Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
9 The Relationship Between Infant and Child Mortality and Subsequent Fertility in Indonesia: 197 ~ - 199 ~ Elizabeth Frankenberg The relationship between infant and child mortality and subsequent fertility at both the aggregate and the familial level has long interested demographers, sociologists, and economists. The idea that mortality decline precipitates fertility decline has its origins in the historical demography of Europe and is a linchpin of demographic transition theory. At the aggregate level two questions have fo- cused on the demographic future of developing countries: (1) Would fertility declines accompany the mortality declines observed in developing countries in the mid-19OOs? (2) How would population growth rates change as mortality declined? Underlying these questions and stimulating interest at the family level is the notion that a couple's fertility is in part a product of the mortality environment in which they are building a family. Actual or expected child deaths prompt a fertility response, so as child deaths become more rare, fertility patterns change. Work on the link between mortality and fertility has sought to answer a variety of questions, some only tangentially related to the "bottom line" of how fertility levels and population growth rates change as mortality declines. Most theoretical models of fertility response make predictions about conscious behav- ioral responses to actual or expected child deaths, given assumptions about couples' family formation goals. A number of authors recognize the potential complexity of those goals when preferences for family size, sex composition, and timing are entertained (Preston, 1978; Ben-Porath, 1976; Wolpin, 1984~. The theoretical literature draws a distinction between strategies of hoarding and re- placement. Both strategies involve the attempt to achieve some target number of 316
ELIZABETH FRANKENBERG 317 surviving children. Hoarding is described as a fertility response to expected mortality, where women give birth to a larger number of children than they ultimately desire in expectation that some will die. Replacement is described as the strategy of having an additional child only in the event that a previous child dies. Other theoretical models are more concerned with how the biological processes surrounding childbearing are affected by a death in ways that in turn affect fertility (Preston, 1978~. Empirical work has focused both on identifying the types of responses to mortality and on quantifying the ultimate effect of child death on completed family size (Olsen, 1980; Mauskopf and Wallace, 1984; Mensch, 1985; Heer and Wu, 1978~. Fertility-related outcomes of interest include completed family size, parity progression ratios, interbirth intervals, and contraceptive use. The increas- ing availability of birth history data, which record the sequence and timing of a woman's births and the survival status of each birth, have fundamentally altered the scope for empirical work on the topic. In this chapter I focus on relationships at the family level in Indonesia, using data from the 1987 and 1991 Demographic and Health Surveys (DHSs). INDONESIA: A DEMOGRAPHIC AND SOCIOECONOMIC OVERVIEW Indonesia, with almost 200 million people inhabiting an equatorial belt of more than 15,000 islands, is the fourth most populous country in the world. The archipelago is home to a diverse group of cultures and ethnicities, including more than 300 distinct language groups. Indonesia has experienced rapid economic growth in the past 25 years, with per capita income rising at an annual rate of about 4.5 percent, from $50 in the late 1960s to $650 in 1992 (World Bank, 1994~. Accompanying this economic growth have been increases in access to schools and health services and, in turn, improvements in measures of human resources. For example, the number of people per physician declined from over 100,000 in the 1950s to around 13,000 in the 1980s (Hugo et al., 1987), while life expectancy at birth increased from 32 to around 53 over this period. Adult illiteracy declined from 61 percent in 1960 to 23 percent in 1990 as the number of primary, junior secondary, and senior secondary schools rose (World Bank, 1994; Hugo et al., 1987; World Bank, 1984~. Of more importance for this chapter than general socioeconomic trends are changes in demographic characteristics in Indonesia over time. Numbers for Indonesia as a whole are not available until the late 1960s, but some information about earlier periods is available for subregions. This discussion is taken from Nitisastro's comprehensive effort to document population change in Indonesia in the 1900s (Nitisastro, 1970~. Estimates of vital rates for Java in the 1930s are available from registration systems for some areas (predominantly urban), as well as from small-scale studies of infant mortality. Nitisastro assembles available
318 INDONESIA: 1971-1991 information and concludes that in the 1930s crude birth rates for Java were higher than 40 per 1,000, while life expectancy at birth was between 30 and 35 years, and infant mortality was between 225 and 250 per 1,000. Estimates of infant mortality rates on Sumatran plantations (where recording of births and deaths was believed to be relatively accurate) yield rates of 160-370 per 1,000. As early as the 1930s, programs relating to hygiene and malaria control began on Java, perhaps reducing mortality in some areas (Hull, 1989~. The 1940s were a period of extreme hardship throughout Indonesia, but especially on Java. The Japanese occupied Indonesia from 1942 to 1945. Be- tween 1945 and 1949 Indonesia fought a war of independence against the Dutch. The decade was characterized by severe population dislocations, food shortages, decreasing fertility, and increasing mortality. Examination of the age structure of the population recorded in a 1958 labor force survey and the 1961 census indi- cates that the number of survivors from births in the 1940s was much smaller than the number of survivors from births in the preceding and subsequent periods (Nitisastro,1970~. During the 1950s, the early years of independence, crude birth rates returned to their prewar levels of more than 40 per 1,000. Simultaneously, a fairly intensive public health campaign began and mortality declined. Esti- mates of infant mortality rates from Jakarta, Surabaya, and Wonosobo (a region of central Java) range from 148 to 178 per 1,000 a considerable decline relative to estimated rates of the 1930s. The 1971 census marks the beginning of the period for which estimates of fertility and mortality patterns are available for Indonesia as a whole, although differences across data sources in sample and content thwart precise documenta- tion of trends. Since the late 1960s both mortality and fertility levels have declined considerably in Indonesia. Scholars believe that in most of Indonesia, the mortality decline began in the 1950s, while fertility did not start to fall until the mid-1960s. Efforts to promote family planning in Indonesia began as early as the 1950s (Hugo et al., 1987~. In 1957 various local groups combined to form the Indonesian Planned Parenthood Organization, which sold contraceptives, invited family planning advisors from abroad to visit Indonesia, and in 1963 conducted a series of seminars on contraception for health professionals from Java and Bali (Hugo et al., 1987~. The activities of the organization were disrupted by the political events of the mid-1960s, but in 1970 the National Family Planning Coordinating Board (BKKBN) was formed with orders to report directly to Presi- dent Soeharto. Estimates of total fertility and infant mortality rates between 1967 and 1991 are summarized in Table 9-1. The table documents a decline in total fertility from 5.61 in the late 1960s to around 3.0 in 1991 a decline that began slowly but accelerated over time. Infant mortality has fallen from 143 per 1,000 in the late 1960s to around 70 per 1,000 in the early 1990s, but the pace of the decline is more erratic than the decline in fertility. The mid-1970s and the mid-1980s appear to be periods of little change in mortality.
ELIZABETH FRANKENBERG TABLE 9-1 Estimated Fertility and Infant Mortality Rates Post-1965 319 Total Infant Fertility Mortality Source Period Rate Period Rate 1971 census 1967- 1970 5.61 1968- 1996 143 1976 SUPAS 1971-1975 5.20 1972-1973 112 1980 census 1976-1979 4.68 1977-1978 112 1985 SUPAS 1980- 1985 4.06 1982- 1983 71 1991 DHS 1984-1987 3.73 1990 census 1986-1989 3.31 1987-1988 70 1991 DHS 1988-1991 3.02 1988-1991 68 1994 DHS 1991-1994 2.85 1991-1994 57 NOTE: SUPAS, intercensal population survey. In reviewing the demographic trends in Indonesia, some evidence supports the contention that mortality decline preceded fertility decline. It appears that mortality began to fall during the 1950s, partially in response to increases in food availability and in public health services, but that fertility did not decline until the mid-1960s. Once the fertility decline began, however, it did so at an accelerating pace, independent of the more erratic pattern of mortality decline. A similar pattern is observed in the historical data on Europe, where the most rapid declines in mortality and fertility appear to have occurred simultaneously (Matthiessen and McCann, 1978~. DATA In the past several decades Indonesia has made demographic and socioeco- nomic data collection a priority. The Indonesian Central Bureau of Statistics has implemented numerous special topic surveys, as well as conducting diennial censuses of the population and of levels of village infrastructure. Data for this chapter are drawn from the 1987 and 1991 Demographic and Health Surveys of Indonesia. These surveys are joint efforts of the Central Bureau of Statistics and Demographic and Health Surveys, with input from the Ministry of Health and from the BKKBN. The 1987 DHS collected information from 11,884 ever-married women between the ages of 15 and 49, living in 20 of Indonesia's 27 provinces. These data put the total fertility rate at 3.4 per woman and the infant mortality rate at 70.2 per 1,000 births for the 1982-1987 period (Central Bureau of Statistics et al., 1989~. The 1991 DHS collected information from 22,909 ever-married women between the ages of 15 and 49. The 1991 survey, which covered all 27 of Indonesia' s provinces, documents a fertility rate
320 TABLE 9-2 Vital Rates, 1984- 1986 INDONESIA: 1971-1991 Standard Standard Vital Rate 1987 Error 1991 Error Ratio z Score Fertility rates 15-19 74 10 95 8 0.82 3.15 20-24 180 15 193 12 0.99 1.45 25-29 167 16 179 12 1.03 1.28 30-34 126 16 134 12 0.97 0.85 35-39 70 13 81 11 0.97 1.21 40-44 30 45-49 10 TFR to 39 3.09 3.41 0.91 Mortality rates Infant mortality 66.8 7.6 75.7 6.4 0.88 1.86 Neonatal mortality 24.5 4.6 32.7 4.2 0.75 2.60 Postneonatal mortality 43.4 6.2 44.5 5.0 0.98 0.26 NOTE: TFR, total fertility rate. SOURCE: 1987 and 1991 Demographic and Health Surveys. Of 3.02 and an infant mortality rate of 68 per 1,000 in 5 years before the survey (Central Bureau of Statistics et al., 1992~.1 DHS data sets are almost uniformly regarded as of high quality, and the Indonesian data are no exception. Table 9-2 demonstrates the comparability of the 1987 and 1991 data. Estimates of age-specific and total fertility rates (up to age 39) are calculated for women in the 20 provinces surveyed in the 1987 data for the period 1984-1986. The 1991 data were reweighted to match the geo- graphic and age distributions prevailing in 1987 in the 20 provinces surveyed as part of the 1987 DHS. I also calculated standard errors, the ratio of the two estimates, and z-scores testing for significant differences between the estimates. These data provide an indication of the magnitude of the differences. The two surveys generate reasonably consistent estimates of age-specific fertility rates for the 1984-1986 period, although the total fertility rate from the 1991 data is higher than that from the 1987 data, in part because the 1991 esti- mates of fertility among those aged 15-19 is considerably higher than the esti- mates from the 1987 data (the z score indicates a significant difference between the two numbers). I also compared mortality rates estimated from the two sur 1The 1994 DHS Final Report became available as this chapter was being written. The 1994 DHS data put Indonesia's fertility rate at 2.85 for the 1991-1994 period and the infant mortality rate at 57 per 1,000 (Central Bureau of Statistics et al., 1995).
ELIZABETH FRANKENBERG 321 veys. Estimates of infant mortality for the 1984-1986 period are higher in the 1991 data (75.7 per 1,000) than in the 1987 data (66.8 per 1,000), although the difference is not significant. The difference stems from discrepancies in esti- mated neonatal mortality (a difference that is statistically significant). Postneo- natal mortality rates for the 1984-1986 period are very similar (43.4 per 1,000 from the 1987 data, 44.5 per 1,000 from the 1991 data). The DHS data contain birth histories, in which respondents provided infor- mation on the dates of birth and death for all their children. Considerable atten- tion has focused on how reliably these dates are reported and the extent to which missing data that require imputation procedures could bias estimates of demo- graphic parameters (Boerma et al., 1994; Sullivan et al., 1990~. The general conclusion of analyses of World Fertility Survey (WFS) and DHS data quality is that estimates of mortality, fertility, and durations of interbirth intervals, particu- larly within 15 years of the survey date, are not seriously biased by date mis- reporting (Sullivan et al., 1990; Lantz et al., 1992; Potter, 1988~. The Indonesia data contain cases in which children's month of birth is not reported by the mother. Failure to report a birthdate is more common when the child died than when the child survived. In the 1987 data, month of birth was imputed for 17.2 percent of the births in the 15 years before the survey. In the 1991 data, month of birth was imputed for 13.6 percent of births occurring in the 15 years before the . . Interview. DESCRIPTIVE STATISTICS Initial efforts to link infant and child mortality to fertility were based on comparisons of the number of children ever born (and children surviving) among women with different child mortality experiences (Olsen, 1980; Mauskopf and Wallace, 1984~. These comparisons were made in part because demographic data were often limited to a woman's report of the number of children ever born and number of children who died. Typically, correlations are positive for several reasons. These reasons highlight the statistical pitfalls associated with analyses of links between child mortality and subsequent fertility. Some of the reasons also explain positive correlations between child mortality and other aspects of reproductive patterns. First, the larger the number of children ever born, the more children who are exposed to the risk of death. Women who give birth to more children will on average have more children die, simply because of a binomial association (the larger the number of trials, the larger the number of deaths) (Olsen, 1980~. Second, if babies die before they are weaned, the cessation of breastfeeding may hasten the return of ovulation after a birth and thus hasten the conception of a subsequent infant (van Ginneken, 1974~. This phenomenon, sometimes re- ferred to as the physiological effect, has been recognized for more than a century
322 INDONESIA: 1971-1991 (Knodel, 1978~: This mechanism has the most scope for effect in settings where average breastfeeding durations are long. Third, women who have given birth to a large number of children are likely to have experienced reproductive patterns (e.g., teenage childbearing, late child- bearing, short birth intervals) that have been shown to be associated with in- creased risks of child mortality in a voluminous number of analyses. In other words, reproductive patterns that generate large numbers of births imply height- ened risk of death (Rinehart and Kols, 1984; Ross and Frankenberg, 1993~. Fourth, some women may have characteristics that predispose them to both high fertility and high mortality. For example, some women may be fatalistic about their ability to control their fertility and the health of their children, whereas others may actively seek out the technologies that change their fertility and their children's health. This difference in level of initiative may well drive differences in fertility and in child health outcomes and is likely to be correlated with charac- teristics such as education. In fact, some would argue that women are not predis- posed to high fertility and mortality, but rather that their allocation of resources reflects a choice of high fertility and mortality. This type of argument appears throughout the demographic and economic literature. Rodriguez et al. (1984) discuss the strong correlation among birth intervals for a given woman, indepen- dent of parity. Trussell et al. (1985) show that this correlation can be only partly explained by factors such as propensity to breastfeed or to use contraception. Pebley and Stupp (1987) discuss the fact that many of the factors associated with risk of child mortality are also associated with birth intervals. Fifth, women who experience child deaths may intentionally bear additional children in an effort to attain their fertility goals. The possibility that women consciously change their fertility patterns in response to actual or anticipated child mortality is of primary interest. Would a particular woman choose a differ- ent number of births or space her births differently if her perceptions of the mortality risks of her children were different or if one of her children died? Designing a statistical test that isolates a conscious response to actual or per- ceived risk of mortality is extremely difficult because of the reasons detailed above that contribute to positive associations between mortality and fertility. Data limitations also pose difficulties for empirical assessments of the de- gree to which the number of children ever born varies by mortality experience. The women of greatest interest are those who have completed their families and stopped having children. But fertility surveys typically collect data from women currently able to reproduce, who may or may not have finished bearing children. In addition, women who have finished having children tend to be older and may not reliably document births and deaths that occurred in the distant past. Differences in completed family size by mortality experience are one of several potential indicators of a fertility response to child mortality. Other mea- sures that are easier to work with, given the existence of surveys of women of reproductive age, are parity progression ratios and birth intervals. Parity progres
ELIZABETH FRANKENBERG 323 sion ratios are defined as the proportion of women at parity x who go on to have an x + 1 birth within a specified time period. Birth intervals are simply the time that elapses between two adjacent births. Because these indicators can be com- puted for women regardless of whether they have finished childbearing, they are well suited for use with demographic data. Comparison of parity progression ratios or birth intervals for women with differing child mortality experiences can indicate the degree to which very specific components of reproductive patterns vary by survival status, but say little about how levels of completed fertility are likely to change as mortality declines. In this section I present cross tabulations and graphs describing the relation- ship between infant and child mortality experience and fertility levels and pat- terns for three periods: 1971-1977, 1978-1984, and 1985-1991. The DHS data from Indonesia appear to be of relatively high quality with respect to the com- pleteness of reporting births, deaths, and associated dates. However, to the extent that dates are misreported or events are omitted in ways that are systematically related to the timing of fertility, the results in my descriptive tables could be affected. I begin with a graph evaluating the effect of an early child death on com- pleted family size in terms of the number children ever born and the number of surviving children. This graph derives from a method described by Wolpin (in this volume) in a recent review of the literature. The method was developed as a means of calculating directly the number of excess births that arise from a death. The method focuses on women at the end of their reproductive span, comparing numbers of children ever born with the number who survived for women whose reproductive histories are nearly identical to a point in time save for the survival status of one child. Any number of reproductive sequences can be compared. I focus on two simple patterns. First I identify women between the ages of 41 and 46 (in 1984 and 1991) who had a first child within the first 2 years of marriage. I stratify these women by the survival status of this first child at 2 years of marriage and compare the number of children ever born with the number of children surviving. The second comparison is based on a subset of women from the first comparison: those whose first child was still alive 2 years after the marriage and who had a second child between the second and fourth year of marriage. Women whose second child was alive 4 years after the marriage are compared with women whose second child did not survive to 4 years after the marriage. These comparisons are presented in Figure 9-1 and Figure 9-2. Figure 9-1 contains the statistics described above for women 41-46 years of age at two points: 1984 and 1991. Several points emerge from the graph. First, family sizes for these women have clearly declined over time. Among older women whose first child was alive after 2 years of marriage, the average number of children ever born is 6.05 in 1984. By 1991 mean children ever born has declined by 0.76 children, to 5.31. Declines of a similar magnitude are observed for women whose second child was alive after 4 years of marriage (see Figure 9
324 7.0 6~0 <~, 5tO . _ o E 4.0 3.0 2.0 1 ,0 -, 0.0 INDONESIA: 1971-1991 Decor, `84 Alive,'B4 Dead, '91 Alive, '91 Survival at the mother's marriage date ~ 2 years |C3CEB| lyrics 1 FIGURE 9-1 Children ever born (CEB) and children surviving (CS) by survival status of the first child for women aged 41-46 with a birth in the first 2 years of marriage. 21. Fertility declines over time are steeper for women whose first child survived to the date of interest than for women whose first child died. When the first or second child survived, the number of children ever born drops by 26 percent over the period. When the first or second child died the decline is around 19 percent. Second, the number of children ever born is always greater for women whose first or second child died by the requisite date than for women whose first or second child was still alive. The differences in children ever born by survival status of the first or second child actually grow over time. In 1984 women whose first child died have around 0.75 more children ever born than women whose first or second child survived. The difference is 1.04 for women whose second child died compared with women whose second child survived. By 1991 women whose first child died have 0.94 more children ever born than women whose first child survived, while women whose second child died have 1.13 more children ever born than women whose second child survived. From these numbers it appears that women who experience a child death also have more births, and that the strength of this phenomenon has increased over time. Although the total number of births is consistently larger for women who lost a first or second child than for women whose first and second child survived, the number of surviving children is smaller for women whose first or second child died. Women who experience a child death early in their reproductive careers go on to have more births, but ultimately smaller numbers of living children than do their counterparts whose early children survive. On average, the difference between the number of children ever born and the number of surviving children is more than two children among women whose
ELIZABETH FRANKENBERG 7 - 4 3 1 O f CJead`~84 Alive,'84 Dead,,91 Al~ve,'91 Sundial at the mother's marriage date ~ 4 years 325 | IC! CEB1 Arcs 1 FIGURE 9-2 Children ever born (CEB) and children surviving (CS) by survival status of the second child for women aged 41-46 with two births in the first 4 years of marriage. first or second child died by the specified date. For women whose first or second child survived to the specified date, the difference between the number of chil- dren ever born and the number of children surviving is considerably less than one child. These differences suggest that mortality risks of children with the same mother are correlated. Mortality is higher among the subsequent children of women who experienced the death of a first or second child than among the subsequent children of women whose first and second children survived to a specified point. I also present other measures of links between child mortality and subse- quent fertility common in the demographic literature. Figure 9-3 displays parity progression ratios by the number of children who survived until the birth of the child at the specified parity. For example, at the birth of the second child (parity two) a woman has had one previous child. If that child survived she has one living child, if it died she has no living children. Rates are calculated for three periods: 1971-1977, 1978-1984, and 1985-1991. A woman is counted as pro- gressing to the next parity if she gives birth to her next child within 42 months of attaining the parity recorded in Figure 9-3. I chose a cutoff date of 42 months because I wanted an interval long enough to capture most women who progress to the next parity (and so longer than mean length of birth intervals) but short enough to guarantee an adequate number of women within the 6-year window. A woman may contribute multiple observations to the figure, depending on which dates she attained which parities. At a general level, several of the patterns that emerge from Figure 9-1 reap- pear in the parity progression ratios. Overall, the proportions of women in each
326 0~7 0~6 0.2 0.1 0.8 .c c0.5 O03 o o 07 1 ~ off 1 0~5 .= 1~, 0 4 go 0 ~ ~ 0~3 cry =. O t 2 0.1 o ~1 0.7- . . - 0.6 0.5 ~ .e 0.4 ~ 0 - ~ 0.3 rL 0-~- O 0.1 & INDONESIA: 1971-1991 A. 1971-lg77 . _ O P~ritv B. lg78-1984 Parity C. 1985-1991 .. .. . .; - - - - - . . - .- -- - - - --,- . . :: ::~:-::-:,::.:_::-:: ::: :::: - . .- - .. ;. ~ ~ ~ - - . -- . - . ..... - ~ - ~ ., . - . -. - ~ ~ -I 2 ... . . . . . . 1~-~ ~ ::.: : :~::: :-: ': ,........... i . i... ~.~-~..~.... .... 1 :'::.:: :' 1:: ::'::.::::-:::': :::': . . .. .. . ~ -I :....-.:.:.. :~ : -A .-~:. .... . ... . -, ~ :~ 3 Parity 4 FIGURE 9-3 Parity progression by survival status of previous children. X O Surv. [1 1 SurY, 532 Surv ~3 Surv. ·0 Surv. C3 1 Surv. .2 Surv. ~3 Surv. ·O Surv. C3 1 Surv. 53 2 Surv. succeding parity/surviving children category have declined over time, as ex- pected in a setting where fertility is declining. For a given parity, the smaller the number of living children, the higher the proportion progressing to the next parity within 42 months. However, exceptions are found among parity-four women for whom two or three of the preceding children have died.
ELIZABETH FRANKENBERG 327 Patterns in the parity progression ratios over time are quite interesting. In 1971-1977, within a parity, changes in parity progression ratios with each addi- tional surviving child tend to increase with parity. For example, for women of parity two, the proportion progressing to parity three within 42 months is 0.689 when the first child has died by the birth of the second, 0.649 when the first child survived to the birth of the second a difference of around 6 percent. For women of parity three, the parity progression ratio for women with no surviving children is 8.7 percent greater than the parity progression ratio for women with one surviv- ing child (0.746 versus 0.681), whereas for women with one surviving child the parity progression ratio is 7.5 percent greater than for women with two surviving children (0.681 versus 0.63~. On the other hand, in 1985-1991 within-parity differences in parity progression ratios by one-child increments in the number of surviving children are greater at the lower parities than at the higher parities. For example, among parity-two women, the parity progression ratio is 0.58 when the first child died and 0.41 when the first child survived, a difference of 30 percent. A similar difference is observed between parity-three women with zero versus one surviving child. The differences are smaller for parity-three women with one versus two surviving children and much smaller for parity-four women. Two statements can be made from these patterns. First, as desired family size shrinks and fertility falls, there is a corresponding decline in the parity at which differences in behavior by number of living children emerge. Second, within-parity differences in parity progression ratios by number of surviving children increase with time. Possibly these widening differentials mark an in- crease in women' s ability to control their fertility through contraceptive use. Birth intervals are another indicator of reproductive patterns that researchers typically investigate in efforts to learn about fertility responses to mortality. Table 9-3 summarizes the average length of time between births. Birth intervals are stratified by period, birth order, and survival status of the initiating (index) child at 1 month and at 12 months after birth. In constructing such a table one wants to avoid the issue of reverse causality. Specifically, it is possible that a woman becomes pregnant again quickly after the birth of the index child and that after this conception the index child dies. Such a phenomenon could occur if the subsequent conception causes the mother to stop breastfeeding the index child, who then dies during the vulnerable period of weaning. In this scenario the short subsequent interval causes the death of the index child, rather than the death of the index child causing a short subsequent interval. To avoid this problem I classify the index child as alive if it was alive 9 months before the birth of the subsequent child. That is, children who died within 9 months of the birth of the subsequent child (and so presumably after the conception of the subsequent child) were counted as alive. The standard argument with respect to birth interval is that the interval after a child that dies in infancy will be shorter than the interval after a child that survives infancy. The shorter interval occurs either because the mother wants to
328 INDONESIA: 1971-1991 TABLE 9-3 Subsequent Birth Interval Lengths by Survival Status of Initiating Child 1971-1977 1978-1984 1985-1991 Birth Survival Order Status at: Dead Alive Dead Alive Dead Alive 1 month 24.3 35.4 22 34.2 21.5 28.8 (310) (6,978) (366) (7,770) (128) (2,475) 12 months 31.8 35.6 27.1 34.5 24.9 29 (234) (6,726) (240) (7,503) (92) (2,376) 1 month 27.5 37.6 24.9 36.7 22.6 30.9 (212) (5,533) (235) (6,005) (59) (1,645) 12 months 32.1 37.8 30 36.9 26.5 31.2 (176) (5,346) (161) (5,835) (91) (1,550) 3 1 month 26 36.4 26.5 36.8 27 31.3 (140) (4,089) (132) (4,306) (64) (1,296) 12 months 30.5 36.7 30.1 37.1 24.4 31.7 (137) (3,935) (171) (4,123) (60) (1,230) 4 1 month 24.9 36.2 27.2 36.1 21.4 30.6 (96) (2,987) (108) (2,904) (32) (918) 12 months 31.7 36.5 33 36.2 24.9 31 (112) (2,866) (113) (2,801) (44) (873) NOTE: Numbers in parentheses below birth interval lengths are sample sizes. "replace" the child she has lost or because the death of an infant ends breast- feeding, which hastens the return of ovulation and thus increases the risk of a rapid subsequent conception. The results in Table 9-3 demonstrate that birth intervals are considerably shorter after a child dies either as a neonate or as a postneonate than after a child who survives those periods, regardless of birth order or time period. The differ- ence in interval lengths by survival status is much larger (25-30 percent) when one compares a neonatal death to a neonatal survival than when one compares a postneonatal death to postneonatal survival (10-15 percent). Prior to the 1985-1991 period, birth intervals are remarkably similar in length when the index child survives. Regardless of birth order, subsequent intervals average 34-37 months. In the 1985-1991 period, however, subsequent intervals are consistently 29-32 months, indicating a fairly dramatic decline in average birth interval length over time. The average length of birth intervals after a child that died has also declined over time, but the decline is not so concentrated in the later period.
ELIZABETH FRANKENBERG 329 The differences between interval lengths after a child who died and interval lengths after a child who survived may arise because women who lose a child consciously try to replace the dead child by becoming pregnant again quickly, or because cessation of breastfeeding hastens conception, or because certain women are predisposed to high fertility (rapid childbearing) and high mortality. One way to rule out the argument that cessation of breastfeeding causes a rapid subsequent conception is by moving forward an interval and considering how the survival status of the preceding sibling affects the interval between the next two children. Results from this exercise are displayed in Table 9-4. Subsequent birth intervals are shorter when the preceding sibling died than when the preceding sibling survived. Absolute differences in interval lengths, however, are much smaller when it is the preceding sibling rather than the index child who died. No clear patterns emerge in differences in interval length by survival status of the preceding sibling. At birth order four, the proportionate difference in interval lengths by survival status declines over time. At other birth orders the proportionate difference is relatively constant with time. All of the indicators considered here suggest that fertility patterns differ for women who have lost a child in comparison with women whose children have survived. Among older women, those who lose a first or second child soon after its birth go on to have more children than women whose early children survived. But despite larger numbers of births, the number of living children is lower for women who lost a child early on, suggesting that these women never "catch up" to their lower-mortality counterparts. At parities of four and below, the propor- tion of women having an additional child within 42 months rises with each prior TABLE 9-4 Subsequent Interval by Survival Status of the Preceding Sibling 1971- 1977 Dead 1978-1984 Alive Dead 1985-1991 Alive Dead Alive 4 35 37.5 33 36.6 29.1 30.8 (668) (5,077) (724) (5,516) (200) (1,504) 33.6 36.4 34 36.8 28.5 31.5 (457) (3,772) (496) (3,942) (156) (1,204) 31.5 36.5 32.9 36.1 28.7 30.5 (360) (2,723) (335) (2,677) (121) (829) 29.9 34.6 34.8 34.3 29.5 30.2 (274) (1,782) (253) (1,848) (68) (563) NOTE: Numbers in parentheses below birth interval lengths are sample sizes.
330 INDONESIA: 1971-1991 child that died. At parities of four and lower, subsequent intervals are shorter when the index child died than when the index child survived. At parities of two through five, subsequent intervals are shorter when the preceding sibling died than when it survived to the birth of the index child. These general patterns prevail across time, in the face of declining levels of fertility and child mortality and increasing availability of contraceptives. The analysis of replacement rates and parity progression ratios suggest that differences in fertility outcomes by child survival experience have widened over time. This phenomenon is consistent with the idea that family planning and health programs make available to women knowledge and technologies that help them control their fertility. It may also be the case that as access to those technologies increases, differentials widen between women with characteristics that predispose them to high fertility and high mortality and women without these characteristics. Accordingly, the question that remains unanswered, despite the results pre- sented above, is whether the death of an infant or young child prompts a con- scious, intentional fertility response in a woman. If her child had survived, would the woman' s fertility be different? Or do her fertility patterns and the mortality experiences of her children arise from other factors? If other factors are the reason, then secular declines in mortality are not likely to affect fertility patterns unless they are accompanied by changes in other factors as well. IS THERE A CAUSAL LINK BETWEEN CHILD MORTALITY AND REPRODUCTIVE PATTERNS? In this section I briefly summarize results from an attempt to pinpoint a conscious, behavioral response of one element of fertility patterns to a child death. The approach, which is described more fully elsewhere (Frankenberg, 1996) relies on birth intervals. The goal of the analysis is not to quantify the effect of child mortality on total fertility as measured by completed family size. Instead, the intent is to design and implement an approach that measures a behav- ioral response to a child death but is relatively free of the statistical problems described above that confound empirical work on this topic. A multivariate counterpart of the bivariate results presented on birth interval length would involve a model of the determinants of the birth interval after the nth child. Such a model can be written as: log Sn = An + Fit + ~nDn-i + En + £, (1) where
ELIZABETH FRANKENBERG X So = length of the interval after the nib child, = characteristics that might affect the interval length, Dn_1 = survival status of the (n - 11~ child, = idiosyncratic error, and £ = mother-specific error. Similarly, for the interval after the (n + 1)~ child, one could write: log Sn+1 = ~n+1 + 13n+1X + ~n+lDn + £n+1 + £ 331 (2) In both models, the parameter estimates are biased if £ iS correlated with Dn_1 or with X. For example, suppose women differ in their level of initiative with respect to use of health and family planning services. Level of initiative is difficult to measure, but is a likely determinant of both child mortality and fertil- ity patterns. If level of initiative is excluded from the models, parameter esti- mates will be biased. One way to rid the parameters of this bias is to estimate the equation: log (_ sn . )= (Fin+1-Fin)X + (6n+1-~n)Dn + (~;n+1-En) + (~-At) (3) where the woman-specific errors drop out of the equation. The intuition behind this model is that the relationship between the interval after the nib and the interval after the (n + 1)~ child is a function of X character- istics and the survival status of the (n - 1)~ child. The approach is one of differences in differences. I am comparing the difference between the nib and (n + 1)th interval when the n - 1 child died to that difference when the n - 1 child survived. In stylized terms I consider two sequences: Bn_1 Dn_l Bn Bn+l Bn+2 n-1 On ~n+1 and ask how the survival status of the (n - 1)~ child in the interval between its birth and the birth of the nib child affects the relationship between the interval after the nib child and the interval after the (n + 1)~ child. Earlier results suggested that the interval after the nib child is shorter when the preceding sibling died than when it survived. Is this because the death causes women to become pregnant again more quickly than they would have had the child survived? Or do other "background" factors that predispose a woman to high mortality and high fertility explain both the short interval and the death of a child? If the explanation lies in other factors, the association between the interval
332 INDONESIA: 1971-1991 after the nib child and the death of the child will be spurious. However, if the other factors have the same effect on the interval after (n + 11~ child as on the interval after the nib child, any difference in the relationship between those two intervals by the survival status of the (n - 11~ child is not a result of predisposing background factors. It may be that these predisposing factors are relatively unimportant and that a simpler estimation strategy would be adequate. This question can be answered by computing a Hausman statistic that compares the results from the differences in differences to a simpler model. A question that immediately arises is how to implement this approach with data. One could analyze the relationship between two adjacent intervals for any woman who experienced this sequence, controlling for factors such as birth order and survival status of children higher than order n - 1. However, it is difficult to construct a set of controls for reproductive history that parsimoniously captures differences in reproductive factors other than the survival status of the (n - 11 child. Alternatively, I impose a series of conditions that force certain similarities on the women in the analysis. The first condition is that women must have had at least four births at the time of the interview. The second condition is that the first four births must all be singletons. The third condition is that first child must have died before the birth of the second child or survived to the birth of the fourth child. The third and fourth conditions are that the second and third child must have survived to the birth of the fourth child. Together, the third and fourth conditions guarantee that the only mortality event the women have experienced over the period analyzed is the death of the first child before the birth of the second child. The last condition is that the birthdates of the children in question must be fully reported (the mother provides the year and month of birth). A total of 7,761 women of the 32,000 women interviewed met these criteria. These conditions are equivalent to stringent controls with respect to repro- ductive patterns. Inevitably, the conditioning selects certain women for the analy- sis while excluding others. I compared the characteristics of the 7,761 women analyzed with the characteristics of other women in the sample who had at least two births by the time of the interview (25,674 women, pooled from the 1987 and 1991 DHS data). The women analyzed are on average about 3 years older and have 1.2 more children than the other mothers of at least two children. The main reason that the women I analyzed are older is that I require them to have at least four children, whereas the comparison group needed only two children. Levels of education and marital status are similar across the two groups. One might expect that the older, higher-fertility women would have less education than their counterparts. Instead, education levels are similar because I eliminated women with imputed birthdates, and these tend to be the less-educated women. The women selected are relatively high-fertility women with relatively low- mortality children. The characteristics that lead to high fertility and low mortal
ELIZABETH FRANKENBERG 333 ity will not affect my results to the extent that they have the same effect on all of a woman's birth intervals. By comparing two birth intervals from the same woman I obtained estimates free from biases caused by unobserved factors that affect both intervals the same way. The hypothesis of interest is that the survival status of the first child affects the ratio of the interval after the third child to the interval after the second child. The death of the first child lowers by one the number of surviving children at subsequent parities and so might affect interval lengths if it interrupts a woman's desired schedule of childbearing. The death of the first child also changes the potential order of children by sex. The second child, which may be the opposite sex of the child who died, becomes the first surviving child. My model contains only three variables: whether the first child survives, the sex of the first surviving child, and the sex of the first surviving child interacted with the survival status of the first child.2 The variables correspond to four categories of women: those whose first child survived and is female, those whose first child survived and is male, those whose first child died and whose second child is female, and those whose first child died and whose second child is male. Although evidence on fertility preferences in Indonesia suggest that Indonesians display a strong prefer- ence for balance in the overall sex composition of their offspring, it is possible that there are preferences for order (Cleland et al., 1983~. The model of interest is the model of log of the ratio of the two intervals (the interval after the third child divided by the interval after the second child). The predicted ratios of the interval lengths are shown in Figure 9-4. From the figure it is clear that the ratios vary by whether the first child survived, but that the effect of the first child's death on the ratio is strongly mediated by the sex of the second child (the first surviving child for these women). When the first child died and the second child is female, the ratio is lower than when the first child survived. When the first child died and the second child is male, the ratio is higher than when the first child survived. Ratios for women whose first child survived are lower when the first child is female than when the first child is male. These results provide evidence of a causal effect of the first child' s survival status on the timing of subsequent fertility, mediated by the sex of the first surviving child. The ratio for women whose first surviving child is female is significantly different from the ratio for women whose first surviving child is male. However, the ratio for women whose first child is female and survives is not statistically significantly different from the ratio for women whose first child 2I tested for the significance of additional covariates related to socioeconomic status, but none were significant. In one version of the model I included a term for survey year to test the appropri- ateness of pooling women from the 1987 and 1991 surveys. The term was not statistically significant and so was not retained in the final analysis. Although an earlier table suggested that the overall samples differ, the subsample of women who contribute to this analysis do not differ by sample membership.
334 1.14 - 1.12 1.10 1.08 1.06 1.04 ° 1.02 ._ ct En: 1.00 1.98 1.96 1.94 INDONESIA: 1971-1991 1.14 1.11 1.07 1.01 1st died, 1st survived, 1st survived, 1st died, 2nd daughter daughter son 2nd son FIGURE 9-4 Ratio of interval after third to interval after second child. dies and whose second child is female. Nor is the ratio for women whose first child is male and survives statistically significantly different from the ratio for women whose first child died and whose second child is male. The ratio models provide a way to check for a causal relationship between the death of a child and timing of subsequent fertility under the assumption that unobserved woman-specific factors affect both fertility and mortality, so that estimates based on comparisons across women are biased. I can check the valid- ity of this assumption, and thus the necessity of using the ratio approach, with a Hausman test. The Hausman test compares the results of the ratio model with the results from a model in which the two birth intervals of interest are pooled: log S = or + OX + ED + £, (4) where S is the interval after the second or the third birth and D indicates the survival status of the first child. Parameter estimates for the ratio model and the pooled model are presented in Table 9-5. The Hausman test yields a statistic of 27.7, which is highly significant. SUMMARY AND CONCLUSIONS This chapter reviews the relationship between infant and child mortality and fertility in Indonesia. Both fertility and child mortality have fallen dramatically since the late 1960s (the point after which national-level statistics are available). Evidence for periods prior to the 1960s suggests that a steady mortality decline began in the 1950s, as food supplies returned to their prewar levels and public health programs were initiated. The decline in mortality appears to have pre- ceded the decline in fertility by 10-15 years. Accompanying and likely contribut- ing to declines in fertility and mortality have been dramatic increases in the availability of health and family planning services since the mid-1960s.
ELIZABETH FRANKENBERG TABLE 9-5 Coefficients from the Differences in Differences and Pooled Models Model Coefficient Standard Error Differences in differences First child dead -0.541 0.0399 First surviving child male 0.0351 0.0153 Male child x first child dead 0.0867 0.0561 F statistic 3.49 Pooled estimates First child 0.0047 0.0215 First surviving child male 0.0122 0.0082 Male child x first child dead -0.1036 0.0301 F statistic 7.38 Hausman statistic 27.7 335 I present a series of descriptive statistics as a means of charting the evolution of the relationship between infant and child mortality and subsequent fertility over time. A number of tentative findings emerge from this exercise. First, completed family sizes, in terms of number of children ever born have declined over time regardless of child survival experiences. However, the decline is steeper for women whose first two children survived their early years than for women whose first or second child died as an infant or toddler. Second, the difference in fertility patterns by survival status of offspring appears to have widened over time. That is, differentials in parity progression ratios and in children ever born by survival experiences are larger in more recent periods. Such a result is consistent with the idea that the increasing availability of contraceptive and health technology has given couples more control over their fertility and thus more scope for responding to child's death or survival. But from these results one cannot tell whether differentials widen with time because all couples use newly available technology to respond to a child's death, or whether some couples assertively use this technology to alter both fertility and children's health, while other couples take a more laissez-faire approach. Third, changes over time in parity progression ratios by the survival status of preceding offspring suggest that over time there has been a decline in the parity at which survival status of preceding children affects the decision to go on to the next parity. Again, this finding is not surprising in a context where fertility has been declining. Analysis of subsequent birth interval lengths reveals that birth intervals are consistently shorter when either the index child or the preceding sibling of the index child died than when it survived. Differences in interval lengths by sur- vival status are larger when it is the survival status of the index child that is
336 INDONESIA: 1971-1991 monitored rather than the survival status of the preceding sibling. This difference almost certainly reflects the fact that the death of a child ends breastfeeding and so speeds the return of ovulation and thus the conception of the next child. This mechanism operates in the comparison of interval lengths stratified by the sur- vival status of the index child, but not in the comparison of interval lengths stratified by the survival status of the preceding sibling. The results for birth interval lengths suggest that intervals were on average six months shorter in the late 1980s than they were in earlier penods. No clear time trends emerge in the differences in interval lengths by survival status of index or preceding children. In the last section of the chapter I summarize results obtained from a model that attempts to establish a causal link between a child death and subsequent fertility patterns. These results provide evidence that for certain groups of women a child' s death changes the pattern of subsequent interval lengths relative to that pattern when a child survives. However, the changes are small in magnitude and occur on average at interval lengths of more than 2 years. Consequently, it is extremely unlikely that these changes have any serious implications for com- pleted levels of fertility. Given that the intervals are well within the range considered to be healthy for women and children, it is also unlikely that the changes contribute to poor health either for mothers or for their offspnng. ACKNOWLEDGMENTS This work was supported by National Institute of Child Health and Human Development grant no. 5R29 HD32627-02. The author gratefully acknowledges comments from Barney Cohen, Irma Elo, Andrew Foster, Mark Montgomery, Samuel Preston, Duncan Thomas, and two anonymous referees. REFERENCES Ben-Porath, Y. 1976 Fertility response to child mortality: Micro data from Israel. Journal of Political Economy 84(2):S163-S178. Boerma, J.T., A.E. Sommerfelt, J.K. van Ginneken, G.T. Bicego, M.K, Stewart, and S.O. Rutstein 1994 Assessment of the quality of health data in DHS-I surveys: An overview. Pp. 1-26 in An Assessment of the Quality of Health Data in DHS-I Surveys. DHS Methodological Re- ports, No. 2. Calverton, Md.: Macro International, Inc. Central Bureau of Statistics [Indonesia], National Family Planning Coordinating Board [Indonesia], and Institute for Resource Development/Westinghouse 1989 National Indonesian Contraceptive Prevalence Survey, 1987. Columbia, Md.: Institute for Resource Development/Westinghouse. Central Bureau of Statistics [Indonesia], National Family Planning Coordinating Board [Indonesia], Ministry of Health [Indonesia], and Macro International, Inc. 1992 Indonesia Demographic and Health Survey, 1991. Columbia, Md.: Macro International, Inc.
ELIZABETH FRANKENBERG 337 Central Bureau of Statistics [Indonesia], State Ministry of Population/National Family Planning Coordinating Board, Ministry of Health, and Macro International, Inc. 1995 Indonesia Demographic and Health Survey 1994. Calverton, Md.: Macro International, Inc. Cleland, J., J. Verrall, and M. Vaessen 1983 Preferences for the Sex of Children and the Influence on Reproductive Behavior. World Fertility Survey Comparative Studies no. 27. Voorburg, Netherlands: International Sta- tistical Institute. Frankenberg, E. 1996 Child Death and the Timing of Subsequent Fertility. Unpublished manuscript, RAND, Santa Monica, Calif. Heer, D., and H. Wu. 1978 Effects in rural Taiwan and urban Morocco: Combining individual and aggregate data. Pp. 135-159 in S.H. Preston, ea., The Effects of Infant and Child Mortality on Fertility. New York: Academic Press. Hugo, G., T. Hull, V. Holland, and G. Jones 1987 The Demographic Dimension in Indonesian Development. Oxford, England: Oxford University Press. Hull, T.H. 1989 Roots of primary health care institutions in Indonesia. Pp. 500-508 in J. Caldwell, S. Findley, P. Caldwell, G. Santow, W. Cosford, J. Braid, and D. Broers-Freeman, eds., What We Know About The Health Transition: The Cultural, Social, and Behavioral Determinants of Health, Vol. II. Canberra: Health Transition Centre, The Australian National University. Knodel, J. 1978 European populations in the past: Family-level relations. Pp 21-45 in S.H. Preston, ea., The Effects of Infant and Child Mortality on Fertility. New York: Academic Press. Lantz, P., M. Partin, and A. Palloni 1992 Using retrospective surveys for estimating the effects of breast feeding and child spacing on infant and child mortality. Population Studies 46(1):121-129. Matthiessen, P.C., and J.C. McCann 1978 The role of mortality in the European fertility transition: Aggregate-level relations. Pp. 47-68 in S.H. Preston, ea., The Effects of Infant and Child Mortality on Fertility. New York: Academic Press. Mauskopf, J., and T. Wallace 1984 Fertility and replacement: Some alternative stochastic models and results for Brazil. Demography 21(4):519-536. Mensch, B. 1985 The effect of child mortality on contraceptive use and fertility in Columbia, Costa Rica, and Korea. Population Studies 39(2):309-327. Nitisastro, W. 1970 Population Trends in Indonesia. Ithaca and London: Cornell University Press. Olsen, R. 1980 Estimating the effect of child mortality on the number of births. Demography 17(4):429- 443. Pebley, A.R., and P.W. Stupp 1987 Reproductive patterns and child mortality in Guatemala. Demography 24(1):43-60. Potter, J.E. 1988 Birth spacing and child survival: A cautionary note regarding the evidence from the WFS. Population Studies 42(3):443-450.
338 INDONESIA: 1971-1991 Preston, S.H. 1978 Introduction. Pp. 1-18 in S.H. Preston, ea., The Effects of Infant and Child Mortality on Fertility. New York: Academic Press. Rinehart, J., and S. Kols 1984 Healthier mothers and children through family planning. Population Reports. Series J. no. 27., May-June. Rodriguez, G., J. Hobcraft, J. McDonald, J. Menken, and J. Trussell 1984 A comparative analysis of determinants of birth intervals. World Fertility Survey Com- parative Studies no. 30. Voorburg, Netherlands: International Statistical Institute. Ross, J., and E. Frankenberg 1993 Findings from Two Decades of Family Planning Research. New York: The Population Council. Sullivan, J., G. Bicego, and S. Rutstein 1990 Assessment of the quality of data used for the direct estimation of infant and child mortal- ity in the DHS surveys. Pp. 113- 140 in An Assessment of DHS Data Quality. Columbia, Md.: Institute for Resource Development. Trussell, J., L. Martin, R. Feldman, J. Palmore, M. Concepcion, and D. Abu Bakar 1985 Determinants of birth-interval length in the Philippines, Malaysia, and Indonesia: A hazard-model analysis. Demography 22(2): 145-168. van Ginneken, J.K. 1974 Prolonged breastfeeding as a birthspacing method. Studies in Family Planning 5(6):201- 206. Wolpin, K. 1984 An estimable dynamic stochastic model of fertility and child mortality. Journal of Politi cal Economy 92(5):852-874. World Bank 1984 Population Change and Economic Development. Washington, D.C.: The World Bank. 1991 Indonesia: Health Planning and Budgeting. Washington, D.C.: The World Bank. 1994 Indonesia: Sustaining Development. Washington, D.C.: The World Bank.
