1While a large body of literature has addressed fertility decline in India and its major states, no study has focused on fertility changes by parity during the course of the fertility transition. In addition, in comparison to China, whose fertility levels and trends have been the object of intense scrutiny and debate in the demographic literature, fertility levels and trends in India have not been examined in detail. It is generally accepted that fertility decline in India started in the 1970s, falling from about 5.5-6 children per women in the early 1970s to around half that level thirty years later (Bhat et al., 1984; Preston and Bhat, 1984; Bhat, 1998; Dyson, 2001; Visaria, 2004). Yet little is known about fertility changes by parity during the fertility transition, although as fertility declines one would expect the main changes to be parity-specific, with a reduction of higher-order births.

2This note analyses fertility changes by parity in India since 1977, taking advantage of nearly 300,000 birth histories collected in three nationally representative sample surveys, the National Family Health Surveys (NFHS). The aim of this study is thus not to discuss in depth the influence of population policy and family planning programmes on parity progression ratios in India, as has been done for China (Feeney and Wang, 1993), but rather to offer a detailed description of the changes affecting childbearing across the country. There are two reasons for this choice. First, since India currently accounts for about one sixth of humanity, the changes affecting parity-specific fertility are of great interest per se. Second, as Dyson (2002, p. 398) noted, “the control over family planning activities has increasingly become the province of individual state governments, rather than of the central government in Delhi”. This recent development makes a national level analysis of the influence of national family planning programmes more difficult to conduct for India as a whole.

3The findings presented here should nonetheless be interpreted in the light of developments in Indian population policies and family planning programmes presented elsewhere (Dyson, 2004; Srinivasan, 2006), as well as in the perspective of the major improvements in education, female autonomy, social and economic development, etc., that have occurred in India over the last 25 years (Dyson et al., 2004).

4To the best of the author’s knowledge, the only parity progression analysis of Indian fertility so far conducted at the country level using nationally representative survey data was based on the 1992-1993 National Family Health Survey (Gandotra et al., 1998). Besides this study, data from the NFHS-1 were used by Kulkarni and Choe (1998) to propose new parity progression-based measures of wanted and unwanted fertility in selected Indian States, and by Mutharayappa et al. (1997) to examine how son preference influences parity progression in India as a whole and in individual states. While these studies provide parity progression ratios at national and state levels, they do so only for the three-year period immediately preceding the survey. However, as we shall see, this choice of a three-year period before the survey is problematic in the case of India. This paper offers a longer perspective on fertility changes by parity over the last 25 years (i.e. from 1977 to 2004).

5The paper begins by presenting the data and the problems associated with the use of birth histories collected in the NFHS. It then describes the method used to assess the fertility changes by parity. Results are given for the progression from the first to the sixth birth and the impact of data quality is briefly discussed. Average lifetime parity is then computed and compared to fertility estimates based on the Sample Registration System (SRS) in order to assess the consistency of the results of the parity-based analysis, but also to consider the quality of the SRS estimates.

I – Data and problems in the use of birth history data collected in NFHS

6Data from the three NFHS were used. The NFHS are nationally representative sample surveys which, alongside a large number of other demographic and health-related questions, record complete birth-history data for all women aged 15-49. [1] Table 1 presents some characteristics of the NFHS samples analysed. [2]

Table 1

Sample characteristics of NFHS, India

Number of women NFHS-1 Apr. 1992 – Sept. 1993 NFHS-2 Nov. 1998 – March 1999 NFHS-3 Nov. 2005 – May 2006 Eligible, aged 15-49 96,981a 96,365 131,596 Interviewed, aged 15-49 89,777a 90,303 124,385 With multiple births 1,672 1,702 1,738 Included in the analysis 88,105 88,584 122,647 With at least one child 77,675 79,152 82,865 With at least two children 64,440 65,937 67,721 With at least three children 47,606 46,779 43,337 With at least four children 31,630 29,528 25,048 With at least five children 19,659 17,422 13,838 (a)Women aged 13-49.Sources: NFHS-1: IIPS, 1995 and computed from NFHS-1 data; NFHS-2: IIPS and ORC Macro, 2000 and computed from NFHS-2 data; NFHS-3: IIPS and Macro International, 2007a and computed from NFHS-3 data.

Sample characteristics of NFHS, India

7It has been shown that birth history data collected in NFHS present some problems for the estimation of fertility. [3] These problems may potentially affect the present parity-based analysis. Bhat (1995) has shown that large-scale misreporting of dates of birth, or of age of children, affected the first NFHS conducted in 1992-1993 (NFHS-1). In addition to the widely observed effect related to poor recall of events that occurred in the past, another problem arises from the structure of the birth history questionnaire and the tendency observed in the responses. It has been shown that birth history questionnaires beginning with a woman’s first birth result in the concentration of births in a period of about 5 to 15 years from the survey date. This effect is produced by the displacement of early births of older women to a date closer to the survey and the shifting of the most recent births backwards in time (Potter, 1977). The misreporting of the birth dates (the “Potter effect”) generally results in an overestimation of fertility decline in the recent past. Comparing fertility estimates computed from NFHS-1 and NFHS-2 data with SRS estimates, Retherford and Mishra (2001) have confirmed that this problem is recurrent in the second NFHS. They concluded that the total fertility rate (TFR) for the three-year period preceding the survey (an indicator commonly calculated using DHS datasets) cannot be computed accurately for India, resulting in considerable underestimation of total fertility. While no study has so far shown that the NFHS-3 presents the same problem, the same pattern is likely to be observed. Indeed, the birth-history questionnaires used in all three NFHS are identical and all begin with a woman’s first birth (IIPS, 1995, p. 358; IIPS and ORC Macro, 2000, p. 386; IIPS and Macro International, 2007b, p. 65).

8It is likely that parity progression ratios computed from NFHS birth history data will be biased by such effects. If so, it is expected that progressions to a given birth will decline in the years preceding the survey due to the displacement of children’s birth dates and that, for the same reason, birth intervals will increase. Likewise, the parity-based average number of children that is ultimately computed from the percentage distribution of women reaching a given parity will also show a decline in the pre-survey period.

II – Method: period parity progression ratios

9Studying fertility changes by parity provides important information because a woman’s decision to have an additional child is more likely to be based on the number of children she already has and the time elapsed since her previous birth (i.e. Coale’s conventional view of the fertility decline as a parity-dependent behaviour) than on her age alone. To assess the changes affecting fertility by parity over the last 25 years in India (from 1977 to 2004), period parity progression ratios (PPPRs) for synthetic cohorts were computed (Hinde, 1998). PPPRs were proposed by Feeney (1983; Feeney and Yu, 1987) and Ní Brochláin (1987) after the original formulation proposed by the French demographer Louis Henry in the mid-twentieth century (Henry, 1953). PPPRs have been applied in diverse contexts in order to understand fertility limiting behaviours (Feeney, 1991; Feeney and Wang, 1993; Feeney and Jianhua, 1994; Hosseini-Chavosi et al., 2006; Spoorenberg, 2009).

10To compute PPPRs for synthetic cohorts, women giving birth in a given year are followed “backwards” in time to the date of their previous child. The calculation of the PPPRs is based “on the (j + 1)th births occurring in a particular year to the women who had their jth births in a range of previous years” (Hinde, 1998, p. 114, emphasis added). The progression from the jth to the (j + 1)th birth, a_j, can be expressed as:

11

12where the series of q_x are proportions computed as the number of women who had their jth birth in the xth year before the current year and their (j + 1)th birth in the current year, divided by the total number of women who had a jth birth in the xth year before the current year minus the number of these women who have already had their (j + 1)th birth before the start of the current year. To compute these ratios, a number of years must be considered in order to go back in time as far as is necessary to capture the vast majority of women who will go on to have another child. As birth intervals of more than ten years are rare in most populations, a good rule of thumb is to go back ten years in the computation of the PPPRs (Hinde, 1998, p. 114). For the progression to the first birth, because childbearing at very young ages occurs in India, the year a woman reached her tenth birthday was selected as the preceding event and a period of 25 years back in time was selected in the computation of the PPPRs in order to include in the numerator all the women who will eventually go on to have a first child. The PPPR can be interpreted as the probability that a woman of parity i will move to parity i + 1 if she maintains the fertility level observed during the given year throughout her reproductive life. Formulas and a detailed example of the computation of PPPRs for synthetic cohorts are given in Hinde (1998, pp. 114-117).

13To assess further the changes in reproductive and family-building behaviours (spacing or stopping patterns), the mean birth intervals by single year are also computed. These intervals are calculated based on women entering the computation of the PPPRs (i.e. women who had their previous birth within a period of ten years before a given year).

14Lastly, from the full set of the period parity progression ratios, the average number of children ever born in a given year t, i.e. the synthetic lifetime average parity, P_t, is computed as the sum of the product of the proportion of women ending at a given parity P_i and their corresponding number of children N_i, using the formula:

15

16where i (from 0 to j) indicates the completed parities; and k indicates the open-ended parity group (in the case of India, women with 6 children or more). The average number of children observed among those women (N_k) in the survey is taken as an estimate of the number of children of women in the open-ended parity group.

17To assess the results of the parity-based analysis, the lifetime average parity, P_t, is compared to independently derived total fertility estimates. In addition, the synthetic lifetime average parity can also shed light on the consistency of other fertility estimates. In the case of India, fertility estimates from the SRS are compared with the results of the parity-based analysis.

18After excluding women who had multiple births (see Table 1 for numbers), who did not declare the year of birth for one of their children (17 women in NFHS-2), who had a child before age 10 (3 women in NFHS-1), who had birth intervals of less than 8 months, or who had an inconsistent birth history (birth date of a child of higher birth order prior to that of a child of lower birth order), the PPPRs were computed based on the birth histories of nearly 300,000 Indian women (299,336) (Table 1). As status variables collected in NFHS reflect a woman’s characteristics at the “survey time” and not at the time of childbearing, the PPPRs were not disaggregated by residence, educational levels, wealth, etc.

19PPPRs were computed for a 15-year period preceding each survey, excluding the survey years, in order to obtain completed annual birth histories. In other words, the results drawn from NFHS-1 (conducted in 1992-1993), from NFHS-2 (conducted in 1998-1999), and from NFHS-3 (conducted in 2005-2006) cover the following periods: 1977-1991, 1983-1997, and 1990-2004.

III – Period parity progression ratios in India, 1977-2004

20Figures 1 and 2 show PPPRs and the mean birth intervals between birth i and birth i + 1 for India from 1977 to 2004. The values are given in the Appendix Tables. The progression to the first birth is not broken down into the progressions from birth to first marriage and from first marriage to first birth (Dommaraju, 2009a, 2009b; Véron, 2008). Rather, progression from age 10 to the first birth is considered.

21Figure 1 shows that the fertility decline in India since the end of the 1970s is due mainly to a reduction in third and higher-order births. Progressions to birth orders higher than two fell by 25-35% between the late 1970s and the early 2000s. This indicates important changes in reproductive and family-building behaviours, with a growing number of Indian women deciding to stop their childbearing at two children. Indeed, stopping of childbearing has affected all parities above two. While in the late 1970s, 9 women out of 10 had a third child, only 6 out of 10 did so 25 years later. In comparison, progression to first and to second birth are both virtually constant over time, with a small decline in the early 2000s. However, this trend probably reflects a tempo effect, i.e. Indian women are postponing their first pregnancy.

22Mean birth intervals (Figure 2) confirm that tempo effects are affecting the progression to first and second births. The mean age at first birth increased by about 18 months between the early 1980s and the early 2000s and the mean interval between first and second births increased by about a half year during that period. By contrast, the mean birth intervals for higher-order births remained fairly stable, indicating that women who advanced to higher parities were not spacing their children differently than in the past.

23The PPPRs and the fertility estimates computed from NFHS data overlap, providing an opportunity to check data consistency. Inconsistencies are clearly visible in Figures 1 and 2 when periods overlap, probably due to the backward displacement of the most recent births to avoid the battery of questions on child health. The transfer of children to a higher age causes the PPPRs to drop artificially in the five years before a survey because fewer births are reported during that period. Likewise, the backward misreporting of children’s birth dates produces longer birth intervals. Moreover, there is a clear inconsistency in mean intervals from age 10 to first birth between NFHS-1 on the one hand, and NFHS-2 and NFHS-3 on the other.

24Nevertheless, Figures 1 and 2 also show consistency. Apart from the five-year period before each survey, the PPPRs and mean birth intervals are very similar in the remaining overlapping portion, indicating that, with the exception of the five years before a survey, the data quality of birth histories collected by the NFHS is good. Observed differences may be partly due to sampling variation, especially at higher parities where the number of births is smaller. The consistency of PPPRs and mean birth intervals in overlapping portions (excluding the five-year period before each survey) provides a first validity check on the results presented. Another useful check is to compare the average lifetime number of children ever born obtained from formula (2) with other estimates of total fertility.

Figure 1

Period parity progression ratios, India, 1977-2004

Period parity progression ratios, India, 1977-2004

Figure 2

Mean birth intervals, India, 1977-2004

Mean birth intervals, India, 1977-2004

IV – Period lifetime parity distribution and average lifetime parity

25Based on the PPPRs, the implied period completed parity distribution is used to compute lifetime average parity. The computed series of lifetime average parities is compared to other estimates of period total fertility. This validation is useful not only to check the consistency of the results given by the PPPRs, but also to assess the quality of the total fertility estimates from other sources.

26Table 2 presents the changes in the implied period completed parity distribution for selected years. When surveys are overlapping, the results of the most recent survey are taken. The changes are impressive. Over the last 25 years, the proportion of women with six or more children has fallen from 45% to 6.2%, while women having three or fewer children has almost tripled from 28% to 76%. In 1977, just 7.5% of Indian women had one or no child, compared with 20.7% in 2004.

Table 2

Implied period completed parity distribution and average lifetime parity, India, 1977-2004, selected years

Completed parity 1977 1981 1985 1989 1993 1997 2001 2004 NFHS-1 NFHS-1 NFHS-2 NFHS-2 NFHS-3 NFHS-3 NFHS-3 NFHS-3 0 4.1 3.4 4.1 3.7 5.8 7.3 8.8 10.9 1 3.4 5.7 4.8 5.3 4.1 6.9 10.3 9.8 2 7.5 13.8 11.6 17.6 16.2 23.5 31.7 33.3 3 13.3 14.0 20.6 23.5 20.0 24.3 22.5 22.3 4 13.6 17.9 20.1 17.9 15.8 15.2 12.1 11.0 5 12.9 13.0 12.4 12.1 10.0 8.4 7.2 6.4 6+ 45.2 32.2 26.2 19.9 28.1 14.4 7.5 6.2 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Average parity* 5.44 4.73 4.44 4.03 4.35 3.45 2.85 2.69 * Assuming an (observed) average parity for women with 6+ children of 8.11 children in NFHS-1, 8.06 children in NFHS-2, and 8.02 in NFHS-3. Source: Author’s calculations based on the three NFHS surveys.

Implied period completed parity distribution and average lifetime parity, India, 1977-2004, selected years

27To assess the quality of the PPPRs, the lifetime period average parity implied by these percentage distributions is computed and compared with existing total fertility estimates. Lifetime average parity and total fertility estimates from the Indian SRS, together with those published in the NFHS reports, are plotted on Figure 3. The lifetime average parity series is higher than the SRS estimates during the 1980s and 1990s. Most of this difference can be attributed to the fact that the SRS underestimates total fertility during the 1980s by about 7%, i.e. 0.3 children per women (Bhat, 2002, p. 133). Once the SRS total fertility estimates are adjusted by +0.3 children (Figure 3, adjusted TF (SRS)), they show a very similar pattern of fertility decline to that of estimated lifetime average parity during the 1980s and 1990s. In the most recent years, the two series deviate from each other. This finding can be attributed to two main factors: first, an improvement over time in the completeness of registration in the SRS; and second, a downward bias in lifetime average parities because of the displacement of children’s birth dates during the five years preceding the surveys. Thus, in Figure 3 the total fertility estimates published for NFHS-1, NFHS-2, and NFHS-3 are significantly lower than the others and the lifetime average parities are always lower in the years preceding a survey than the total fertility estimates computed for the same years from data of later surveys. This problem illustrates the “Potter effect” discussed earlier.

Figure 3

Lifetime average parity for synthetic cohorts (three-year moving average) and total fertility estimates

Lifetime average parity for synthetic cohorts (three-year moving average) and total fertility estimates

V – Discussion and conclusions

28Analysing the fertility transition in India using period parity progression ratios (PPPRs) for synthetic cohorts shows that the main cause of the fertility decline is a reduction in third and higher-order births. Figure 4 shows the trends observed in India since 1977 and decomposes the change in the lifetime average parity based on the PPPRs by successive birth order between 1977 and 2004. It shows that the progression to first birth alone would have reduced fertility by less than half a child from 5.4 to 5 children per woman. The changes in the progression from the first to second birth reduce fertility by an additional 0.3 children to 4.7 children per woman. The fall in the probability of having a third child has the largest impact on fertility, reducing it by 1.1 children per woman.

Figure 4

Decomposition of the 1977-2004 fertility decline in India by progression to each successive birth

Decomposition of the 1977-2004 fertility decline in India by progression to each successive birth

29Fertility changes at work in India indicate that a two-child family model is emerging in the country. According to NFHS data, 33% of Indian women had two children only (Table 2) and 54% had two or fewer children in 2004, compared with 7.5% and 15%, respectively, 25 years earlier in 1977. These figures provide strong evidence of the changes affecting reproductive and family-building behaviours over the course of fertility transition in India. That the two-child family model is emerging in national-level data should not come as a surprise. It has already been observed in the southern States which can be considered as socioeconomically more advanced and progressive, and which very often lead the way, providing indications of likely future demographic changes in the other Indian States and the country as a whole (Guilmoto and Rajan, 2001; Dyson, 2002). A study of fertility changes by parity disaggregated by region or State would probably reveal that the Southern States, such as Kerala and Tamil Nadu, where fertility levels are around or just below replacement level, exhibit distinctive reproductive and family-building behaviours in comparison to the northern and central States where fertility stands today at higher levels, with about 2.5-3 children per woman (IIPS and Macro International, 2007a).

30Use of the PPPRs in the analysis of Indian survey data yields fairly consistent fertility levels and trends in comparison to traditional total fertility estimates obtained from the SRS. The PPPRs present an alternative tool to estimate the consistency and quality of total fertility estimates derived from other data sources. This is true for India, but also for Iran (Hosseini-Chavosi et al., 2006) and Mongolia (Spoorenberg, 2009) where lifetime average parity estimates can be compared with several total fertility estimates (census, civil registration). For India, the PPPR-based lifetime parity estimates provides further proof that fertility in India during the 1980s and 1990s was higher than the levels shown by the SRS-based total fertility estimates. This finding corroborates previous research findings (Retherford and Mishra, 2001; Bhat, 2002). In addition, when excluding estimates for the five years before each survey, the overlapping portion of the estimates allows assessment of the quality of the SRS-based total fertility estimates during the 1990s. Figure 3 indicates that the quality of birth registration by the SRS improved during the 1990s. It is very likely, however, that the lifetime parity estimates for the year 2000 and later are biased downward and cannot be used as a basis for comparison.

31In describing fertility changes in India by using PPPRs, issues of data quality arise. They are not new and have already been reported in other studies. To address these issues, diverse proposals have been made on how to improve the training of interviewers and the data collection process in surveys (Arnold 1990; Pullum 2006). A frequent suggestion is to change the cut-off date that determines the administration of the extensive health questionnaire. Another possibility is to apply the extensive questionnaire to a subsample selected on a criterion other than the date of birth of children. The subsample to which the shorter questionnaire was applied could thus serve as a control to determine the extent of the shifting of children to older ages and its impact on demographic indicators.

32Lastly, regression techniques can also be used to model trends in the PPPRs, mean birth intervals, or lifetime average parity, and to correct such trends for the displacement of births. In regressions, fertility estimates could be weighted according to their consistency or some estimates could be excluded. This kind of approach seems promising since, as shown in the present paper, retrospective fertility estimates from successive surveys are generally consistent.