a state that has been considered a benchmark for gender equality, including in assessments based on proportion of sexes,2 10 had a male:female ratio of 1.11 (95% CI 1.01 to 1.21).

Table 2 presents the odds ratios (OR) with 95% CI of having a male infant for each covariate, while adjusting for all other covariates. Higher income was associated with increased ORs of having a male infant (p = 0.09). Compared with households in the poorest income quartile, the second richest and the richest had an increased OR of 1.05 (95% CI 1.00 to 1.10) and 1.06 (95% CI 1.01 to 1.12) of having a male infant. The ORs related to having a male infant in the second poorest quartile were not different from those in the poorest quartile. Households where the head had a post-secondary education had an OR of 1.15 (95% CI 1.04 to 1.27) compared with those with no formal education. The ORs of having a male infant in households where the head had a primary or secondary education were not different from those with no formal education. Education of the spouse of the household was not associated with odds of having a male infant (p = 0.44). The odds of having a male infant did not differ by social caste (p = 0.45).

Using the pre-PNDT period as the reference, the ORs associated with having a male infant were slightly higher in the post-PNDT period (OR 1.03, 95% CI 0.99 to 1.07, p = 0.09).

Place of residence (urban/rural) was not associated with the odds of having a male infant (p = 0.19). Christians had a lower OR of having a male infant compared with the reference group, Hindu (OR 0.89, 95% CI 0.81 to 0.99); other religion-based differentials in the ORs of having a male infant were not substantially different from Hindus.

State-level variations were observed in the odds of having a male infant (p = 0.01) (fig 1). Using the state of Kerala as a reference, the only state that had a higher OR of having a male infant was Punjab (OR 1.17, 95% CI 1.01 to 1.36). While Haryana and Gujarat had marginally increased ORs of having a male infant, compared with Kerala, these were not statistically significant (Haryana: OR 1.04, p = 0.59; Gujarat: OR 1.03, p = 0.66). The remaining states had similar odds of having a male infant when compared with Kerala, with the exception of Karnataka (OR 0.85, p = 0.01) and Himachal Pradesh (OR 0.87, p = 0.07).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Adjusted odds ratios and 95% confidence interval for having a male infant by state of residence. Note: Adjusted for income quartiles, education, caste, religion, place of residence, number of children and time fixed effects.

Table 3 presents the results from two multivariable regression models stratified by periods before and after the implementation of the PNDT Act. There was a positive association between income quartiles and the odds of having a male infant during the pre-PNDT period (p = 0.04). The OR of having a male infant was 1.09 (95% CI 1.02 to 1.17) in the richest quartile, followed by 1.07 (95% CI 1.01 to 1.14) in the second richest quartile, when compared with the poorest quartile. During the post-PNDT period, there was no association between income quartiles and the odds of having a male infant (p = 0.65). In the post-PNDT period, households where the head had a post-secondary education had increased OR of having a male infant (OR 1.21, 95% CI 1.04 to 1.42) when compared with households where the head had no formal education. In the pre-PNDT period, while those with a college education had an increased OR of having a male infant, this was not statistically significant (OR 1.10, p = 0.17). Other education-based differentials in the odds of having a male infant were not substantial in both pre- and post-PNDT periods. The only other noticeable differential related to the place of residence; in the post-PNDT period, households in urban areas were more likely to having a male infant than households in rural areas (OR 1.10, 95% CI 1.03 to 1.17).

Punjab had a substantially increased OR of having a male infant compared with Kerala (OR 1.30, 95% CI 1.06 to 1.72) in the post-PNDT period, followed by Rajasthan (OR 1.25, 95% 1.02 to 1.53). Haryana, Bihar, Gujarat and Tamil Nadu had increased ORs of having a male infant (OR range 1.16–1.06), compared with Kerala, but were not statistically significant (p range 0.17–0.57). For the remaining states, the odds of having a male infant were not different from that of Kerala.

DISCUSSION

A social analysis of the proportion of sexes among infants in India draws attention to three salient findings. First, the association between the odds of having a male infant and SES varied depending on the measure of SES as well as in relation to the period of observation. There was an association between income and the odds of having male infant, such that the odds of having a male infant increased with income quartiles. However, this association was weak and not substantial in the post-PNDT time period. There was a consistent and positive association between fathers/male head of the household with post-secondary education and the odds of having a male infant. The odds of having a male infant did not differ between caste groups. It thus appears that sex imbalance was particularly concentrated among the high SES groups, mainly along income and male educational dimensions, even though, in recent years, there seems to be substantial weakening of the income effect on the odds of having a male infant. The second major finding relates to the state variations in the odds of having a male infant, over and above those that can be attributed to individual demographic and socioeconomic characteristics. Notably, and contrary to expectation, Kerala is not particularly different from other Indian states in terms of its distribution of sexes among infants. The third finding relates to the lack of influence of the PNDT Act on the odds of having a male infant.

Before we elaborate on the above findings, it should be noted that the “normal” sex ratio at birth across human populations is understood to be somewhere between 105 and 107, with a median of 105.9.23 Even though about 5% more boys than girls are born everywhere, women are hardier than men and, given similar care, survive better at all ages, including in utero.24 Thus, the sex differentials in mortality by age 1 year across most countries are more trivial. Consequently, the male advantage in the sex ratio at birth is unlikely to account for the imbalances in infant sex ratio.

The finding that the concentration of males is greater among high SES groups has been suggested previously.9 11 14 For instance, mothers with higher education had significantly lower adjusted female:male ratios at birth (683, 95% CI 610 to 756) than mothers with no formal education (869, 95% CI 820 to 917).14 However, in our study, we did not find a statistically significant effect of the education of the spouse of the head of the household (typically female and possibly the mother). Regardless, the null finding for “female” education from our study and the positive finding from the earlier study14 are contrary to the conventional wisdom that has argued that improving maternal education is associated with lower levels of gender inequality among children. Our study had the added ability to examine the effect of the education of the head of the household (typically male and possibly the father) and found a positive association between education and the odds of having a male infant.

A positive association between SES and sex imbalance is paradoxical given that high SES has been consistently associated with improved health and other well-being outcomes.16 The combined effect of persistent and intense son preference,17 along with the increasing affordability and accessibility of technologies for sex determination,11 13 however, makes the concentration of male infants among the high SES group somewhat less counterintuitive. Higher incidence of dowry marriages and prevailing inheritance practices favouring sons, both more common among high SES groups, are key motivations for an intense preference for sons among high SES groups.25

Meanwhile, a son-preference norm could simply be “dormant” among low SES groups,26 ie, low SES households do not possess sufficient resources to practise discrimination. For instance, low SES households, with reduced access to the technologies available for fetal sex detection, may be less able to participate in the high SES group behaviours of fetal sex detection followed by selective abortion.27

The weakening of the income effect on the odds of having a male infant in the post-PNDT period (which is the period after 1999/00) seems to support this explanation. India has experienced a substantial improvement in economic well-being since the late 1990s, possibly making technologies available for fetal sex detection followed by selective abortion considerably more affordable.28 The higher OR of having a male infant in urban areas (which also have higher per capita income on average) compared with rural areas in the post-PNDT period suggests that the availability of fetal sex determination technologies are more concentrated in urban areas. Indeed, none of these patterns can be attributed to the implementation of the PNDT Act. Rather, it is suggestive of the impact of the overall changes that India experienced since 1999/2000 (among which rapid economic growth was a salient one) that changed the pattern of concentration of male infants in the high SES groups to a pattern where all groups had similar odds of having a male infant.

The finding that the proportion of sexes among infants in Kerala is no different from that in other Indian states in terms of its likelihood of having a male infant is in sharp contrast to previous interpretations. Indeed, Kerala has been argued to have a “normal” proportion of sexes that mirrors the distribution observed in the developed world.2 10 While Punjab’s grim position of a highly disproportionate distribution of sexes has been well recognised,15 27 our findings suggest that Kerala (and other Indian states) have no reasons to be cheerful either. This finding is in agreement with a previous study that reported a ratio of 765 females per 1000 males in Kerala for second-order births, conditional on the first birth being female.14 Meanwhile, female:male ratios for the second birth in Kerala, conditional on the first birth being male, were not substantially different, with almost all states (including Punjab) showing a female:male ratio >1000.14 Collectively, these findings suggest that disproportionate distribution of sexes is a concern afflicting all Indian states even though there are variations in the intensity of the sex imbalance.

Our findings suggest that the proportion of sexes among infants did not change after the implementation of the PNDT Act. Scepticism about the effects of well-intentioned policies in India is not uncommon, with the PNDT Act being no exception.29 However, the possible ineffectiveness of this policy needs to be interpreted within the larger socioeconomic and cultural context.

Specifically, parties seeking (ie, prospective parents) and parties offering (ie, physicians) the illegal service of determining the sex of a fetus have no incentive (other than moral conscience) to comply with the PNDT Act. Indeed, parents and physicians have incentives not to comply; for parents, it enables selection of the sex of their unborn child (social incentive for producing sons) and, for physicians, it enables an opportunity to provide a service for which there is substantial demand (economic incentive). It has been estimated that the business of sex determination followed by sex-selective abortion is worth at least US$100 million per year.11 Obviously, for the PNDT policy to work, compliance by medical professionals is absolutely critical. It has been argued that physicians in India have historically supported sex-selective abortions, and the evidence suggests that they continue to do so.11 30 If the current estimates of abortion of female fetuses is true (eg, 10 million in the last 20 years, or 1 million within the next 5 years),11 14 it is unlikely that this illegal process is being facilitated by a minority of errant physicians. At the same time, the role of physicians in facilitating sex-selective abortions needs to be evaluated via systematic empirical research.

To the extent that the PNDT Act has been put into operation, it has been, and continues to be, tardy.15 31 For instance, the PNDT Act requires the registration of ultrasound machines and these data are expected to be collected by states and reported to a central monitoring board. However, due to either shortages of personnel or lack of motivation, these data are rarely collected or reported in a timely and systematic manner,15 reflecting ineffective implementation of the PNDT Act.

Besides the influence of sex-selective abortion on infant sex ratio, the role of female infanticide in determining infant sex ratio cannot be ruled out; this is a phenomenon that has been documented in India for well over 100 years.32 In recent years, the phenomenon of female infanticide has not only been seen to persist in the states that had a historical precedence, such as Uttar Pradesh and Bihar,6 33 but has also been observed in other parts of India including Tamil Nadu.34^–36 While female infanticide was traditionally largely confined to the upper castes, the spread of the dowry custom, marginalisation of women in socioeconomic domains, as well as lower caste men emulating upper caste customs have been advanced as possible reasons for the spread of female infanticide.28 33 While not undermining the distressing practice of female infanticide, existing data seem to suggest that it is likely to be concentrated in certain groups and places and is unlikely to explain the overall sex imbalance that is observed across India.2 Furthermore, sex differentials in infant mortality in India are considerably narrower and even absent.37

As caveats to the study, we note the limitation of the study’s use of the infant sex ratio to assess the proportion of sexes as opposed to the more direct measure of proportion of sexes at birth. The inadequacy of vital statistics registration in India prohibits the availability of reliable and comparative figures on the proportion of sexes at birth.10 12 As a consequence, it was not possible to disentangle the unique role of sex-selective abortion from other practices such as female infanticide that could influence infant sex ratio in a population. The second limitation relates to the use of a sample survey as opposed to the census to study the proportion of sexes among infants. The Indian census, unfortunately, does not tabulate the population by sex for children under 1 year or by education of the parents. Furthermore, the Indian census does not ascertain income. Together, these limitations prohibited the use of census data for this analysis. At the same time, as we used a sample survey to analyse the sex ratio, inferences should be restricted to patterns and associations, and not to the precise magnitude of the sex ratio. Finally, our interpretation of the possible effect of the PNDT Act on the odds of having a male infant is restricted to the provisions of the Act that was enacted in 1994 and implemented in 1996. Whether the subsequent amendment in 2003 yields a different outcome could not be effectively tested in this study, as the most recent time period in our study was 2004, which might be too early to assess the impact of the expanded 2003 version of the PNDT Act on the proportion of sexes.

The above discussion raises a fundamental question: can public policy address the problem of sex imbalance in India? Ethicist Bernard Dickens states that, “attempts to end son preference by prohibition are failing in India, and appear too peripheral on their own to relieve sex bias”,38 and argues that prohibitions in countries such as India (with intense son preference) pose a far greater risk to women’s and girl children’s lives and their health,38 through maternal mortality, infanticide, malnutrition, neglect, low status and so on. It is true, as our findings suggest, that policy solutions that are asymmetrical to the social context within which they are intended to operate are likely to be ineffective. However, the alternative cannot be the laissez-faire approach that is sometimes suggested. For instance, it is argued that the shortage of prospective wives will eventually enhance the social and economic value of daughters, overturning their vulnerability and the force of male domination.38 Indeed, such an approach is perilous; it amounts to turning a blind eye or even justifying the lack of moral, social and legal responsibility of the medical establishment in contributing to this problem, while also rationalising violence again women.11

Abnormality in the proportion of sexes reflects failed human development at the most fundamental level. While the problem is chronic in countries such as India, China, South Korea, Taiwan and Hong Kong, the practice of sex selection is becoming increasingly prevalent in Vietnam, Azerbaijan, Armenia and Georgia,39 and has been reported in Nepal, Jordan, Bangladesh and Pakistan.11 The excess concentration of males has also been discussed as a threat to peace and security, both within countries40 and globally.41 Analyses of the social patterning in the proportion of sexes among infants in India suggest that improvements in socioeconomic circumstances or introducing policies that are not aligned with societal norms and preferences are unlikely to be effective instruments in normalising the imbalances in the distribution of sexes.