Socioeconomic Inequalities in Childhood Mortality

The last three decades have witnessed substantial reductions in childhood mortality in most developing nations. Despite this encouraging picture, analysis of WFS and DHS survey data shows that socioeconomic disparities in survival chances have not narrowed between the 1970s and 1980s, and in some cases, have widened. Changes in mother’s education and father’s occupation contributed only modestly to secular declines in mortality. In most countries studied, no more than 20 per cent of the national trend could be accounted for by compositional improvements. The median contributions of improvements in mother’s education and father’s occupation were ten and eight per cent, respectively. The last three decades have witnessed substantial reductions in childhood mortality in most developing countries. The record of socioeconomic progress has been less even. Nevertheless today’s mothers are more likely than the mothers of the previous generation to be educated, to be living in an urban setting and to be married to a man with a non-manual occupation. We also know, particularly from extensive analysis of WFS data, that the survival chances of children vary widely between socioeconomic strata, with the educational attainment of the mother being a particularly strong predictor. From this perspective, two important questions arise. Are socioeconomic disparities in child survival widening or narrowing? And to what extent can mortality decline be attributed to changes in the socioeconomic composition of populations? Answers to these and other questions relating to changes in the age pattern of mortality are sought by a joint analysis of WFS and DHS survey data for 15 developing countries that have participated in both survey programs. We make no pretence that these countries are representative of the developing world or of particular regions, but their number and geographical spread are sufficient to permit tentative generalizations. Life-table measures of mortality for five-year calendar periods are produced at the national level and for socioeconomic subgroups. This approach permits an unusually long historical dimension to the study of trends. It also encounters severe problems of data consistency between the WFS and DHS which are assessed in the first substantive section of the paper. In a few countries, the two sets of estimates are incompatible. These cases are dropped from the analysis. In the majority of cases, however, they match well and may be regarded (with caution) as a single continuous historical record. * The authors wish to thank the two anonymous reviewers, whose comments on an earlier version of the paper were very helpful. Any remaining errors and oversights are entirely the responsibility of the authors. 2 SOCIOECONOMIC INEQUALITIES IN CHILDHOOD MORTALITY CLELAND, BICEGO AND FEGAN HEALTH TRANSITION REVIEW VOL. 2 NO 1 1992. Design of the analysis The data collection procedures of WFS and DHS are well known and require no elaboration here (Singh 1984, IRD 1990). With regard to mortality measurement, both survey programs used methods that are identical in their essential features: an initial count of live births by sex and survival, followed by a birth history in which date of birth, survival status and age at death are recorded for each child. From these data, period measures of mortality are obtained by standard life-table procedures. For each survey, three periods of interest were defined: the five years preceding the survey plus the year of fieldwork; and the two preceding quinquennia. Thus the entire analysis is restricted to events and exposure within the 15 to 16 years before each survey. In calendar terms the study period covers a little over two decades: from the mid-1960s (the most distant WFS period) to the mid-1980s (the most recent DHS period). To press the analysis beyond this span would have encountered increasing problems of data quality and truncation. Even within this shortened time span, mortality estimates suffer from the progressive loss of children born to older mothers, as attention shifts from most recent to less recent periods. However, exploratory analysis confirmed that the truncation bias was negligible within the 15year period under study and could be ignored. One of the potential advantages of birth histories over summary measures of mortality from censuses and household surveys is their potential for examination of the age-specific probabilities of death. While most of the results in this paper take the form of 5q0 estimates (termed ‘overall childhood mortality’), trends in infant mortality (1q0) and child mortality (4q1) are also presented. The placing of the main emphasis on values of 5q0 is justified in terms of both their relatively low sampling error and their robustness to errors in reporting age at death. As noted earlier, the substantive focus of the paper is on socioeconomic differentials in childhood mortality. The choice of characteristics for inclusion in the analysis was limited to those measured in a reasonably comparable manner in both survey programs, and by the additional consideration of sample distributions. There were only four serious contenders: maternal education, paternal education and occupation, and urban-rural residence. In view of its strong association with child survival, maternal education was an obvious choice. The existence of level of school reached and number of years completed on both WFS and DHS files enabled us to define three groups (no schooling, some primary schooling, some secondary or higher schooling) in a consistent manner for both WFS and DHS. The handling of paternal characteristics was less straightforward. Ideally we required a measure of economic status but neither WFS nor DHS has devised a satisfactory way of measuring this multifaceted concept. Paternal education is no doubt a predictor of income and standard of living but it is highly correlated with maternal education and, for this reason, was unlikely to yield results that differed appreciably from the maternal education results. Father’s occupation has the advantage of identifying position within the economic structure more adequately than education, but is typically not well measured. A recurrent problem with this variable is the existence of an often large but ambiguous sales and service category. On the reasonable assumption that a combination of paternal education and occupation might represent socioeconomic status better than either single characteristic, the following categories were defined: (a) white collar: professional, managerial or clerical workers with at least five years of schooling, plus sales or service workers with secondary or higher schooling. (b agrarian: self-employed farmers or agricultural employees. (c) blue collar: all skilled and unskilled manual workers, plus professional, managerial or clerical workers with less than five years schooling, plus sales and services workers with less than secondary schooling, plus missing values and never married mothers. SOCIOECONOMIC INEQUALITIES IN CHILDHOOD MORTALITY 3 CLELAND, BICEGO AND FEGAN HEALTH TRANSITION REVIEW VOL. 2 NO 1 1992. The sensitivity of estimates to the decision to place all missing values and never-married mothers in the blue collar category was assessed by running new tabulations with these two responses placed in a separate category. The only case where these adjustments made a noticeable difference to the blue collar estimate was Kenya, but, as will be shown later, this country was excluded from the main study on other grounds. The third and last characteristic selected for the analysis was urban-rural residence. Though it can be demonstrated that much of the urban-rural differential in childhood mortality merely reflects differences in educational composition, the inclusion of the variable is warranted because of its relevance to health policies. The demonstration that large differences in mortality exist between the two strata may lead, and occasionally have led, to shifts in health resources. All WFS and DHS files contain a coding of type of place of residence, often in the form of dichotomy but sometimes with a more detailed classification of urban areas. Large and smaller urban centres were always grouped together to form a single urban stratum. We were unable to verify that the distinction between urban and rural localities was always identical in both surveys for the same country. However, the policy of both survey programs is to follow the official or census definitions of each country. It is possible that official definitions were changed in the period between the two surveys, but it is most improbable that this occurred in sufficient instances to invalidate our analysis. To summarize, period measures of morality were calculated at the national level and for subgroups defined in terms of maternal education, paternal occupation/education, and rural-urban residence. Trends were calculated in a straightforward manner from estimates for different periods; and changes in differential mortality were assessed in form of absolute and relative differences for subgroups at different points in time. Consistency and reliability of WFS and DHS estimates The value of this analysis rests largely on the degree of consistency between WFS and DHS estimates. This issue is addressed in Figure 1, which presents an overview of the two sets of estimates. For several countries the conjuncture of the trend lines is perfect. Mexico, Egypt, Tunisia, and Dominican Republic fall into this category. In a further group (Sudan, Senegal, Morocco, and Ecuador), the match is very close, almost certainly within the bounds of sampling error. The remaining seven countries are more problematic. In six instances, the DHS estimate for the period approximately 10 to 14 years before the survey is lower than the WFS estimates for the most recent quinquennium before the survey. These disparities suggest – though do not prove – that deaths occurring in the more distant past suffer greater omission than recent deaths. In the remaining case, Ghana, the reverse occurs: DHS estimates are higher than the corresponding WFS estimates. The magnitudes of these discrepancies are shown in Table 1 in some form of estimates for the same five-year period, mostly taken from an earlier analysis of DHS data quality (Sullivan, Bicego and Rutstein 1990). Ghana provides by far the most serious instance of obvious error. For the period centred on 1977 the WFS estimate of childhood mortality is 38 per cent lower than the corresponding DHS figure. A WFS analysis by Adansi-Pipim (1985) showed that WFS also yielded lower mortality estimates than the 1971 Supplementary Enquiry; thus the probable cause of the WFS-DHS discrepancy is severe omission of deaths in the Ghana WFS, even for the most recent period. Table 1 Comparison of WFS and DHS estimate of overall childhood mortality for the same five-year reference period Midpoint of Level of 5q0 Percentage 4 SOCIOECONOMIC INEQUALITIES IN CHILDHOOD MORTALITY CLELAND, BICEGO AND FEGAN HEALTH TRANSITION REVIEW VOL. 2 NO 1 1992. reference period WFS DHS difference Senegal 1976.0 261 259 +1 Ghana 1977.2 120 166 -38 Sudan (North) 1976.5 149 141 +5 Peru 1975.5 144 132 +9 Egypt 1975.5 208 206 +1 Morocco 1977.7 144 155 -8 Indonesia (Java/Bali) 1973.8 158 140 +11 Kenya 1975.5 143 116 +19 Dominican Rep. 1972.9 127 117 +8 Ecuador 1977.3 116 110 +5 Tunisia 1976.1 103 101 +2 Mexico 1974.4 94 93 +1 Thailand 1972.9 88 74 +15 Colombia 1974.0 105 84 +19 Sri Lanka 1973.3 85 59 +31 Source: Sullivan et al. 1990 In Sri Lanka, Kenya and Colombia, DHS estimates are appreciably lower than WFS ones for the overlapping period. In Sri Lanka the root problem may be related to the conditions under which the DHS was conducted. Part of the country had to be excluded from the sample frame and no doubt it was more difficult to execute a high-quality national survey in the 1980s than in the 1970s, for obvious reasons. In Kenya, the defect also lies primarily with the DHS; comparison with other sources suggests omission of dead children particularly by older DHS respondents. Diagnosis of the problem in Colombia is more difficult. Indeed, the longer historical perspective provided in Figure 1 indicates that the discrepancy may be restricted to the overlapping period and is not as serious as the 19 per cent gap for this period might imply. On the basis of these comparisons we decided to exclude Ghana, Sri Lanka and Kenya from the main analysis of socioeconomic differentials. In all three cases, there were solid grounds for doubting the validity of national estimates for one or other of the two surveys, even for the most recent period. If national estimates are suspect, then even more uncertainty must surround subnational figures. For the remaining twelve countries, the two sets appeared sufficiently consistent to justify more detailed examination. Figure 1 Trends in overall childhood mortality, circa 1965 to circa 1985 SOCIOECONOMIC INEQUALITIES IN CHILDHOOD MORTALITY 5 CLELAND, BICEGO AND FEGAN HEALTH TRANSITION REVIEW VOL. 2 NO 1 1992. Consistency between WFS and DHS, of course, does not guarantee accuracy. Indeed a comparison with the series compiled by Hill and Pebley (1989), who drew upon a variety of data sources and report only for countries having information judged to be of reasonable quality and consistency, indicates major differences for several countries. The most startling example is Tunisia, where WFS and DHS trends are totally consistent but are much lower than the Hill-Pebley estimates. For the period centred on 1972, Hill and Pebley give a 5q0 value of 180, compared to values from WFS of 130 for 1970 and 106 and 104 for 1976 from WFS and DHS. It is inappropriate to attempt here a resolution of these and other differences in estimated childhood mortality levels and trends, because it would involve detailed country-specific evaluations. It must suffice to stress that mortality trends for many developing countries are not known with certainty and that the results presented below should be interpreted with caution. Interpretative emphasis should be given to general patterns rather than country-specific results. National levels and trends We start the presentation of results with an examination of national levels and trends in childhood mortality. By comparing the estimate obtained in the WFS for the period 10 to 14 years prior to the 6 SOCIOECONOMIC INEQUALITIES IN CHILDHOOD MORTALITY CLELAND, BICEGO AND FEGAN HEALTH TRANSITION REVIEW VOL. 2 NO 1 1992. survey (centred approximately on 1965) with the most recent DHS estimate (centred approximately on 1985), a 20-year perspective on change is obtained. In eleven out of twelve countries, substantial declines in childhood mortality are recorded (Table 2). The maverick in this group is North Sudan, but a special degree of caution is required for this country because of the possibility that both WFS and DHS suffer from omission. Farah and Preston (1982), for instance, obtained much higher estimates of childhood mortality from the 1973 Census than those derived from WFS. Table 2 National levels of overall childhood mortality: circa 1965 and 1985 Level c.1965 c.1985 Decline Abs Percentage Senegal 295 191 104 35 Egypt 248 107 140 57 Indonesia (Java/Bali) 206 95 111 54 Peru 195 111 84 43 Morocco 187 102 85 45 Dominican Republic 159 88 70 44 Ecuador 156 82 74 47 Tunisia 151 65 86 57 Sudan (North) 148 127 21 14 Thailand 131 45 86 66 Mexico 128 61 67 52 Colombia 123 43 80 65