Tim Miller > Curriculum Vitae > Malnutrition and mortality among Bolivian children > Chapter Three

THE HEALTH OF BOLIVIAN CHILDREN: 1960-1990

Are Bolivian children healthier today than those born a generation ago? If so, what has accounted for this improvement? These questions are addressed here to provide a context for the examination of the health of Bolivian children in the late 1980s.

Extreme poverty in Bolivia leads to a lack of societal resources for both the improvement of child health and the monitoring of such improvements. Consequently, we know very little about what has been happening to Bolivian children in the last thirty years. Nevertheless, there are three reasons to present the scant historical information we have. First, certain broad patterns emerge from these disparate data sources which inform the subsequent analysis of the 1989 Demographic and Health Survey (ENDSA). Second, it is important for the reader to bear in mind our general level of ignorance as a consequence of this paucity of data. By remaining mindful of our ignorance, we reduce the likelihood of mistaken judgments and policies. Third, even though it is difficult to draw conclusions from these data, it is important to think carefully about the nature of possible explanations for improvements in health during a health transition.

What is meant by the term "health?" How can we tell if the health of Bolivian children has improved? In this dissertation, a healthy population of children is defined as one experiencing low levels of malnutrition, morbidity, and mortality. A decline in these measures over time would demonstrate that the health of Bolivian children has improved. We turn now to an examination of the changes in malnutrition, morbidity, and mortality over the last thirty years.

Malnutrition

Protein-energy malnutrition is often gauged through physical measurement of children. When judged against appropriate standards, height-for-age, weight-for-age, weight-for-height, and arm circumference are effective measures of growth retardation. These measurements are commonly, and somewhat incorrectly, referred to as measuring malnutrition. In most developed countries, the chief cause of growth retardation is protein-energy malnutrition. There are other contributing causes in Bolivia, however, such as hypoxia in high-altitude environments, specific nutrient deficiencies, and frequent illness episodes. Hence, in certain circumstances, these measurements might be better thought of as measuring hypoxia or diarrheal incidence than measuring malnutrition. This should be kept in mind in the discussion that follows.

Height-for-age, weight-for-age, and weight-for-height measurements were made in two large Bolivian surveys taken at the beginning and end of the 1980s. The Encuesta sobre Estado Nutricional de la Población took physical measurements of about 6,000 children aged 6 - 59 months in 1980. The Demographic and Health Survey (ENDSA) took physical measurements of about 2,600 children aged 3-36 months in 1989.

These physical measurements are compared to an international standard. (Issues surrounding this standard and its use in defining malnutrition will be discussed in Chapter 4). Children who are more than two standard deviations below their respective standard's median are classified as stunted (low height for age), underweight (low weight for age), or wasted (low weight for height). Stunted children are likely to have suffered from chronic, long-term growth retardation. Stunting is often referred to as mild malnutrition or undernutrition. Underweight children are likely to have suffered from a recent weight-loss from sudden food shortage or recent illness. It is a measure of short-term growth difficulties. Wasted children are suffering from chronic and severe growth retardation. This is often referred to as severe malnutrition.

These two surveys are not strictly comparable. As noted above, different aged children were surveyed. Figure 3.1 presents the measurement statistics for these two surveys. The proportion of children stunted appears unchanged over the decade, at roughly 40%. There appear to be dramatic increases in underweight and wasted children. The proportion of children classified as wasted increased 8-fold (from 0.2% to 1.6%). The proportion underweight increased 13-fold (from 1% to 13.3%).

Could the nutritional status of Bolivian children have deteriorated over the decade? As noted above, the differences in age groups sampled makes comparison of the two surveys difficult. Other evidence consistent with a deterioration in health status is provided by Morales and Rocabado (1988) who report an increase in low birthweight babies born in hospitals in the city of La Paz during the late 1980s.

There are no data sources which directly measure malnutrition prior to the 1980s. Indirect estimates of malnutrition based upon the food purchasing power of wages indicate worsening nutrition since the mid-1970s. A joint study (ORSTROM, 1986) by the Bolivian Instituto de Alimentación y Nutrición (INAN) and the French Office de la Recherche Scientifique et Technique Outre - Mer (ORSTOM) revealed a 13% decline in the purchasing power of the average industrial wage between 1975 and 1983 and a startling 32% decline in the food purchasing power of the minimum wage between November 1982 and January 1985 during Bolivia's recent economic crisis.

In summary, it appears that the major form of malnutrition in Bolivia is mild malnutrition or chronic undernutrition as reflected in the 40% of children classified as stunted by height-for-age data. We lack good data for a time-series on nutrition. The data we do have supports the view that child nutrition levels during the 1980s remained unchanged or perhaps worsened.

There are three types of data sources for morbidity: hospital and clinic records, health interview surveys, and infectious disease registration systems. All are subject to extreme bias in Bolivia. Hospital and clinic records are biased because they represent only those Bolivian children who have access to health facilities. Health interview surveys are biased as they are subject to recall errors, and diagnosis is based on parental observation combined with personal perceptions of what constitutes illness and disease. Infectious disease registration systems are very incomplete in Bolivia. Observed changes in disease reporting are as likely to reflect changes in the completeness of the registration system as changes in the actual incidence of diseases.

The Bolivian Ministry of Health published a list of leading causes of visits to Ministry health facilities for infants and children in 1970 (see Figure 3.2). From these records, diarrhea and respiratory problems are the leading causes of health facility visits in 1970, accounting for 67% of illness among infants and 50% of illness among 1 to 4 year-olds. The other leading causes of illness in 1970 were nutritional deficiencies and measles. The full impact of the latter on morbidity is not reflected in these statistics. Both nutritional deficiencies and measles are likely to contribute to diarrhea and respiratory illnesses.

Unfortunately, the Ministry has not published any further statistics on health facility visits. A health interview survey conducted in 1983 can be used for comparison. The Encuesta de Prevalencia de Medicamentos of 1983 provides data on illness since birth for children under 5. The data are presented in Figure 3.3. Here, we note that diarrhea and respiratory illnesses (such as whooping cough and pneumonia) continue to be leading causes of child morbidity.

The EPM-1983, because it is a health interview survey, is subject to two principal biases. First, those diseases which require laboratory testing or trained clinical observations will be under-reported in the surveys and will be disproportionately reported by those parents wealthy enough to seek out medical care or by parents in areas where there is high access to medical care. Second, personal differences in perceptions of when a child is sick and what the child is suffering from are often based on shared cultural health beliefs. In Bolivia, this may mean that observed differences in disease among population groups are

due in large measure to cultural differences in conceptualization of illness and disease rather than to differences in ecology, standard of living, or access to medical care.

The registration system for infectious disease morbidity maintained by the Dirección Nacional de Epidemiología provides a time series on morbidity changes since 1965. A steep rise and decline in the number of cases of measles and whooping cough is evident in Figure 3.4. These reflect changes in the reported cases of measles and whooping cough due to an actual change in incidence, a change in propensity to report cases, or some combination of the two. Hence, it is hazardous to infer changes in incidence based on this data. For example, decline after 1982 occurred prior to mass immunization programs. During the early 1980s, the Bolivian economy was in crisis, undergoing a hyperinflation coupled with a severe contraction. A likely explanation of this decline is the failure of this registration system during the crisis.

In summary, an overview of morbidity shows the importance of diarrheal and respiratory diseases as sources of child illness. The full impact of immuno-preventible diseases remains unknown.

Mortality

What do we know about changes in causes of death among children over this period? Because of the poor vital registration system, there are no good time series data on these changes. The vital statistics system in Bolivia, as in most developing countries, records only a fraction of the total deaths occurring in the country. In 1981, the vital registration

system recorded 11,883 deaths to children under 5 in Bolivia. This is about 40% of an estimated 31,000 deaths to children under 5.

Shown in Figure 3.5 are lists of causes of death from a variety of sources (hospital and clinic records, civil registration, and a health interview survey). There are so many differences between these sources that it is impossible to compare the disease distribution of the early 1980s and late 1980s. Diarrhea, respiratory, and birth problems are consistently among the leading causes of death. These account for half to 3/4 of all deaths in the four data sources. Surprisingly, diseases preventable by immunization do not appear to play a large role. It is difficult to gauge their impact using statistics on cause of death since they are often a secondary cause of death. Many diseases such as measles compromise the body's immune system so that the child is very susceptible to death from diarrhea or pneumonia.

In lieu of vital registration information on births and deaths, surveys and censuses can be used to determine levels and trends in infant and child mortality in the country. There are 5 sources for a time-series on mortality: la Encuesta Demográfica Nacional de 1980 (EDEN I), el Censo Nacional de Población y Vivienda de 1976 (CNPV 76), la Encuesta Demográfica Nacional de 1980 (EDEN II), la Encuesta Nacional de Población y Vivienda de 1988 (ENPV 88), and la Encuesta Nacional de Demografía y Salud de 1989 (ENDSA 89).

In all these sources, information was collected on age of the mother, number of children born, and number of children dead. This allows for estimation of child mortality by the use of Brass and Feeney indirect estimation techniques. Empirically, these techniques

have been shown to work rather well for establishing general levels and trends. Using DHS surveys, estimates derived from these techniques can be compared with direct estimates of child mortality from maternal histories. Hill (1992) recently reviewed several such comparisons and found no major biases in the method.

National estimates of infant mortality (1q0) and toddler mortality (4q1) are presented in Figures 3.6 and 3.7 using Brass and Feeney indirect estimation techniques. Both the 1q0 and 1q4 are derived from data by assuming a fixed age-pattern of mortality (see Appendix 3.1). Therefore, such numbers cannot be used to test for the hypothesis that the age-structure of mortality has changed over time. Some fear that many health interventions aimed at infants simply serve to postpone death until shortly after infancy. Hence, one might expect to see infant mortality rates falling rapidly while mortality in the second year of life increases. In this data, if child mortality is improving, both estimates of 1q0 and 4q1 must fall together, because the age structure of mortality is assumed to be constant.

There are two interesting points to be made. Most obviously, why don't the curves derived from the various sources match? Why don't estimates of infant mortality for the mid-1970s based on the 1988 ENPV match those based on the 1976 CNPV? In general, there are biases expected from the indirect estimation method for mortality estimates derived from the responses of older women and younger women. It may be that women in general tend to underreport deaths of children. If this is the case, then older women will tend to forget more of their children's deaths than younger women. Hence, child mortality estimates for about 20 years before the census which are derived from the reports of older women will be underestimated. The reports of young women (age 15-24) are used to provide recent estimates of child mortality. Child mortality typically follows a U-shaped curve with children of very young and very old mothers facing higher

risks. This means an infant mortality estimate derived entirely from young mothers' reports will be an overestimate. Therefore, the indirect estimation method will yield trends in infant mortality that need to be raised at their most distant tail and lowered at their most recent tail.

The second point, made stronger by the presumed bias in these estimates, is that child mortality is falling. The figures indicate a steady decline in infant mortality beginning as early as the late 1960s, in full swing by the 1970s, and accelerating throughout the 1980s.

Using data from the 1976 Population Census and the 1988 Population Survey, we can analyze changes in the infant mortality rate over the decade. Table 3.1 presents the infant mortality rates for children based upon education of the mother (none, basic, intermediate, middle or more), area of residence (rural or urban), and region of residence (Altiplano, highland valleys, or tropical lowlands) for 1976 and 1988. Within each group, excess mortality risk is calculated by dividing the infant mortality rate by the lowest of the group rates. By comparing changes in this excess risk between 1976 and 1988, those groups which most benefited from the decline can be determined.

Excess Excess

IMR Risk IMR Risk

TOTAL 153 102

urban 125 1.00 83 1.00

rural 171 1.37 120 1.45

altiplano 152 1.29 110 1.38

valleys 173 1.47 109 1.36

tropics 118 1.00 80 1.00

none 185 2.68 128 2.17

basic 146 2.12 116 1.96

interm 104 1.51 93 1.58

middle 69 1.00 59 1.00

There was virtually no change in urban/rural differentials. In 1976 infants born in rural areas faced a risk 37% greater than those born in urban areas. In 1988, they faced a risk 45% greater. There were modest changes noted between various regions. Compared with the lowest mortality risk group, those in the tropical lowlands, the higher risk faced by those on the Altiplano increased from 29% to 38%. There has been a modest improvement in the health risks faced by infants of the highland valleys compared with those of the tropical lowlands. In 1976 they faced an increased risk of 47%; which declined to 36% by 1988.

The biggest changes are seen with respect to education of the mother. In 1976, infants of mothers with no education faced a mortality risk 2.7 times greater than that of infants of mothers with at least a middle school education. In 1986 this had declined to 2.2 times the risk. This relative risk represents an enormous social inequality. Yet, this great decline is very noteworthy. These data are represented graphically in Figure 3.8. The steepness of both the 1976 and 1988 curves reflect the extreme degrees of social inequality. The flatness of the 1988 curve with respect to the 1974 curve illustrates the growing equalization in health.

To summarize, indirect estimation techniques for child mortality are the most reliable sources of information on changes in the health of Bolivian children in the last 30 years. These data indicate a steady decline in child mortality during this time. The decline cannot be dated accurately, but certainly the decline was well underway in the mid 1970s and continues strongly to this day. The decline occurred equally throughout the country; similar rates of decline were found between urban and rural areas and between the three major ecological zones. It appears that the rate declined much more rapidly among the poorly educated. The leading causes of child death throughout this period have been diarrhea, respiratory illness, and birth traumas, and to an unknown extent immuno-preventable diseases such as measles. Morbidity data also show diarrhea and respiratory diseases as the leading causes of child sickness in Bolivia. Trends in the incidence of immuno-preventible diseases could not be assessed due to presumed variation in the quality of the registration system over time. Malnutrition data point toward chronic undernutrition as the main form of malnutrition in Bolivia. Good time series data on these changes are lacking, but there is no evidence of improvement and some evidence of worsening nutrition among Bolivian children throughout the 1980s.

What accounts for the improvement in health as reflected in the mortality decline? From the mid-1970s to the mid-1980s the infant mortality rate in Bolivia declined from 140 to 100 deaths for every 1,000 births. This has meant that over the decade an approximately 5,000 children survived each year. Currently, there are 75,000 Bolivian children and teenagers who owe their lives to this improvement in health. Why are these people alive? What has changed?

There has been a long running dispute among medical historians, demographers, and other social scientists over the causes of mortality decline. In its crudest form, the debate pits improvements in nutrition, income, and the general standard of living against improvements in social organization, public health, and scientific knowledge. A chief proponent of the former was McKeown (1976). In assessing the causes of the rapid mortality decline in Europe during the late 19th century, McKeown noted the limited roles played by medical and public health activities. Whooping cough, measles, and scarlet fever all began to decline prior to medical immunizations. McKeown further claimed that the reductions in mortality were principally due to a decrease in airborne diseases such as TB and measles rather than decreases in waterborne and foodborne diseases against which public health measures are more effective. He observed that during this period exposure to such airborne pathogens likely increased due to crowded living conditions accompanying industrialization. Therefore, he deduced that resistance to disease must have increased, most likely due to nutritional improvements.

His analysis can be criticized in several ways. First, he clearly understates the impact of medical and public health activities. Smallpox and diphtheria vaccinations were quite successful medical interventions of the period. And much of the mortality decline was due to reduction in deaths due to diarrhea, which is precisely the type of illness against which public health measures are most effective. Second, his argument is based upon deductive reasoning. He fails to establish that nutritional improvements ever occurred over the period or that these were spatially and temporally correlated with the mortality decline. The central problem is a lack of nutritional data for the period. To the extent that trends in heights in a population reflect changes in nutrition rather than changes in disease environment, support for McKeown's thesis is given by a study on historical trends in heights of British military recruits which demonstrate increases in height during this period (Flood, Wachter, and Gregory, 1990).

Arguments in support of the roles played by social organization, public health, and general scientific advance are seen in the works of Preston and Caldwell. Preston (1980), in an analysis of the positive relationship between life expectancy at birth and income per capita in the 20th century, observed a distinct upward shift in the curve between the 1930s and 1960s. He proposed that this shift in the relationship between income and health was due to changes in the efficiency of public health technology. It was thus changes in technology rather than income which accounted for the majority of health improvements in the 20th century. Caldwell (1986) noted that there are several countries with very low per capita income which nevertheless have relatively low mortality. These countries are distinguished from high mortality countries by their provision of local health services, high levels of education among women, and the political will to devote the appropriate expenditures to the cause of mortality reduction.

Are the improvements in health now underway in Bolivia principally the result of nutritional improvements, of economic development, or of a general elevation in the standard of living? As noted earlier, the nutritional status of the average Bolivian child has not improved over time. In fact, it is possible that it has worsened throughout the 1980s. Hence, the Bolivian mortality decline is exceptional in that it has occurred seemingly independently of any nutritional improvement.

What of changes in the standard of living over time? Using a restrictive definition of standard of living as reflected in the gross domestic product per capita, we can assess these changes. Shown in Figure 3.9 is the course of the Bolivian gross domestic product (GDP) since 1948. The general upward trend seen in the post-war period is typical, part of a worldwide economic growth trend. The abrupt decline in the late 1970s was part of the global recession attributable to oil price shocks. The early 1980s witnessed a severe crisis in the Bolivian economy, a huge hyperinflation reaching a peak annualized rate of 28,000% in August 1985 coupled with a severe economic contraction. The economy did not grow at all during this period. Population has grown steadily: at 2.2% in the 1950s, 2.3% in the 1960s, 2.5% in the 1970s, and 2.7% in the 1980s. Also shown in Figure 3.9 is GDP per capita since 1970s. Here one notes the surprising result that though the total economy in 1987 was double its size of 1957, the average standard of living of Bolivians

in 1987 was lower than that of Bolivians two decades earlier. While improvement in GDP per capita may account for the initial declines observed in child mortality, it cannot explain the continued decline throughout the late 1970s and 1980s.

The GDP per capita is a poor measure of changes in the standard of living for several reasons. First, it is a measure of the modern, monetized economy and therefore does not measure activity that is not monetized or measured. Three important examples in Bolivia are the natural environment, the traditional economy, and the black market economy (most notably production associated with cocaine trafficking). Second, GDP per capita is an average and does not measure changes in the distribution of income over time. Third and most importantly, many of the forces of economic modernization are reflected in changes in the stock of personal and public wealth which are not measured in the GDP per capita. A notable example is the building of public infrastructure such as roads. To an isolated village the most significant change in the standard of living may be the building of a road. Although per capita income may have returned to its 1950 level, it is certainly not true that village life in Bolivia today is the same as it was in 1950. Thus, owing to data limitation, general economic modernization as a factor in mortality decline cannot be examined.

At the start of the 1980s, the United Nations declared it to be the decade of clean water. Provision of safe drinking water and adequate sanitation for everyone was declared to be attainable within the decade. Nevertheless, public health and sanitation efforts in Bolivia, as in most developing countries, were minimal in the 1980s and universal access to safe water and sanitation remained a difficult and distant prospect. The percentage of households with access to safe water and adequate sanitation remained virtually unchanged between 1976 and 1988. Reduction in waterborne or water-washed pathogens such as those responsible for diarrhea is unlikely to have occurred through changes linked to the provision of water and sanitation.

The expansion of primary health care has been linked to the declines in child mortality observed in Costa Rica (Roxero-Bixby, 1986) and Nicaragua (Sandiford, et. al., 1991). The case of Nicaragua is particularly relevant. There, the mortality decline certainly occurred during a period of declining living standards during the civil war. Sandiford, et.al., after documenting the IMR decline, proceeded to investigate several possible causes: income, nutrition, breastfeeding, maternal education, immunizations, health services, and provision of water and sanitation. Of these, they concluded that it was the expansion of primary health care which accounted for the decline. They lacked direct measures of such changes but report a decreasing number of doctors per hospital bed, which they deduced meant a shift of doctors from hospitals to the clinics where primary health care takes place.

Similar problems in measuring changes in the health infrastructure exist in Bolivia. The measures we do have are consistent with an expansion of primary health care coinciding with the observed mortality decline. The changes in the numbers of hospitals, and in the numbers of health centers, medical posts, and sanitary posts are compared in Figure 3.10. There was a dramatic expansion of the latter during the 1970s and early 1980s. Changes in health manpower also confirm this expansion of primary health care. Shown in Figure 3.11 are the changes in the number of doctors, trained nurses, and auxiliary nurses

employed in the public health sector. One notes the increase in the number of auxiliary nurses relative to doctors and trained nurses in the public health sector.

Although such measures are consistent with an expansion of public health care, this is unlikely to be the leading cause of the Bolivian mortality decline. First, the modern medical care system affects the lives of few Bolivians. Even if primary health care expanded dramatically over the period, current estimates are that only about 1/3 of Bolivians receive adequate basic health care. (AID, 1988).

Second, much of the health infrastructure is of poor quality. Many facilities are inadequate, lacking basic water and sanitation hook-up. Equipment is frequently broken and poorly maintained, medical supplies are often severely limited, operating hours are erratic, clinics are understaffed, and medical staff often lack proper training to ensure safe and effective care. A frequently cited reason among campesinos for choosing clinics as a last resort is their impression of clinics as extremely dangerous places. This perception of danger is partially unfounded: as the clinics are used as a place of last resort, children are brought in so gravely ill that they cannot be saved. Nevertheless, the clinics' reputation is often well deserved owing to the lack of training among medical personnel.

Third, the most effective preventative and curative modern medical interventions were not implemented until the mid-1980s. Mass immunizations against measles, diphtheria, whooping cough, polio, and tuberculosis begun in the mid-1980s as part of the World Health Organization's Universal Child Immunization Campaign are now beginning to bring these diseases under control. Similarly, the widespread use of oral rehydration therapy starting in the mid-1980s, coupled with Bolivia's successful educational campaign against cholera in the early 1990s are just beginning to have a large impact on child health. Due to these efforts it is likely that substantial further declines in infant mortality are underway and will continue throughout the 1990s. The case for expansion of primary health care as a causal factor in Bolivia's mortality decline to date is a weak one, although, this expansion will no doubt be a factor in an increasingly rapid decline in infant mortality throughout the next decade.

Due to strong opposition from political and religious organizations, family planning efforts are minimal in Bolivia. Political authorities seem to be motivated by the belief that high levels of unwanted childbearing are good for the country: a larger population is equated with a larger economic and military power base. Many political authorities believe that the country is sparsely populated in many regions and that, in order to develop economically, Bolivians must settle these areas. In addition, the history of Bolivia is one of continued loss of territory to her neighbors, which has lead to the hope that population settlement in frontier areas will prevent further loss of territory. The Catholic Church in Bolivia is a major political and religious force in the country. Its opposition to family planning mainly has been effective in limiting access to family planning through its considerable influence on governmental policy. It appears to be less effective in reducing the demand of women for family planning. According to the recent DHS survey, only about 20% of non-contracepting, sexually active women cited religious reasons for not using family planning.

In the face of high perceived need for family limitation and in spite of limited availability of modern family planning, Bolivia has begun the fertility transition. The TFR in Bolivia has steadily declined from an estimated 6.6 children per woman in 1970 to 4.9 children per woman in 1990 (Guzman, Torrez, and Schkolnik, 1991). The biggest decline occurred in the early 1980s during the severe economic crisis. Figure 3.12 displays this transition.

An examination of the proximate determinants of fertility can aid in better understanding this transition and its impact on infant mortality. One approach is the Bongaarts method which can be used to quantify the extent to which fertility is reduced from its maximum level by three key proximate determinants. Guzman, Torrez, and Schkolnik (1991) applied this method to Bolivia using data from the 1989 Demographic and Health Survey. The main determinant was found to be lactation and abstinence (54%), followed by marriage patterns (24%), followed by contraceptives (22%). The rather limited role played by contraception is expected given that only 30% of the reproductive-aged population used any method of contraception (mainly abstinence) and only 12% were using modern methods (mainly IUD, pill, and female sterilization).

What has happened to these three main determinants over time? Could changes in these three determinants have caused the observed fertility reduction? A survey on contraceptive use in 1983 revealed roughly 26% using any method and 11% using a modern method. Hence, I would conclude that changes in contraceptive use could not have accounted for the observed decline. With respect to marriage, the mean age at first marriage has remained unchanged around age 20. This is partial evidence that despite the economic crisis, marriage patterns have largely been unchanged. Third, lactation and abstinence are the major determinants of fertility levels in the country. We lack data on changes in breastfeeding behavior over time. With modernization the average length of breastfeeding has probably declined and hence may not explain the observed fertility decline.

Two other factors in the Bongaarts model: coital frequency and induced abortion are possible explanations for the observed decline. It is possible that levels of migration in search of work increased during the economic crisis of the 1980s. Often, such migrations mean separation of spouses and decline in coital frequency for considerable periods. It is possible this may have contributed to the observed fertility decline.

The most likely explanation of the decline is an increase in induced abortion in response to the economic crisis. As noted earlier, the economic crisis in Bolivia was severe and likely to have dramatically impacted the ability of Bolivian families to feed their children. Women may have responded to such desperate circumstances by seeking abortions. An estimate of 1.5 lifetime abortions per woman was calculated from the Bolivian DHS using indirect estimation techniques (Foreit and Nortman, 1992). This implies that one out of every five pregnancies is aborted, making induced abortion the major method of conscious fertility control in Bolivia. That abortion -- an illegal and frequently dangerous operation in Bolivia-- is used so frequently reflects the desperate situation in which Bolivian women find themselves.

Fertility reduction may have lowered the infant mortality rate in the country by decreasing the proportion of high risk pregnancies. Given that the fertility reduction coincided with a severe economic crisis and the fact that abortion is the main method of conscious fertility control, it is likely that it was those women whose newborns would have been at highest mortality risk who resorted to abortion. If this were the principal cause of the reduction in infant mortality, then this reduction, rather than indicating the improvement in the health of Bolivian children, is a reflection of the increasingly desperate circumstances of Bolivian families.

There has been a major expansion in the educational system since the 1950s as seen in Figure 3.13. Given the strong educational differentials in infant mortality, changes in the infant mortality rate associated with this compositional shift in the population would be expected.

The change in the aggregate infant mortality rate since the mid-1970s can be viewed as a combination of two factors: change in rates and change in composition. The aggregate infant mortality rate is a weighted average of the sub-group rates. The aggregate rate will change when the sub-group rates change, when the distribution of the overall population among the sub-groups changes, or some combination of the two. For example, the IMR in Bolivia can be considered to be the weighted average of the IMR in urban areas and the IMR in rural areas, when these rates are weighted by the proportion of children born in urban and rural areas respectively. If place of residence is a determinant of infant mortality, then migration from the high mortality rural area to the lower mortality urban areas would reduce the overall infant mortality rate in Bolivia. This is a compositional effect. Without any compositional change, the urban IMR, or the rural IMR, or both

could fall. The resulting reduction in the overall IMR is a rate effect. In reality, the aggregate rate falls due to both rate and compositional effects.

Much of the beneficial effect of modernization on infant mortality is presumably due to compositional effects. A modernizing country is characterized by increasing urbanization or expansion of educational systems. These result in compositional shifts in the population which lead to reductions in the aggregate infant mortality rate. A formal decomposition of the change in the aggregate infant mortality rate into compositional and rate effects cannot be used to establish causality. However, it does yield an estimate of the magnitude of possible composition effects which may accompany modernization.

Using data from the 1976 Census and the 1988 Survey, the univariate effects of the purely compositional shifts in childbearing by maternal education, place of residence, and region of residence on the aggregate infant mortality rate can be analyzed. The 1976 Census yielded an IMR estimate of 153. The 1988 Survey yields an estimate of 102, an overall decline of 51 deaths per thousand births. By applying the 1986 composition of births to the 1976 sub-group rates, we can estimate the change in the IMR that would have occurred purely due to compositional shifts in births which occurred over the 12 year period. The results of this formal decomposition are presented in Table 3.2.

Table 3.2

Aggregate Infant Mortality Rate, 1976 153

Aggregate Infant Mortality Rate, 1988 102

Net Change, 1976-1988 -51

MATERNAL EDUCATION Rates, 1976 Composition, 1988

None 185 17%

Basic (1-3 years) 155 10%

Basic (4-5 years) 138 33%

Intermediate 104 15%

Middle, University, Advanced 69 25%

IMR under pure composition effect 125

Percent of observed reduction

attributable to composition effect 55%

PLACE OF RESIDENCE Rates, 1976 Composition, 1988

Urban 125 51%

Rural 171 49%

IMR under pure composition effect 148

Percent of observed reduction

attributable to composition effect 10%

REGION OF RESIDENCE Rates, 1976 Composition, 1988

Altiplano 152 46%

Valleys 173 27%

Llanos 118 27%

IMR under pure composition effect 149

Percent of observed reduction

attributable to composition effect 8%

NOTE: The pure composition effect is calculated by assuming unchanging rates. If the data for 1976 were available, one could also calculate a pure rate effect by assuming unchanging composition. The sum of the pure composition and pure rate effect will not be 100%. In the case of Bolivia, this sum would exceed 100% -- indicating that both compositional and rate changes have acted in concert to reduce infant mortality.

A compositional shift which has received a lot of attention among Bolivian demographers is the urbanization of the country. This factor is minor, however, in the observed reduction in infant mortality. Less than 10% of the decline might be attributed to a pure compositional shift based on rural to urban migration. Similarly, there have been regional shifts in the population from the highland valleys to the Altiplano and especially to the topical lowlands. These too are minor effects. Less than 8% of the decline could be accounted for by a pure compositional shift.

Not all forces of modernization may act to reduce infant mortality through compositional shifts. Bongaarts (1988) surmised that expansion of family planning programs accompanying modernization could increase the aggregate mortality rate. As mothers from higher social strata adopt family planning, the social composition of births would shift toward lower educated women whose children are at higher risk. Due to this compositional shift, the aggregate mortality rate would rise.

When Bongaarts's hypothesis is viewed in the broader context of modernization, a different picture emerges. The social composition of births can be viewed as a weighted average of the fertility rates of the various social strata, when these rates are weighted by the proportion of women in each stratum. In the case of Bolivia, and perhaps typically in developing countries undergoing a fertility transition, there is an increase in the proportion of women who are highly educated while there is a simultaneous reduction in their fertility. The net outcome might be a shift in the social composition of births towards or away from social strata or classes which tend to have lower infant mortality.

In the case of Bolivia, the social composition of births seems to have shifted in the direction of more highly educated women. As seen in Table 3.2, the pure compositional shift in childbearing would account for fully half of the observed decline in infant mortality rate. Though these results are consistent with the view that expansion of educational systems result in large declines in infant mortality, the formal decomposition cannot prove this causal relationship.

It is possible that some more general phenomenon is at work and the expansion of the educational system and the reduction in infant mortality are both the result of this third factor. And too, if a causal relationship were to exist, we could not ascertain from this data set whether education provides a specific know-how or a certain status or a desire to use modern health technologies -- all of which result in improved child survival. However, it is known that for the majority of Bolivians going to school means developing a fluency in Spanish. In Bolivia, this fluency has traditionally been a prerequisite for access to modern ideas, attitudes, and technology. Perhaps, education is best thought of as an indicator of access to the modern world.

Often, authors attempting explanation of the historical or contemporary health transition have seemingly appealed to Arthur Conan Doyle to support their arguments. In the words of Sherlock Holmes: "Whenever you have eliminated the possible, whatever remains, however improbable, must be correct." (cf. McKeown, 1976). The trouble with this type of argument is that it means that explanation will tend to lie with those variables which are difficult to eliminate because they are difficult to measure. Therefore, there should be a reluctance to attribute Bolivia's mortality decline to poorly measured explanatory variables: increases in induced abortion in response to the economic crisis or diffusion of new ideas about health and the role of women. Similarly, there should be reluctance to dismiss evidence which has been poorly measured such as changes in the standard of living.

Expansion of the educational system and the accompanying shift in the composition of births toward higher educated women could have accounted for nearly half of the observed mortality decline. But, until a direct link between education and infant mortality has been established, this remains a highly speculative causal factor. Several commonly held explanations of mortality decline have, however, been eliminated: urbanization, increases in the GDP per capita, improvements in nutrition, and expansion of primary health care.

This brief overview provides the context for an analysis of the health of Bolivian children in the late 1980s. There is little information concerning the health of Bolivian children over the last thirty years. It appears that mortality rates began to fall quite rapidly during the last 15 years. While eliminating several of the leading possible explanations, this analysis found no clearly identifiable cause of the mortality decline in Bolivia.