Skewed Sex Ratios at Birth and Future Marriage Squeeze in China and India, 2005–2100

Abstract

I examine the potential impact of the anticipated future marriage squeeze on nuptiality patterns in China and India during the twenty-first century. I use population projections from 2005 to 2100 based on three different scenarios for the sex ratio at birth (SRB). To counteract the limitations of cross-sectional methods commonly used to assess the severity of marriage squeezes, I use a two-sex cohort-based procedure to simulate marriage patterns over the twenty-first century based on the female dominance model. I also examine two more-flexible marriage functions to illustrate the potential impact of changes in marriage schedules as a response to the marriage squeeze. Longitudinal indicators of marriage squeeze indicate that the number of prospective grooms in both countries will exceed that of prospective brides by more 50% for three decades in the most favorable scenario. Rates of male bachelorhood will not peak before 2050, and the squeeze conditions will be felt several decades thereafter, even among cohorts unaffected by adverse SRB. If the SRB is allowed to return to normalcy by 2020, the proportion of men unmarried at age 50 is expected to rise to 15% in China by 2055 and to 10% in India by 2065. India suffers from the additional impact of a delayed fertility transition on its age structures.

Appendices

Appendix A: Parameters for Population Projection

Table 3 summarizes the parameters used for the projections of China’s and India’s populations until 2100 (see also Guilmoto 2010). Projections were done with the cohort-component method of the software package Spectrum (DemProj model), under the assumption of no international migration during the period under examination. Population figures and demographic parameters were borrowed from the medium variant of the world projections by the United Nations (2007). For the 2005 baseline year, population and demographic parameters come from the United Nations estimates, and these parameters incorporate in particular the current level of female excess mortality among children. However, two adjustments were necessary for China. First, the sex distribution derived from the 2005 1% survey—in which male adults are severely underestimated—was corrected by using the estimated sex ratio for the entire population in 2005 and the cohort-specific sex ratio for adults aged 15–54 during the 2000 census. Second, China’s fertility in 2005 was taken as 1.6 children per woman, a medium value derived from available TFR estimates by Cai (2008), Lutz et al. (2007), Retherford et al. (2005), and Zeng (2007). Values used for the 2050–2100 period were derived from two different sets of long-term projections prepared by IIASA (Lutz et al. 2008) and by the United Nations (2004).

Table 3 Parameters used for demographic projections: 2005–2100

The first two SRB scenarios described in the text are used to calculate the share of male and female births in each five-year birth cohort in 2005–2100. SRB figures used for 2005 (120 for China and 113 for India) are taken from the 2005 1% survey for China and from estimates by the Sample Registration System for India. The most recent estimate by China’s National Bureau of Statistics puts the SRB at 121 for 2008. It remains that SRB levels used here are only estimates in view of the possibility of underenumeration of girls in China and of sample issues in India.¹⁷ I have not attempted here to provide an alternative set of SRB estimates, and only the next round of censuses in both countries will provide the more detailed figures required to reestimate SRB levels. But in order to account for the possibility of female underregistration in China, I have also run a set of simulations with lower SRB (see Appendix C).

For the simulation based on normal SRB (with SRB constant at 105), I also corrected the 2005 baseline data for China and India because they are affected by past SRB imbalances. To do this, I modified the 2005 sex distribution of the population below 25 years in order to reflect a normal sex ratio distribution. For India, I computed correction factors for each five-year period since 1980 based on SRB estimates (see Kulkarni 2010), and I applied them to the 2005 sex ratio for all age groups below 25 years to compute the corrected sex and age distributions used in this projection. In the absence of a similar series for China, I used the average sex ratios of five-year age groups computed from United Nations estimates for the world after removing China and India.

Appendix B: Parameters and Procedure for Marriage Simulations

The nuptiality parameters are given in Table 4, starting with estimates for 2005. Because neither China nor India publishes marriage rates by age, I had to resort to indirect estimates based on successive age and sex distributions by marital status to compute the nuptiality table for 2000–2005 by five-year age group from age 15 to 50. The marriage schedules for China were computed longitudinally by combining the distributions from the 2000 census and the 2005 1% survey. Similarly, I used the two NFHS rounds of 1998–1999 and 2005–2006 (IIPS 2007) for India. The 2005 marriage rates from ages 15 to 50 have been used to compute the weighted adult sex ratio (computed on age distributions weighted by age-specific marriage rates) displayed in Figs. 1 and 2.¹⁸

Table 4 Nuptiality parameters used for marriage simulations: 2005–2100

To simulate marriages in five-year cohorts, I used the marriage probabilities derived from marriage schedules as well as a correction factor in order to account for remarriage (below age 50). This correction coefficient for remarriage was derived from two sources: the proportion remarried in the 2000 and 2005 Chinese surveys, and the proportion of remarried men and women in India according to the latest NFHS data for India. From the standard marriage schedules, I also derived other marriage schedules used in the models for the second half of the twenty-first century by increasing the proportion marrying after age 25 (women) and 30 (men). The marriage simulations are limited to unions occurring before the age of 50.

The simulation process starts from the 2005 five-year age distributions by sex and marital status. The number of expected first marriages for men and women in 2005–2010 is calculated from the marriage probabilities and the single population in 2005.¹⁹ The MSI (Figs. 1 and 2) is computed as the sex ratio of these expected first marriages after correction for remarriages. Because of the male preponderance, this ratio is usually above 100, and the final number of simulated first marriages is determined by using one of the marriage models described below. The number of men and women remaining single in 2010 is then computed for each cohort from the projected populations and the simulated numbers of first marriages during 2005–2010 according to each marriage function. With the computed 2010 distributions by age and marital status, the same procedure is then repeated successively until 2100, with specific nuptiality parameters for each five-year interval (Table 4).

This article uses the following marriage models:

• 1. In the female dominance (FD) model (Table 1), the number of simulated marriages is set by the number of expected female marriages. In the case of male surplus, male marriages are uniformly reduced across all age groups to match the available supply of brides. In this model, female nuptiality patterns are also allowed to change gradually over the period 2005–2050 (rise in age at first marriage in China and India, change of remarriage coefficients in India), but the difference in age at marriage between men and women remains constant throughout.

• 2. The delayed male marriage (DMM) model (Table 2) is a variant of the previous FD model, in which the male age at marriage is made to rise faster than the female age at marriage; the spousal age difference increases by two years in China and India from 2005 to 2050 and remains constant afterward.

• 3. In the harmonic mean (HM) model (Table 2), the simulated number of marriages is the harmonic mean of male and female expected marriages (Keilman 1985). With M representing the expected number of male marriages and F representing female marriages, the actual number of marriages, μ, is computed as . The adjustment ratio (simulated number / expected number) is applied to expected marriage rates by age and sex. The adjusted probability of marriage is capped to one in each age group.

Appendix C: Sensitivity Analysis of Female Birth Underreporting

As indicated previously, SRB levels used here are only estimates since data for China and India are not based on exhaustive civil registration sources; my simulation exercise remains therefore potentially sensitive to the quality of the estimates. To assess the sensitivity of the model, I ran a different set of population projections and marriage simulations based on lower SRB levels. I restricted the analysis to China, where SRB is both higher and more likely to be affected by estimation issues than India. In particular, Chinese fertility is notoriously underestimated by recent surveys and censuses, and female births might be more underreported than male births, resulting in exaggerated SRB estimates.

Based on this hypothesis, I postulate here that the real SRB is 113 in 2005, with male and female underreporting rates of 15% and 20%, respectively. This corresponds to an excess of 5% in underreporting rates for female births. Applying this rather high level of selective female underregistration will help to assess the sensitivity of the simulations. All other population and marriage parameters are kept similar to those of the rapid-transition scenario shown in Table 4.

Results in Table 5 demonstrate that a lower SRB contributes to reducing the male excess in the marriage market and to increasing the male probability to marry. However, the overall impact of lower SRB (113 in 2005) seems barely visible before the 2040s and remains modest thereafter. For instance, the MSI diminishes from 157 in 2040 to 151 according to the lower SRB scenario. The difference between the two series almost disappears after 2060. The estimated proportion of unmarried men at 50 also rises more slowly during 2050–2080 with these parameters, but the difference with the standard SRB scenario never exceeds 1%. This sensitivity analysis suggests therefore that SRB overestimation may exert but a minor effect on the simulation results.

Table 5 Marriage squeeze indicators and proportion of unmarried men at age 50 according to a lower SRB estimate, China: 2005–2100