|
Items in cart [0] |
Questions? Call 888-624-8373 |
| Copyright © 2009. National Academy of Sciences. All rights reserved. Terms of Use and Privacy Statement |
Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 384
11
Child Mortality and the Fertility Transition:
Aggregated and Multileve! Evidence
from Costa Rica
Luis Rosero-Bixby
INTRODUCTION
Is decreasing child mortality a prerequisite a necessary condition for de-
creasing fertility? Can decreasing child mortality alone trigger the fertility tran-
sition? These questions have important policy implications. If improving child
survival is a precondition for birth control, family planning programs in the least
developed regions are unlikely to succeed, especially if these programs have a
vertical organization independent of child health interventions. In turn, if reduc-
ing child mortality is a sufficient condition, family planning programs may be
somewhat superfluous: "Development is the best contraceptive." In this chapter
I address the issue of whether reduced child mortality is crucial for the fertility
transition by examining the empirical evidence from Costa Rica, a developing
country that managed to decrease both child mortality and birth rates. Here I
examine Costa Rica's record at the aggregate and the individual level.
A strong association between child mortality and fertility is well documented
in the literature. Countries with low infant mortality almost always have low
birth rates (Heer, 1966; Mauldin et al., 1978~. Couples that have lost a child are,
in turn, less likely to use contraception, tend to have more children, and have
shorter birth intervals (Taylor et al., 1976~. However, this association is neither
proof of causation nor indicates the direction of causation. The association may
have three closely linked origins: First, child mortality and fertility share a
common set of determinants, such as mother's education, access to health ser-
vices, breastfeeding practices, and less observable traits such as a preference for
"high-quality" children or a less fatalistic outlook on life (Hanson et al., 1994~.
Second, lower fertility may lessen child mortality by acting on such potential
384
OCR for page 385
LUIS ROSERO-BIXBY
385
mechanisms as maternal depletion associated with pregnancies and lactation
(Trussell and Pebley, 1984), sibling competition for scarce family resources and
maternal care including breastfeeding (Pebley and Millman, 1986), and transmis-
sion of infections in child-crowded environments (Blacker, 1987; Haaga, 1989~.
The third possibility the one addressed in this chapter is that the direction of
causation runs from child survival to fertility. Although disentangling these three
types of causal links is an impossible task with the data available, by statistically
controlling the effect of particular variables and by paying attention to the se-
quence of events over time, I try to reach some conclusions about the third causal
link: the role of child mortality on contemporary fertility transitions.
There are several explanations for the postulated effect of child mortality on
fertility. In the classic demographic transition theory, high fertility is in part a
response to high levels of infant and child mortality (Notestein, 1953; Davis,
1955~. Parents have many children to replace those who have died (replacement
effect) or parents set excess fertility goals in anticipation of their children's
deaths (insurance effect) (Lloyd and Ivanov, 1988~. Increased probabilities of
child survival may thus be a necessary condition for fertility decline: Parents will
not control their fertility unless they have assurance their children will survive
(Taylor et al., 1976~. Moreover, improvements in child survival may be a suffi-
cient condition for fertility decline as soon as parents realize that it is no longer
necessary to have many children or to suffer the economic consequences of larger
families (Preston, 1978~. Under these circumstances, fertility transition would be
merely an adjustment process to conditions brought about by reduced child mor-
tality (Carlsson, 1966~. Where long periods of breastfeeding are the norm, a
physiological mechanism may also be important: The death of an infant may
substantially reduce the breastfeeding period and, consequently, the period of
temporary infecundity after childbirth, which would result in shorter birth inter-
vals and more children born by the end of the reproductive life (Cochrane and
Zachariah, 1983~.
Two earlier studies of the Costa Rican record show significant effects on
reproductive behavior of couples whose children have died (Rutstein and Medica,
1978; Mensch,1985~. The replacement strategy is the most likely effect of one's
own children's deaths. This effect seems reasonable and is well documented in
populations that already control fertility, as it was in the Costa Rican samples
analyzed in these two studies. In pretransition societies, however, it is hard to
believe that couples turn on and off their fertility in response to child deaths.
Data for European populations in the past usually do not show significant re-
placement effects (Knodel, 1978~.
A third Costa Rican study, based on the 1976 World Fertility Survey, does
not find a significant effect of child mortality in the community upon the repro-
ductive behavior of individual women (Heer and Rodriguez, 1986~. The insur-
ance strategy is the most likely causal mechanism for the postulated effect of
community-level child mortality. This effect has been seldom studied in other
OCR for page 386
386
AGGREGATE AND MULTILEVEL EVIDENCE FROM COSTA RICA
countries because of the lack of reliable multilevel data. One of the problems for
building multilevel data sets is the definition of community and then having
reliable estimates of child mortality or other contextual variables. The study of
Heer and Rodriguez used the canton as the unit for computing contextual vari-
ables. This Costa Rican administrative unit is, however, in many cases too large
and internally too heterogeneous to be considered a meaningful entity; its bound-
aries are often arbitrary and the relevant context for the canton' s edges may be
that of neighboring cantones. In this chapter I overcome these limitations by
using geographic information system data to obtain a precise and compact defini-
tion of "context" (the area within a radius of 5 kilometers in rural areas and 1
kilometer in urban areas from the index household). In addition, in this chapter I
focus on the crucial event, adoption of family planning for the first time, as the
dependent variable. In the studied cohorts, which grew up in a natural fertility
environment, this represents a breaking point with the past.
After a brief analysis of Costa Rican national trends during this century, I
examine the role of child mortality on the fertility transition at the macro- and the
micro-level. The analysis at each level looks first at bivariate associations and
then moves into multivariate associations with the purpose of isolating net ef-
fects. The macro-level analysis is based on data from 89 Costa Rican counties,
which are small geographic entities defined on the basis of the country's admin-
istrative division in cantones and distritos. The multivariate analysis of this data
set includes traditional regression models of the effect of child mortality on
fertility at three points in time and between these points. It also models the event
"fertility transition," as operationalized by two dependent variables: onset and
pace of decline. The multivariate analysis of the onset of the fertility transition is
carried out using Cox regression. The individual-level analysis uses a survey
conducted in 1984 that included a lifetime calendar of contraceptive use. The
analysis tests the hypothesis that contextual child mortality patterns influence the
adoption of fertility control. This analysis focuses on the individual-level equiva-
lent of fertility transition the timing in the adoption of birth control rather
than on reproductive behavior in general, and it is restricted to the cohorts that
lived through the fertility transition (i.e., women aged 15-34 in 1960~. Given that
the analysis combines contextual or macro-level indicators of child mortality and
other variables to explain individual-level reproductive behavior, this analysis is
referred to as "multilevel."
SECULAR TRENDS IN COSTA RICA
Costa Rica experienced one of the earliest and fastest, although incomplete,
fertility transitions in the developing world. The total fertility rate fell from 7.3 to
5.5 between 1960 and 1968, the year an energetic national family planning pro-
gram started, and then to 3.7 by 1976, the year the decline abruptly stopped
(United Nations, 1985~. After fluctuating erratically around 3.7 births, the rate
OCR for page 387
LUIS ROSERO-BIXBY
387
began to decline again by 1986, although at a slow pace. According to the most
recent estimate, the total fertility rate is 2.7 births per woman in 1995 and the
contraceptive prevalence rate is 75 percent in 1993 (Caja Costarricense de Seguro
Social, 1994~.
There is no unique and simple explanation for the fertility transition in Costa
Rica. Some authors stress the role of education, especially among women (Stycos,
1982~. Other authors note the pervasiveness of the decline across all regions and
social sectors to stress the role of the government in erasing differentials and
redistributing income (Behm and Guzman, 1979~. Others underscore the explo-
sive increase in the supply of contraceptives by the private sector and later by the
public sector that took place in the 1960s (Tin Myaing Thein and Reynolds,
1972~. It is obvious that the official family planning program, launched in 1968,
had nothing to do with the early fertility decline, which occurred mostly among
the urban, middle class. It is, however, accepted that the rapid propagation of the
process to rural and urban working classes was catapulted by the energetic family
planning program launched in 1968 (Rosero-Bixby et al., 1982~. Recent litera-
ture also stresses diffusion, or social interaction, processes as one of the key
determinants of the transition (Rosero-Bixby and Casterline, 1994, 1995; Knight,
1995~.
Although most studies mention child mortality decline among the socioeco-
nomic transformations that may have brought some of the fertility transition,
none considers it a crucial factor.
At the fertility transition onset, the country' s infant mortality rate was 76 per
1,000 and the child mortality risk (the probability of dying before the age of 5
years) was 96 per 1,000, which are high levels at current standards but were not
in the late 1950s. In absolute terms, most of the possible reduction in child
mortality rate had already taken place by 1960. It had declined by two-thirds
from levels of about 300 per 1,000 observed by the 1910s (Figure 11-1~. In
relative terms, however, the fastest infant and child mortality reductions occured
in the 1970s. The child mortality rate of about 18 per 1,000 in 1990 is one-fifth
that of 1960. This acceleration in the pace of child and infant mortality decline in
the 1970s has been linked to three factors (Rosero-Bixby, 1991a): public health
interventions, which probably were the most important; development, including
social improvements and a vigorous and sustained growth in the economy (Fig-
ure 11-3), and the fertility decline.
Figure 11-1 compares the evolution during the twentieth century of the gen-
eral fertility rate, the child mortality rate, and an indicator of the economy (im-
ports per capita, one of the few indicators with a series covering the entire pe-
riod). The data for this figure come from official statistics, which in Costa Rica
are reasonably reliable. The only apparent association between the general fertil-
ity rate and the child mortality rate is the acceleration in the relative decline in the
latter during the fertility transition.
Actually, this acceleration starts a few years after the fertility transition
OCR for page 388
388
so
0 0
0 0
_ 0
$sn 0661 u! Sl3°dm! mideo Had
0 0 0
0 0 0
~ cat
to
i I J .. ~I I ~
,2°o_O
o
Q
0-~
,3
- o~o
AS
0~
a_
i_
·= 1
LL
-
).?
0
.=
o`.~
f1'
o - o
0 0 0 0
~0 ~0
co co cat cat
0 0 0
~0 up
odor ad sane tow Pl!~o pue M!I!pa~
to
~
to
~
cot
to
to)
to
cat
a,
.=
cd
· ~
cd
o
sit
o
· ~
cd
· ~
sit
cd
cd
sit
· ~
so -
cd
cd
sit
·
c~
s~
o
1
OCR for page 389
LUIS ROSERO-BIXBY
389
onset. This temporal sequence suggests that the direction of causation, if any,
may run from fertility to child mortality. The rapid, concurrent economic growth
suggests, in turn, that both child mortality and fertility declines were part of a
broader transformation process in society and living standards.
Figure 11-1 hardly supports the thesis that child mortality decline was key
for the fertility transition in Costa Rica. In particular, the record until 1964 is that
neither the secular decline nor the short-term fluctuations (mostly due to measles
epidemics) in child mortality was sufficient to alter fertility or start the transition.
The small fluctuations in fertility during those early years are mostly linked to
marital disruption, a marriage boom in the 1950s, and declines in widowhood.
The only way that Figure 11-1 could be compatible with the "sufficient condi-
tion" thesis would be if the effect of child mortality requires long lags, or thresh-
old doses, to act.
The data in Figure 11-1 are not conclusive as to whether a certain minimal
level of child survival is required for fertility transition the necessary condition
thesis. If such a prerequisite exists, the Costa Rican experience indicates a child
mortality rate threshold of 100 per 1,000 or higher.
The rate did not need to be as low as, say, 50 for people to adopt family
planning; the fertility transition started at levels substantially higher than this. In
turn, it would not be appropriate to conclude that the fertility transition is not
possible at a rate of, say, 200 per 1,000. The transition did not start in Costa Rica
in 1945 when the child mortality rate was approximately 200 per 1,000 either
because of prevailing mortality conditions or because of the absence of other
conditions, such as the availability of contraceptives or the rising costs of child-
bearing.
MACRO-LEVEL BIVARIATE ASSOCIATIONS
The covariations in fertility and child mortality across geographic units may
cast some light on the nature of the association between these two variables.
These covariations may be studied in a rich data set for 89 Costa Rican counties.
The unit of analysis, the county, is a small geographic unit usually on the order of
20,000 inhabitants, defined on the basis of the Costa Rican administrative divi-
sions, cantones and distritos. Indicators available in this data set are the marital
fertility rate (births per married woman aged 15-44) in 1965,1975, and 1985; and
the child mortality rate (probability of dying before the fifth birthday) lagged 2
years. The numerator for computing the marital fertility rate is a 5-year average
from the country's vital statistics; the denominator is an estimate based on the
1963,1973, and 1984 censuses. The data on births were validated with estimates
obtained by projecting the census populations backward. The child mortality
rates in 1963 and 1973 were estimated using a variation of the Brass method
(United Nations, 1983) on data from the 1973 and 1984 censuses on the propor-
tions of surviving children by mother's age. The child mortality rate in 1983 is a
OCR for page 390
390
AGGREGATE AND MULTILEVEL EVIDENCE FROM COSTA RICA
5-year average from vital statistics corrected by the ratio between the census-
based estimate and a vital statistics estimate in 1973 (no correction was larger
than 20 percent). Eleven counties with seemingly unreliable estimates were
excluded from the original pool of 100 counties (details reported in Rosero-
Bixby, l991b).
Bivariate Macro-Level Associations
Figure 11-2 shows a hypothetical scatterplot for interpreting the data of
bivariate covariations in fertility and child mortality levels. The figure's four
quadrants result from combining high and low child mortality levels with high
and low fertility rates. Most populations should fall in quadrants II and IV, the
Low
Child Mortality
High
l
I
._
._
IL
ll
Empty if
sufficient
condition
"Development is the
best contraceptive"
o
o o
o
O O O
O O
0Go 0
OOcPO O
o
O c
o
o
~ O
o
o o
o
O O O
0 Go 0
OCPO
o
o
00
o
lll
o
0 0°
0 0o
o
Empty if
necessary
condition
"Vertical family planning
programs cannot succeed"
IV
FIGURE 11-2 Expected scatterplot for the causal associations of child mortality on
fertility.
OCR for page 391
LUIS ROSERO-BIXBY
391
high-high and low-low quadrants. Quadrant I of low child mortality and high
fertility is expected to be nearly empty if a decreasing child mortality is a suffi-
cient condition to bring down fertility rates. Quadrant III of high child mortality
and low fertility should not contain observations if lowering child deaths is a
precondition for fertility decline. If real data show a substantial number of
observations in quadrants I or III, the corresponding hypotheses should be re-
jected (although the contrary does not mean that the hypotheses are true). A few
observations in quadrants I and III may occur in real data as a consequence of
measurement errors or effect lags.
How do the Costa Rican counties behave in comparison with this hypotheti-
cal association? Figure 11-3 shows the scatterplots for 1965, 1975, and 1985.
The 1985 marital fertility rate is also plotted against the child mortality rate 22
years earlier to give an impression of the effect of considering lengthened lags. A
striking feature in the figure is the fast pace of change in both fertility and child
mortality. In just 10-year intervals, there are remarkable shifts in the cloud of
observations toward the origin. Although the expected positive correlation oc-
curs in the four cross sections, the correlation becomes weak by 1985.
By 1965, although most counties lay in the high-high zone (quadrant II),
some have moderate child mortality and high fertility rates as counter evidence
for the "sufficient condition" thesis: Birth rates continued to be high in spite of
moderate infant mortality. By 1975, counties with either high mortality or high
fertility are extinct species for practical purposes. Very few counties ever fall in
the region of low fertility and high mortality (quadrant IV). The "necessary
condition" thesis is neither supported nor rejected by these data.
By the mid-1980s, all counties but one had a low child mortality rate of less
than 50 per 1,000. The variation in fertility rates is somewhat broader and the
correlation coefficient is a modest 0.31 a suggestion that at these low levels
there may be little connection between the two variables.
To what extent do child mortality levels in the past influence current fertility
levels? An effect may come about if knowledge of the chances of child survival
acquired in childhood and adolescence (news of child deaths heard at home from
parents and other adults) is what make couples pursue an insurance reproductive
strategy later on life (i.e., have as many children as possible to ensure that some
will survive). Figure 11-3 examines this point by plotting the 1985 marital
fertility rate against a 22-year lagged child mortality rate. The new correlation
coefficient (0.65) is substantially higher than that for the 2-year lagged child
mortality rate (0.31~. This higher association may be, however, just an artifact of
other variables such as the county's level of socioeconomic development. In
particular, less developed counties, which tended to have a high fertility rate in
1985, had high child mortality in the 1960s; these counties, however, do not have
high child mortality in the 1980s, thanks to health programs implemented in
Costa Rica in the 1970s (especially primary health care interventions), which
OCR for page 392
392
~ o
ho.
ll
.______1°f ~ ______
of
~916L {q {~!1!~11~!~W °
rat
o o
o o
I
~o
o o
No
~ (Oo°OooN
\-~300)
o o
No
Ago
~-
o
o
o o o o o o
o o o o o o
(D ~Cr) Cal
996 ~ {q {~!1!~} 1~!~W
to
to
Cal
o
to
to
Cal
Q
. _
o
~5
S
o o o ~ / - ct
11
Oo 1
o
Q
. _
Id
o
~5
S
- o
OCR for page 393
393
(V)
o
11
1 o
o I> 0
o
~986 ~ {q {~!1!~} 1~!~W °
o
o o
1°
o
o
o o
oo
o o
o o
(D ~
o o
o o
~ Cr)
986 ~ {q {~!1!~} 1~!~W
o o
o o
Cal
Cal
.o
Do
o
to
- ~ct)
~o
Q
' ° ~d
o
~5
S
- o
o
_ o
C~
o
~\°
_v
o
. ~o ct)
Q
.~
o
~5
S
- o
oo
~D
C~
.=
o
·_.
Ct
C~
o
·_.
Ct
s~
o
. _.
~ O
~ .~
0=
·= ~
C~ ~
~ -
.0 o
;~ ,=
.=
C=
O ~
O
O
·=
11
OCR for page 394
394
AGGREGATE AND MULTILEVEL EVIDENCE FROM COSTA RICA
erased most of the socioeconomic differentials in infant mortality (Rosero-Bixby,
1986~.
The scatterplot for the 22-year lagged effect shows a substantial number of
counties in quadrant III, which can be taken as evidence against the thesis that
low child mortality during childhood and adolescence is a precondition for con-
trolling fertility later in life.
Multivariate Macro-Level Effects on Fertility Rates
To what extent are the bivariate association examined so far and its fluctua-
tions over time a manifestation of third determinants shared by both child mortal-
ity and fertility? I addressed this question by statistically controlling the effect of
potential confounders in multivariate regression models. In addition to child
mortality, regression equations explaining the general marital fertility rate in-
clude seven indicators of socioeconomic, programmatic, diffusionist, and geo-
graphic conditions at the aforementioned three points in time (unless otherwise
indicated, these indicators are lagged 2 years from the marital fertility rate and
come mostly from the 1963, 1973, and 1984 censuses). Given that these indica-
tors are considered only to control their potentially confounding effects on the
relationship between child mortality and fertility, neither a theoretical construct
nor details about their meaning and operationalization are given here (details in
Rosero-Bixby, l991b).
Multiple regression models estimated for the three cross sections of child
mortality and marital fertility rates (1965, 1975, and 1985) show relatively mod-
est net effects of child mortality lagged 2 and 12 years on marital fertility (Table
11-1~. The net elasticities (i.e., the percentage of change in the marital fertility
rate resulting from a 1 percent change in the child mortality rate) range between
0.02 and 0.18. All but the elasticity for the 2-year lagged child mortality rate in
1985 are statistically significant. As with the bivariate correlation coefficient, the
association weakens in the cross sections before and after the fertility transition.
Although these estimates do control the potentially confounding effect of
other variables in the model, there is no guarantee that the model is fully identi-
fied and thus that all spurious associations have been purged; there is always the
possibility that the child mortality rate is picking up the effect of a confounding
variable that was not included in the model. Regression models on change rates,
rasher then on levels, ameliorate this possibility by purging all of a county's
characteristics and residuals that do not vary over time, such as systematic regis-
tration errors or cultural constants influencing both fertility and child mortality.
(Random and other errors in the data are, however, magnified by the computation
of changes, and the "regression to the mean" phenomenon may introduce consid-
erable noise in the variance of changes (Bohrnstedt, 1969) and can bias toward
zero the estimated effects (Freedman and Takeshita, 1969~.)
OCR for page 400
400
AGGREGATE AND MULTILEVEL EVIDENCE FROM COSTA RICA
course or the 45th birthday, whichever came first. The sample size usable in the
analysis was only 470 women, which means a limited statistical power for detect-
ing small associations (this sample size has less than 80 percent statistical power
for detecting relative rates in the range of 0.6-1.7~.
The "exposure" variable contextual child mortality comes from the 1973
and 1984 censuses, coded into a geographic information system. Although mor-
tality estimates were readily available at the county or even distrito level, conven-
tional geographic units were not considered because of the large heterogeneities
between and within them and the frequent changes in their boundary layout.
Instead, a standard definition of "context," independent of administrative bound-
aries, was adopted, namely, the area within a radius of 1 kilometer in urban areas
and of 5 kilometers in rural areas from the index women's reported place of
residence during most of adolescence. The contextual child mortality was defined
as the cumulative proportion of children who had died to women aged 40-49 who
belong to the respondent's cohort and live in the previously defined context.
Therefore, the proportion of children dead among women aged 40-49 in the 1973
census were assigned to respondents born in 1926-1936 and the proportion from
the 1984 census, to respondents born in 1937-1947.
Other contextual indicators used in the multilevel analysis were:
· Completed fertility. Children ever born per woman aged 40-49 in the
aforementioned census, cohorts, and radius.
· Proportion of households under the poverty line in the aforementioned
radius. Poverty defined with unmet basic needs criteria (absence in a household
of any two of seven minimum items, including running water, sanitation, electric-
ity, dwelling's materials, a kitchen, bedrooms, and a radio or TV set).
· Family planning supply estimated as the per capita density of services
(weighted by the inverse of distance) in a radius of 30 kilometers from the index
household.
At the individual level, the following respondent's characteristics were consid-
ered:
Completed years of formal education.
Wealth index, number of commodities existing in the household by 1984.
Calendar year at first sexual intercourse, which defines the starting point
of the exposure and also time-trend effects.
Age at first sexual intercourse.
Whether the respondent has ever married (a time-varying covariate).
Bivariate and Multilevel Effects
Does the pattern of adoption of family planning differ with contextual child
OCR for page 401
LUIS ROSERO-BIXBY
401
mortality levels? The presence of censored observations in the data calls for the
use of life table techniques to estimate the adoption curves. Figure 11-5 shows the
cumulative adoption curves for six contextual levels of child mortality. These are
cumulative "survival" curves estimated by the Kaplan-Meier method (Kaplan
and Meier, 1958~. It is evident that women from low-mortality contexts adopt
contraception quicker and in higher proportions than women in high-mortality
contexts. The curves suggest three child mortality levels for differentiating the
incidence of contraception: less than 75 per 1,000, 75 to 124, and 125 and higher.
The Kaplan-Meier median waiting time for adopting family planning by these
three groups is 4, 10, and 25 years since first intercourse, respectively (Table 11-
3~. The association could not be clearer.
There are large differences in the timing of the adoption of family planning
related to other macro- and micro-level characteristics, but none appears to be as
important or large as that observed between various child mortality groups (Table
11-3~. However, the level of contextual child mortality itself seems strongly
associated to these other characteristics. For example, the median family plan-
ning adoption time varies from 21 years since first sex among the poor to 3 years
among the wealthy, but the child mortality proportion also varies greatly from
1.0
o 0.8_
o
Q
g
tt 0.6
-
tin
Q 0.4_
o
._
Q 0.2_
IF
IL
0.0
Or ,
C~
~-
~1 ~
r
I
1
CM 50-74r
- , ,
r - - I
CM 75-99
r
~ ~~~ CM 100-124
r - - I
-
CM 150~-
art' 1 ' ' ' ' 1 ' ' ' ' 1
l
' ' ' ' 1 ' ' ' ' 1 ' ' ' ' 1
,
1 CM 125-149
0 5 10 15 20
Years since first sexual intercourse
FIGURE 11-5 Family planning adoption curve by level of contextual child mortality
(CM).
OCR for page 402
402
AGGREGATE AND MULTILEVEL EVIDENCE FROM COSTA RICA
TABLE 11-3 Contextual Child Mortality and Median Duration until Adoption
of Family Planning by Selected Variables
Variable
and Categories
Median Adoption
Yeara
Contextual Child
Mortality
Total 469 10.2 102
Contextual-level
Child mortality
29-74 140 4.0 57
75- 124 209 9.9 98
125-290 120 24.9 164
Family planning supply
1968- 1972
Moderate 200 12.7 123
High 269 9.0 87
Completed fertility
2.7-3.9 108 3.4 59
4.0-5.4 160 8.0 88
5.5-8.9 201 18.4 137
Households under poverty
line
<10% 309 6.5 82
10% or more 160 20.3 143
Individual-level
Education (years)
0-2 135 18.4 125
3-6 231 9.6 99
7 Or more 103 2.6 81
Wealth group
Poor 121 20.6 128
Low 216 11.2 102
Medium/high 132 3.2 80
Birth cohort
1926-1936 231 18.1 130
1937-1947 238 5.9 76
Year first sex
1937-1959 293 15.3 115
1960-1969 139 4.9 83
1970-1983 37 1.9 73
Age at first se'
11-16 98 18.4 112
17-24 274 10.0 103
25-47 97 3.8 92
aKaplan-Meier estimate. Time counted since first sexual intercourse (women aged 38-58 years in
1984, ever sexually active).
OCR for page 403
LUIS ROSERO-BIXBY
403
128 to 80 per 1,000, respectively. Child mortality variation overlaps substan-
tially with all these other variables, and it is thus quite possible that most of the
bivariate association is attributable to these confounding factors. Thus, statistical
control of these confounding effects with multivariate models appears to be man-
datory.
Multivariate and Multilevel Effects
Adjusted effects of contextual child mortality on the individual-level rate of
adoption of family planning were estimated with Cox multivariate regression.
The explanatory variables in the model were categorized to accommodate curvi-
linear effects. Two of the explanatory variables were allowed to vary over time:
contextual family planning supply (zero until 1968 and the average for 1969-
1979 thereafter) and individual marital status. Because preliminary models
showed interaction between cohort and child mortality, the two variables were
combined. Table 11-4 shows the estimated effects as rate ratios of adopting
family planning.
Among the older cohorts of women born in 1926-1936, a contextual child
mortality rate under 125 per 1,000 increases the rate of adoption of family plan-
ning by 51 percent. For the younger cohorts born in 1937-1947, crossing a child
mortality threshold of 75 per 1,000 increases the likelihood of adopting family
planning by 36 percent (2.38 . 1.75 = 1.36~. Net child mortality effects are
statistically significant. The extreme shifts in mortality levels over time makes
them, however, a moving target. For example, studying the effect of crossing the
line of 125 deaths per 1,000 only makes sense for older cohorts because virtually
no one among the younger cohorts has been exposed to a contextual mortality
rate of 125 or higher.
Other variables with significant net effects on the adoption of birth control
are marital status, the year when the observation started, and household wealth, as
well as the contextual level of family planning supply. In contrast with previous
results, there are no significant diffusion effects from contextual fertility in this
data set, perhaps because the explanatory variable is cohort-based, rather than the
period-based total fertility.
Is a moderate contextual child mortality rate a precondition for the adoption
of family planning? The cumulative adoption curves in Figure 11-6 show that a
contextual child mortality rate of 125 or higher may be a serious obstacle for
adopting family planning but it is not an absolute impediment: About 20 percent
of couples in this category have adopted birth control after 10 years of sexual
activity. More interestingly, a Cox model estimated only for contexts where
child mortality is 100 or higher shows that the adoption rate may increase sharply
in these contexts with the supply of family planning services or with the
household's wealth (Table 11-5~. Thus, the obstacle of high child mortality can
be circumvented.
OCR for page 404
404
AGGREGATE AND MULTILEVEL EVIDENCE FROM COSTA RICA
TABLE 11-4 Rate Ratio of Adopting Family Planning Estimated with a Cox
Regression Model
Variable
N Rate Ratio 95% Confidence Interval
Contextual-level
Child mortality,
mothers' cohort
2125, old 109 1.00 Reference Group
<125, old 122 1.51 1.00-2.27
275, young 128 1.75 1.08-2.82
<75, young 110 2.38 1.39-4.06
Family planning supplya
None 527 1.00 Reference Group
Moderate 158 1.76 1.20-2.60
High 176 1.59 1.09-2.34
Completed fertility
2.7-3.9 108 1.00 Reference Group
4.0-5.4 160 0.88 0.64-1.21
5.5-8.9 201 0.99 0.64-1.53
Households under poverty line
<10% 309 1.00 Reference Group
10% or more 160 0.80 0.56-1.13
Individual-level
Education (years)
0-2
3-6
7 or more
Wealth group
Poor
Low
Medium/high
Age at first sex
11-16
17-24
25-47
Marital statusa
Premarital
Ever married
Year first sex
135
231
103
1.00
1.19
1.33
121
216
132 2.57
1.00
1.50
98 1.00
274 1.20
97 1.13
259
602
Reference Group
0.90- 1.57
0.93-1.91
Reference Group
1.10-2.03
1.77-3.73
Reference Group
0.87-1.64
0.68-1.87
1.00
2.47
1.04
Reference Group
1.79-3.40
1.00- 1.07
NOTE: N indicates number of observed segments for these variables.
aTime-varying covariate.
OCR for page 405
LUIS ROSERO-BIXBY
1.0-
~ 0.8_
o
Q
.g
E
in
0 0.4-
~ 0.2_
._
0.0
405
r I
, I _ _
1-- 1
1 1 1
CM < 75 (N=35)
l
, CM75-124(N=71)
r
CM 125+ (N=77)
I , j , , I , , , j I , . . .
.
l
o
5
10
Years since first sexual intercourse
15
20
FIGURE 11-6 Family planning adoption curve among unlikely adopters by level of child
mortality (CM).
Conversely, lowering child mortality seems by itself a factor for increasing
the adoption rate of birth control. Figure 11-6 shows the adoption curves for
women who, according to the regression estimates, were unlikely adopters of
contraception, women with minimum wealth and education and no family plan-
ning services. Among these unlikely adopters, the cumulative adoption curve
clearly shifts upward with lower contextual child mortality.
DISCUSSION
In a series of focus group discussions conducted with Costa Rican women in
their 50s in 1993 (i.e., from the cohorts that lived through the fertility transition),
high child mortality was not perceived as a reason for having large families in the
past nor was its reduction seen as a reason for the shift to the small family of
today (Rosero-Bixby and Casterline, 1995~. Although these discussions focused
on the diffusion of the family planning message, the child survival hypothesis
was explicitly raised by the moderators in all groups. The suggestion that a
decline in child mortality may have played an important role in the fertility
transition did not resonate in the focus groups. However, two possible links with
low child mortality emerged spontaneously in the discussions. Namely, (1) that
OCR for page 406
406
AGGREGATE AND MULTILEVEL EVIDENCE FROM COSTA RICA
TABLE 11-5 Rate Ratio of Adopting Family Planning Estimated with a Cox
Regression Model for Contexts with Child Mortality of 100 or Higher
Variable N Ratio 95% Confidence Interval
Family planning supply
None 260 1.00 Reference Group
Moderate 97 1.92 1.04-3.51
High 57 1.91 0.99-3.72
Wealth group
Poor
Low
Medium/high
Marital statusa
Premarital
Ever married
Year first sex
81
92
36
120 1.00
294 1.84
1.00
1.63
2.20
1.05
Reference Group
1.09-2.44
1.33-3.66
Reference Group
1.08-3.17
1.01- 1.09
NOTES: N indicates number of observed segments for this variable.
aTime-varying covariate.
the family planning message often diffused in waiting rooms of health centers
where increasing numbers of mothers were taking their children for preventive or
curative care:
I heard about family planning for the first time when I brought my sick daugh-
ter to the clinic. I started to listen to the other women. They would say, "Did
you hear from Carmen that they are going to offer family planning here, that
they are going to bring pills." Once I heard a woman telling that she used
condoms and got pregnant. She didn't know what had happened inside, or if
the condom was torn, the thing was that she got pregnant. Those women came
from everywhere (Rosero-Bixby and Casterline, 1995:65~.
(2) that the burden of helping their mothers to rear a large family was a motiva-
tion for wanting a small family. Increased child survival is an obvious reason for
larger families:
My mother had two pairs of twins and a lot of other kids. I would come home
from school and had to go pick coffee and help my mother to sew because we
needed the money to raise so many kids. It was changing diapers and washing
all the time. And I got the idea that having lots of kids was a kind of slavery
(Rosero-Bixby and Casterline, 1995:74~.
Neither the focus group discussions nor the statistical record at the aggregate
and individual levels support the claim that decreasing child mortality is critical
for decreasing fertility. However, decreasing child mortality may facilitate the
fertility transition, and high child mortality may delay the transition.
OCR for page 407
LUIS ROSERO-BIXBY
407
Just as there are developing countries that, in spite of moderate infant mortal-
ity, continue having high birth rates, the decline of child mortality in Costa Rica
during several decades did not affect fertility trends. The data show that one
cannot expect that crossing a child mortality threshold of 200 or 150 per 1,000
will automatically bring about fertility decline. Not even falling below 100 per
1,000 child deaths will generate an automatic response. In short, decreasing child
mortality does not appear to be a sufficient condition for fertility decline, nor can
the Costa Rican fertility transition be explained solely in terms of an adjustment
process to moderate child mortality rates.
The data are inconclusive regarding the thesis that reduced child mortality is
a condition for fertility decline. Supporting the thesis is the fact that practically
no Costa Rican county, nor for that matter population in the world (Hanson et al.,
1994), has experienced low fertility and high child mortality simultaneously.
This statement is, of course, conditional on what one considers high child mortal-
ity. If one draws the line at a child mortality rate of 100 per 1,000 or higher, it is
indeed almost impossible to find populations with controlled fertility. A closer
look at the data show that in a substantial number of Costa Rican communities the
onset of the fertility transition occurred at child mortality levels above 130 per
1,000. Moreover, individual-level data of the cohorts that changed fertility in
Costa Rica show that the rates of adoption of family planning in contexts of high
child mortality were far from zero. More interestingly, the data for those contexts
show that adoption of family planning increases sharply with the presence of
such conditions as family planning services and higher living standards. The
obstacle of high child mortality does not seem impossible to beat.
Evidence suggesting that moderate child mortality may facilitate the fertility
transition comes from the significantly earlier transition onset in counties with
lower child mortality, as well as from the earlier adoption of family planning
among women from low-mortality contexts. These effects persist after control-
ling for the potentially confounding effect of standards of living, education,
supply of family planning services, and the like.
To discuss the causal mechanisms behind this association one should first
understand how reproductive decisions are made before and during the fertility
transition. Following Fishbein's theory of reasoned action, reproductive behav-
ior may be shaped by "the person's beliefs that the behavior leads to certain
outcomes and his evaluation of these outcomes," and "the person's beliefs that
specific individuals or groups think he should or should not perform the behavior
and his motivation to comply with the specific referents" (Fishbein and
Middlestadt, 1987:363~. The discussion about replacement and insurance strate-
gies for having children (or before this, for having sex or using contraceptives)
assumes the perfectly rational type of behavior implicit in the belief that behavior
leads to predictable outcomes. It is probable, however, that reproduction, like
many human actions, is not based mainly on day-to-day conscious decisions but
guided by cultural norms and reference groups. High fertility as a response to
OCR for page 408
408
AGGREGATE AND MULTILEVEL EVIDENCE FROM COSTA RICA
high mortality may be implicit in these cultural norms. The distinction between
replacement and insurance strategies does not seem meaningful at this collective
level. Under a routine-dictated behavior, situations may occur in which an
individual's self-interest clashes with the prevailing cultural precepts and leads
the person to exercise conscious decision making. The onset of fertility transition
could be one of these situations. Enlarged families resulting from improved child
survival rates could be one of the reasons for questioning routine reproductive
behavior. The second quote above from a Costa Rican woman suggests this
possibility.
Although the Costa Rican experience cannot be extrapolated uncntically, the
findings in this chapter suggest the following three policy considerations:
1. It would be a mistake for a government to expect an automatic fertility
decline to follow a fall in child mortality. The latter does not seem to be a
sufficient condition for the former.
2. High child mortality is not a good reason for not providing family plan-
ning services since a sizable proportion of couples may adopt family planning in
contexts of high child mortality.
3. Fertility reductions are more likely to occur and family planning pro-
grams more likely to succeed in contexts where child mortality is low. A family
planning intervention coupled with a child survival program will probably have
more effect than a vertical program of only family planning services.
REFERENCES
Blacker, J.G.
Behm, H., and J.M. Guzman
1979 Diferencias socioeconomicas del descenso de la fecundidad en Costa Rica 1960-1970.
Pp. 158-183 in Se'timo Seminario Nacional de Demograf~a. San Jose, Costa Rica:
Direccion General de EstadIstica y Censos.
1987 Health impacts of family planning. Health Policy and Planning 2(3): 193-203.
Bohrnstedt, G.W.
1969 Observations on the measurement of change. Pp. 1113-1136 in E.F. Borgatta, ea., Socio-
logical Methodology. San Francisco: Jossey-Bass.
Caja Costarricense de Seguro Social
1994 Fecundidad y Formacio'n de la Familial Encuesta Nacional de Salud Reproductiva 1993.
San Jose, Costa Rica: Caja Costarricense de Seguro Social.
Carlsson, G.
1966 The decline of fertility: Innovation or adjustment process. Population Studies 20(2): 149-
174.
Cochrane, S.H., and K.C. Zachariah
1983 Infant and Child Mortality as a Determinant of Fertility: The Policy Implications. World
Cox, D.R.
Bank Staff Working Paper no. 556. The World Bank, Washington, D.C.
1972 Regression models and life tables. Journal of the Royal Statistical Society Series B
34: 187-202.
OCR for page 409
LUIS ROSERO-BIXBY
409
Davis, K.
1955 Institutional patterns favoring high fertility in underdeveloped areas. Eugenics Quarterly
2(1):33-39.
Fishbein, M., and S.E. Middlestadt
1987 Using the theory of reasoned action to develop educational interventions: Applications to
illicit drug use. Health Education Research 2(4):361-371.
Freedman, R., and J.Y. Takeshita
1969 Family Planning in Taiwan: An Experiment in Social Change. Princeton, N.J.: Princeton
University Press.
Haaga, J.G.
1989 Mechanisms for the association of maternal age, parity and birth spacing with infant
health. Pp. 96-139 in A.M. Parnell, ea., Contraceptive Use and Controlled Fertility:
Health Issues for Women and Children. Committee on Population, National Research
Council. Washington, D.C.: National Academy Press.
Hanson, L.A., S. Bergstrom, and L. Rosero-Bixby
1994 Infant mortality and birth rates. Pp. 37-48 in K.S. Lankinen, ea., Health and Disease in
Developing Countries. London: MacMillan Press.
Heer, D.
1966 Economic development and fertility. Demography 3(2):423-444.
Heer, D., and V. RodrIguez
1986 The Impact of Child Mortality Levels on Fertility Behavior and Attitudes in Costa Rica.
Unpublished research report, The Population Council, New York.
Irwin, K.L., L. Rosero-Bixby, M.W. Oberle, N.C. Lee, A.S. Whatley, J.A. Fortney, and N.G.
Bonhome
1988 Oral contraceptives and cervical cancer risk in Costa Rica: Detection bias or causal
association? Journal of the American Medical Association 259(1):59-64.
Kaplan, E.L., and P. Meier
1958 Nonparametric estimation from incomplete observations. Journal of the American Statis-
tical Association 53:457-481.
Knight, R.
1995 The Diffusion of Information and Adoption of Contraception in Costa Rica. Chapter
presented at the International Seminar on the Population in Central America. Universidad
de Costa Rica, San Jose, Costa Rica, October 21.
Knodel, J.
1978 European populations in the past. Family-level relations. Pp. 21-45 in S.H. Preston, ea.,
The Effect of Infant and Child Mortality on Fertility. New York: Academic Press.
Lee, N.C., L. Rosero-Bixby, M.W. Oberle, C. Grimaldo, A.S. Whatley, and E.Z. Rovira
1987 A case-control study of breast cancer and hormonal contraception in Costa Rica. Journal
of the National Cancer Institute 79(6): 1247- 1254.
Lloyd, C.B., and S. Ivanov
1988 The effects of improved child survival on family planning practice and fertility. Studies
in Family Planning 19(3): 141-161.
Mauldin, W.B, B. Berelson, and Z. Sykes
1978 Conditions of fertility decline in developing countries, 1965-1975. Studies in Family
Planning 9(5):89-147.
Mensch, B.
1985 The effect of child mortality on contraceptive use and fertility in Colombia, Costa Rica,
and Korea. Population Studies 39(2):309-327.
Notestein, F.W.
1953 Economic problems of population change. Pp. 13-31 in Proceedings of the EightInterna-
tional Conference of Agricultural Economists. London: Oxford University Press.
OCR for page 410
410
AGGREGATE AND MULTILEVEL EVIDENCE FROM COSTA RICA
Pebley, A., and S. Millman
1986 Birth spacing and child survival. International Family Planning Perspectives 12(3):71-
79.
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.
Rosero-Bixby, L.
1986 Infant mortality in Costa Rica: Explaining the recent decline. Studies in Family Planning
17(2):57-65.
1991a Socioeconomic development, health interventions, and mortality decline in Costa Rica.
Scandinavian Journal of Social Medicine 46(Suppl. N):33-42.
l991b Interaction Diffusion and Fertility Transition in Costa Rica. Unpublished Ph.D. disserta-
tion, University of Michigan, Ann Arbor.
Rosero-Bixby, L., and J. Casterline
1994 Interaction diffusion and fertility transition in Costa Rica. Social Forces 73(2):435-462
1995 Difusion por interaccion social y transicion de la fecundidad: Evidencia cuantitativa y
cualitativa de Costa Rica. Notas de Poblacio'n 61:29-78.
Rosero-Bixby, L., and M.W. Oberle
1989 Fertility change in Costa Rica 1960-84: Analysis of retrospective lifetime reproductive
histories. Journal of Biosocial Science 21(4):419-432.
Rosero-Bixby, L., M. Gomez, and V. RodrIguez,
1982 Determinantes de la Fecundidad en Costa Rica. Analisis Longitudinal de Tres Encuestas.
Costa Rica: Direccion General de EstadIstica y Censos, World Fertility Survey, Interna-
tional Statistical Institute.
Rosero-Bixby, L., M.W. Oberle, and N.C. Lee
1987 Reproductive history and breast cancer in a population of high fertility, Costa Rica, 1984-
5. International Journal of Cancer 40(6):747-754.
Rutstein, S.O., and V. Medica
1978 The Latin American experience. Pp. 93-112 in S.H. Preston, ea., The Effects of Infant
and Child Mortality on Fertility. New York: Academic Press.
Stycos, J.M.
1982 The decline of fertility in Costa Rica: Literacy, modernization, and family planning.
Population Studies 36(1):15-30.
Taylor, C., J. Newman, and N. Kelly
1976 The child survival hypothesis. Population Studies 30(2):263-278.
Tin Myaing Thein, and J. Reynolds
1972 Contraception in Costa Rica. The Role of the Private Sector, 1959-1969. San Jose, Costa
Rica: Asociacion Demografica Costarricense.
Trussell, J., and A. Pebley
1984 The potential impact of changes in fertility on infant, child, and maternal mortality. Stud-
ies in Family Planning 15(6) :267-280.
United Nations
1983 Manual X: Indirect Techniques for Demographic Estimation. New York: Department of
International Economic and Social Affairs.
1985 Socio-economic Development and Fertility Decline in Costa Rica. New York: Depart-
ment of Economic and Social Affairs.
Representative terms from entire chapter:
family planning
