African Population Studies Union for African Population Studies ISSN: 0850-5780 Vol. 12, Num. 1, 1997

African Population Studies/Etude de la Population Africaine, Vol. 12, No. 1, March/mars 1997

Gugulethu Gule, Ph.D.

Statistics and Demography Department, University of Swaziland

Code Number: ep97005

Abstract/Résumé

This paper identifies factors that account for areal differences in childhood mortality in Swaziland. An analytic model developed by Mosley and Chen (1984) is used to study the determinants and their mechanisms A multivariate analysis is used to investigate mortality variation. Data used in this paper are from the 1983 Swaziland National Nutrition Survey and the 1986 Swaziland Population Census. The results show that ecological zone, access to a major road and maternal education are the most important determinants of geographic variation in mortality in Swaziland. Together, these explain about 57 percent of the variation in mortality. Of particular interest is the finding that the effect of building a road is almost equivalent to that of increasing the number of women with secondary education and above by 45 percent, both leading to some 2 percent reduction in childhood mortality.

Depuis 1960 le volume des recherches menées sur la prise de décision en ce qui concerne la reproduction a considérablement augmenté. La communauté des chercheurs est actuellement essentiellement préoccupé par l'amélioration des connaissances dans les domaines spécialisés. D'autre part, l'insuffisance des connaissances et des capacités demeure un problème majeur. On sait très peu des déterminants de la population au Nord du Nigéria, où les informations sur les indications démographiques de base ont montré des différentiels croissants selon les régions à la fois au niveau de la fécondité et des systèmes de santé. On sait très peu des déterminants ou conséquences des différences régionales au niveau de la fertilité pour la vie des femmes et des transitions qui s'opèrent au niveau de la fécondité. Il est de plus en plus clair que la position qu'occupent les femmes dans la famille et dans la communauté a un impact majeur sur la fécondité au Nigéria mais les implications au niveau des politiques et des programmes ne sont toujours pas claires. La tâche de la communauté des chercheurs dans les années à venir consistera à aborder la question de savoir comment la position des femmes peut être améliorée dans les sociétés qui refusent d'éduquer leurs filles et de permettre à leurs femmes de travailler hors du domicile conjugal.

Introduction

Childhood mortality is one of the important indicators of a country’s general medical and public health conditions, and consequently, the country’s level of socio-economic development. Its decline is therefore not only desirable but also indicative of an improvement in general living standards. Historically, rapidly declining mortality and sustained high levels of fertility have been the principal driving forces behind the high and accelerating population growth rates in sub-saharan Africa. Rapid population growth is undesirable because it hampers socio-economic development, which in turn may undo the progress in mortality reduction. It follows then, that efforts to reduce mortality should be accompanied by effective fertility reduction strategies.

Swaziland has experienced a very sharp decline in childhood mortality. The childhood mortality rate (i.e. number of under-five deaths per thousand live births) fell from 214 deaths per thousand in 1960 to 120 in 1988 (Gule, 1990). Preliminary results of the 1991 Swaziland Demographic and Housing Survey indicate that childhood mortality continues to decline at a rapid pace, the estimate being 89 per thousand in 1991 (Gule, forthcoming).

The improvement in childhood mortality is the result of a combination of factors. Amongst these are the expansion of health care facilities and services, disease-specific health interventions, changes in diets and health behaviour, and the success of developmental programmes such as the Water and Sanitation Programme. The significant improvement in Swazi women's educational attainment must also have played an important role in reducing mortality.

Despite the rapid decline, Swaziland's childhood mortality levels are still very high, compared to those of other countries in the Southern African region. For example, Swaziland's infant mortality rate is 72 per thousand, while those of Botswana and South Africa are 39 and 46 respectively (Gule, forthcoming; Population Reference Bureau, 1995). There is, therefore, a need for the Ministry of Health and Social Welfare to intensify its mortality reduction campaign.

Though childhood mortality has declined sharply at the national level, the pace of decline varies substantially from one area of the country to another. Consequently, vast geographic differentials in childhood mortality exist. For example, the highest childhood mortality rate (203 per thousand) at Ezidwedweni, in the south-eastern part of Swaziland, is more than double that of the lowest rate (90 per thousand) in Manzini, the commercial city (Gule, 1990).

To facilitate a guided and effective reduction in childhood mortality, sub-groups of the population that experience very high mortality need to be identified and targeted for mortality reduction. Narrowing mortality differences among population sub-groups will result in the reduction of the overall level of childhood mortality.

The objective of this paper is to identify factors that account for the large differences in areal mortality in Swaziland, using multivariate analysis. The knowledge of these factors is important for policy formulation and implementation because it will enable policy makers to formulate appropriate policies for the reduction of childhood mortality by paying special attention to the disadvantaged sub-groups.

Analytic framework and hypotheses

An analytic framework developed by Mosley and Chen (1984) is used in explaining the mechanisms through which various determinants operate to affect mortality. The framework categorizes the determinants into two groups: indirect determinants (e.g., education and availability of sanitation facilities); and direct or proximate determinants (e.g., immunization and child-feeding practices). Proximate determinants have the closest or most direct effect on mortality. Indirect factors, on the other hand, are the most "distant" from mortality, and they operate through one or more proximate factors to affect mortality.

The principal hypotheses tested in this paper are that childhood mortality is lowest in areas:

1. that have a high proportion of educated women;
2. that have access to a major road;
3. that are located in the Highveld and Middleveld; and
4. that have a high level of utilization of modern delivery health facilities.

Amongst the numerous factors that have been found to be associated with childhood mortality, maternal education has been shown to have the greatest impact on children's survival chances. Education is highly and negatively correlated with childhood mortality even when other factors correlated with education have been controlled (Farah and Preston, 1982). Although there is general consensus about the importance of maternal education, there is little agreement on the mechanisms through which education operates to affect mortality. Nayar (1985) contends that educated people are more aware of the location of health care facilities and are more likely to utilize them. At home, they may take better care of their children by providing more nutritious food and practising hygienic habits (e.g., washing hands before handling food). Jain (1988), on the other hand, contends that increased levels of education result in better utilization of available health facilities. A more economically oriented argument by Schultz (1979) states that better educated mothers earn more in the labour market and consequently their household incomes are elevated, thus enabling them to purchase goods and services to improve child health.

Women's educational attainment has improved significantly in Swaziland. The proportion of women aged 15-59 with secondary education and above (i.e. 10 or more years of schooling) increased from 16 percent in 1976 to 31 and 40 percent in 1986 and 1991 respectively. Though education levels have improved vastly at the national level, some areas still lag behind. Consequently, we expect areas with relatively high educational attainment to have lower levels of childhood mortality.

Studies of the effect of access to roads on mortality are lacking in demographic research. Though there is a general understanding that access to roads may have an effect on mortality, there is no study that has empirically investigated this relationship. The only study that alludes to the importance of roads does not present empirical evidence. Nayar (1985) speculates that access to good roads in the Kerala State of India may particularly have contributed to lower mortality in this State compared to the others.

Swaziland has a relatively good network of motorable roads, even in very remote areas. However, not all the roads have a good public transportation system. In some areas there is virtually no public transportation. In this study, a road is considered to be major, and inherently good if, in addition to being motorable, it has a frequent and reliable public transportation system.

In Swaziland, access to a major road almost always implies better access to social amenities like schools and modern health facilities. In addition, an area that has access to a road is open to interactions with people from all walks of life, and this interaction may have an impact on people’s attitudes and knowledge. For instance, women may obtain knowledge of the location of modern health facilities. Furthermore, areas with good access to roads are more likely to benefit from visits by government extension officers from the ministries such as those of Education, Health and Agriculture, as well as from non-governmental organizations. For these reasons, we expect areas with access to major roads to have lower childhood mortality.

Distinct childhood mortality differentials by ecological region of residence have also been observed. The mortality differences are a result of regional climatic and disease prevalence conditions. In their study in Kenya, Anker and Knowles (1977) found that malarial endemicity in different regions had a strong effect on childhood mortality levels.

Swaziland has a variety of topographic and climatic conditions. The country is divided longitudinally into four ecological zones: the Highveld, Middleveld, Lowveld and Lubombo Plateau. The mountainous Highveld has a temperate climate, with occasional frosts in winter. The Middleveld, with a gently undulating topography, has a favourable warm climate and is therefore the most agriculturally developed. The Lowveld’s very hot climate and flat terrain are favourable to disease organisms and carriers. Malaria and other diseases like diarrhea are therefore prevalent in this zone (Swaziland Government, 1986). The sparsely populated Lubombo Plateau has climatic conditions similar to those of the Middleveld. Given the significant differences in climatic conditions and disease prevalence by ecological zone, we expect childhood mortality to also vary substantially by ecological zone.

Place of delivery is also an important determinant of mortality, particularly neonatal mortality. Children delivered in modern health facilities usually exhibit lower rates of mortality (Jain, 1988). However, in some cases, mortality among children delivered in modern facilities is observed to be higher because mothers use these facilities mostly when they have pregnancy complications.

The level of utilization of modern health delivery services is not very high in Swaziland. For example, only about 57 percent of the women delivered their children in modern health facilities (i.e. hospitals or health centers) in 1983. The remainder delivered in their homes or in the homes of traditional birth attendants. Because of differential levels of utilization of delivery health facilities per area, we expect areas with higher levels of utilization to have lower levels of childhood mortality.

Data and methods

The analysis in this paper is based on data from the 1986 Swaziland Population Census and the 1983 Swaziland National Nutrition Survey. The census provides information on childhood mortality and on indirect determinants (i.e. ecological zone of residence, land tenure system, maternal education, type of toilet facility and fuel used, source of water supply and employment status). Due to the fact that the census does not provide information on all the variables this paper intends to analyze, supplementary information (i.e. information on proximate determinants) is obtained from the survey. Mapping procedures were used to link the two data sets.

The primary objective of the survey was to assess the health and nutritional status of Swazi children. Its sample, consisting of 3,337 women aged 12-49 years and 5,370 children age 0-59 months, was drawn from 125 enumeration areas, using a two-stage random sampling technique (Swaziland Government, 1986). The three questionnaires administered provide information on length of breast-feeding, diarrhea prevalence, travel time to the nearest health facility, place of delivery, measles immunization, clinic attendance and nutritional status of children.

A multivariate regression technique is used to identify factors that account for geographic variation in childhood mortality in Swaziland. The childhood mortality rate, q(5), in each commune (defined later) is the dependent variable in the estimated regression equation. It is computed by using a computer programme developed by Ewbank (1986) based on reports of women aged 15-29 years. This program is one of several variants of Brass' original method of estimating childhood mortality (Brass, 1964). In estimating q(5), the Coale and Demeny East model life table is used because it best represents Swaziland's age pattern of mortality in childhood (Gule, 1990).

In the 1986 Census 1,077 enumeration areas were identified. The major difficulty in using these areas as units of analysis lay in the fact that it was impossible to estimate mortality rates for some areas because they had very few inhabitants. In order to overcome this problem, larger areas, termed "Communes" for the purposes of this research, were created. A Commune is defined as a group of adjacent enumeration areas , preferably sharing a common areal name. The grouping yielded 115 Communes, the average population size of each Commune being 5,922, with an areal size of 160 square kilometres (58.3 square miles).

Because data on proximate determinants were not available at the commune level from census data, each Commune was assigned a proximate variable value for the Nutrition Survey enumeration area or areas located in it. We assumed that the commune variable was adequately represented by the enumeration area variable. For instance, we assumed that the prevalence of diarrhea in an enumeration area was similar to prevalence in the commune in which the enumeration area was located. This could result in coefficients biased toward zeros, since prevalence may vary significantly from one enumeration area to another within the same commune.

Though 115 Communes were created, not all the communes were used in the regression analysis. The 88 communes selected for the present analysis were those in which at least one Nutrition Survey enumeration area was located. In cases where there were more than one enumeration area, the average value for the enumeration areas was assigned to the commune. Mapping procedures were used to match communes and enumeration areas.

In order to determine the mechanisms through which variables of interest affect mortality, variables were introduced into the above equation at different stages. Variables perceived to have an indirect effect on childhood mortality (e.g., maternal education and access to roads) were introduced first, while proximate variables (e.g., place of delivery and nutritional status of a child), considered to have a more direct effect on mortality, were introduced last.

A brief description and measurement of each predictor variable used in this study is shown in Table 1. Education, employment and availability of a toilet are highly correlated with one another. In order to avoid problems of multicollinearity, the latter two variables were excluded from the above regression equation. Education was selected from among the collinear variables because it reflected the omitted variables.

Results

The greatest variation of areal mortality is explained by indirect variables, which account for 59 percent of the variation in areal childhood mortality (Table 2, Model 1). Proximate variables, on the other hand, account for only 26 percent of the variation (Model 2).

Indirect variables

The most interesting finding from Table 2 is the highly significant negative relationship between access to major roads and childhood mortality. The effect of access to roads on childhood mortality persists even when all variables are controlled. This variable alone accounts for about 38 percent of the geographic variation in mortality. Even the control for education, which tends to reduce the independent effect of many variables, does not affect the persistent independent effect of this variable.

An examination of correlation coefficients (not shown) reveals that residents of areas with access to a major road tend to be more educated and have better access to and are more likely to utilize modern health facilities. Consequently, the inclusion of these variables in the regression equation significantly reduces the ROAD coefficient.

In addition to improved access to and utilization of health facilities, we speculate that access to a major road alters social interaction patterns. Through interaction individuals may adopt lifestyles, attitudes and practices that may be beneficial for the survival of their children. For instance, women may obtain knowledge about the location of modern health facilities. Furthermore, areas with access to roads are more likely to benefit from visits by government extension officers from such ministries as Education, Health and Social Welfare and Agriculture, as well as from non-governmental organizations.

At the univariate level, childhood mortality is significantly lower in government and company towns than on Swazi nation land. However, the coefficient of the variable GOVT becomes insignificant once other variables, particularly education and access to roads, are controlled. This implies that lower mortality observed in government towns can be attributed to the fact that residents of these towns are relatively more educated and have better access to roads than their counterparts residing on Swazi nation land. These results are similar to those found in other studies which attributed the effect of urban residence to differences in education levels among rural and urban residents (United Nations, 1985). Though the coefficient of COMPANY declines when many other variables are controlled (Model 3), the variable still remains significant.

Ecological zone of residence, which is considered as a proxy for disease prevalence, has a strong and statistically significant effect on childhood mortality, even when many other variables are held constant. Communes located in the Lowveld, where malaria and other diseases like diarrhea are more prevalent, tend to have relatively higher childhood mortality than communes located in the Highveld and Middleveld. If data were available, one would examine if ecological zone maintains its significant effect on mortality when malaria endemicity is controlled.

Education plays a major role in accounting for mortality differentials in Swaziland. It alone accounts for 40 percent of the variation in areal mortality. This strong negative effect on mortality persists even after many other factors are controlled. Since education is so highly correlated with the employment and availability of toilets indexes, it is difficult to determine how much of this persistent effect is attributable to education alone, and how much to the other variables.

Proximate variables

Malnutrition, measured by the variable STUNT, is related to childhood mortality at the univariate level. However, the nutrition-mortality link disappears once other proximate variables are controlled (Model 2). This may be due to the fact that malnutrition and mortality are affected by similar factors. The surprising aspect is the apparent reversal of the sign of this variable's coefficient when certain variables, like education and access to roads, are held constant.

Univariate coefficients indicate that communes with longer periods of breast-feeding have higher mortality, a result contrary to our expectation. Results of the Nutrition Survey show that, in Swaziland, shorter periods of breast-feeding are associated with modernity (i.e. the length of breast-feeding is shorter in urban areas and among educated women). If indeed breast-feeding is associated with modernity, then controlling for other modernization variables should eliminate the significance of this variable in accounting for mortality variation.

Breast-feeding remains significant when other proximate variables are controlled, and it maintains its positive relationship with mortality. However, the expected negative relationship between breast-feeding and mortality emerges when other modernization variables like education are integrated into the regression equation, even though the coefficient becomes insignificant.

When only proximate variables are controlled, diarrhea prevalence is highly significant. When indirect variables, particularly the ecological zone dummy variables, are also included in the equation, the positive effect of diarrhea prevalence on mortality is eliminated. This may indicate that the ecological dummy variables measure not only the prevalence of malaria but also the prevalence of diarrhea.

 Both measles immunization coverage and clinic attendance have no significant relationship with childhood mortality. The lack of significance for the immunization variable may have resulted from the failure to locate vaccination cards in low mortality areas at the time the survey was being conducted.

At the univariate level, travel time to the nearest health care facility (TIME) is very important in accounting for areal mortality variation. However, when other proximate variables are controlled (place of delivery in particular), the explanatory power of this variable is eliminated, thus indicating that most of the effect of this variable on mortality is transmitted through other variables.

Place of delivery is the only proximate variable that maintains its independent effect on mortality when all other variables are controlled (Model 3). Much of its effect on mortality appears to be transmitted through the access to roads and education variables, which also measure access and utility of modern health facilities.

Conclusion

Standardized coefficients were used in gauging the relative importance of various independent variables in accounting for childhood mortality variation. The coefficients indicate that ecological zone, access to roads, and education (listed in the order of importance) are the most important determinants of areal mortality variation in Swaziland (see Table 3). As expected, communes located in the Highveld or Middleveld, those that have access to a major road and others with a high proportion of women with secondary education and above, have the lowest levels of mortality. These factors have an independent effect on mortality, thereby implying that their effect on mortality is not solely transmitted through other variables. They explain 57 percent of the variation in childhood mortality (not shown in Table 3). Among the proximate determinants, place of delivery is the most important. As expected, communes that have a high proportion of females delivering in modern health facilities exhibit lower mortality.

The most important and interesting finding of this study is the importance of access to roads in accounting for mortality variation. This variable maintains a highly significant effect on mortality even when many other variables are controlled. Communes that have access to a road have childhood mortality that is about 18 per thousand (2%) lower than that of communes with no access to roads when all other variables are held constant.

In examining the ordinary beta coefficients presented in Table 2 we observe that the effect of building a road in a commune is almost equivalent to increasing the number of women with secondary education and above by approximately 45 percent, both leading to some 18 per thousand (2%) reduction in childhood mortality. It is interesting to note that, using the infant mortality rate (infant deaths per thousand live births) in each commune as the dependent variable instead of childhood mortality in the regression equation yields relatively similar results. Building a road in a commune and increasing the number of women with secondary education and above by 47 percent have a similar effect, both leading to a 13 per thousand reduction in infant mortality. These results are important for policy formulation since they present different strategies with a similar impact on infant and childhood mortality. The choice of one strategy at the other's expense is dependent on many factors, including cost-effectiveness, cultural barriers that may be encountered and the time

frame under consideration. For example, building a road in a commune may have a faster impact than educating young girls.

A surprising and strange finding of this study is that the effect of indirect determinants is stronger than that of proximate determinants, a result contrary to our expectation, based on the analytical framework presented earlier. We expect the inclusion of proximate determinants in the regression equation to significantly reduce the size of the coefficients of indirect variables and their importance in accounting for mortality variations so as to indicate that the latter operate through the former to affect mortality. However, the results seem to indicate that proximate determinants operate through indirect determinants because the inclusion of indirect determinants reduces the size proximate determinants' coefficients and their significance (Table 2, Models 2 and 3).

This phenomenon may be an artifact of the assumption made in the measurement of the proximate variables (i.e. that a commune proximate variable is adequately represented by the enumeration area variable). The violation of the assumption could bias the coefficients toward zero. But if that is so, why are these variables significant when they alone are included in the regression equation? Another possible explanation for the phenomenon is that, by the nature of their measurement, the proximate variables are really indirect variables and therefore measure the same thing as indirect variables. Perhaps a different picture would emerge if individuals, rather than areas, were used as units of analysis in performing regression analysis and all variables were obtained from the same data set.

TABLE 1: Description and Measurement of Predictor Variables

Notes:

^a The variables are measured for each commune.
^b Omitted category in statistical analysis (pertains to categorical variables represented by dummy variables).
^c Mapping procedures were used to determine if a major road passed though a commune.
^d Stunting (chronic undernutrition) is defined as height-for-age that is more than 2 standard deviations below the National Center for Health/Centers for Disease Control mean value.
^e Computed by utilizing the prevalence/incidence method and a 24-month period (Eelens and Donne, 1985).

Source: 1983 Swaziland National Nutrition Survey and 1986 Swaziland Population Census

Table 2: Multiple Regression Coefficients and Univariate Coefficients Measuring the Effects of Indirect and Proximate Variables on Childhood Mortality, Measured by q(5)⁺

NI Not Included
⁺ q(5) per thousand
* Significant at the .10 level
** Significant at the .05 level
*** Significant at the .01 level

Sources: 1983 Swaziland National Nutrition Survey and 1986 Swaziland Population Census

Table 3: Ordinary and Standardized Beta Coefficients Measuring the Effect of Predictor Variables on Childhood Mortality, Measured by q(5)^a

Notes:

^a q(5) per thousand
* Significant at the .10 level
** Significant at the .05 level
*** Significant at the .01 level

Source: Table 2

References

• Anker, R. and J. Knowles. 1977. "An Empirical Analysis of Mortality Differentials in Kenya at the Macro and Micro Levels." Population and Employment Working Paper, No. 60. Geneva: International Labour Organization.
• Brass, W. 1964. Uses of Census or Survey Data for Estimating Vital Rates. A Paper Presented to the Economic Commission for Africa, African Seminar on Vital Statistics, E/CN.14/AS4/S.
• Eelens,F., and L. Donne. 1985. The Proximate Determinants of Fertility in Sub-Saharan Africa: A Factbook Based on the Results of the WFS. Inter-university Programme in Demography, Working Paper 1985-3. Brussels: Vrije Universiteit.
• Ewbank, D. 1986. Evaluation of Model Life Tables for East Africa. Seminar on Comparative Studies of Mortality and Morbidity: Old and New Approaches to Measurement and Analysis. Italy: Siena, 7-12 July.
• Farah, A.M. and S. Preston. 1982. "Child Mortality Differential in Sudan." Population and Development Review, Vol. 8, No. 2, June, 1982, pp. 365-383.
• Gule, G. 1990. Childhood Mortality in Swaziland: Levels, Trends and Determinants. Ph.d Dissertation In Demography. Philadelphia, PA: University of Pennsylvania.
• -----. forthcoming. "Mortality in Swaziland." In S. Dlamini (ed.), Report of the 1991 Swaziland Demographic and Housing Survey, Central Statistics Office, Mbabane, Swaziland.
• Hobcraft et al. 1982. Socioeconomic Factors in Infant and Child Mortality: A Cross National Comparison. A Paper Presented at the Annual Meeting of the Population Association of America, San Diego, May.
• Jain, A. 1988. "Determinants of Regional Variations in Infant Mortality in Rural India." In A. Jain and L. Visaria (eds.), Infant Mortality in India: Differentials and Determinants. Sage Publications, pp. 127-167.
• Mosley, W., and L. Chen. 1984. "An Analytical Framework for the Study of Child Survival in Developing Countries." In Mosley and Chen (eds.), op. cit., pp. 25-45.
• Nayar, P. 1985. "The Case of Kerala, India." In Vallin and Lopez (eds.), Health Policy, Social Policy and Mortality Prospects. International Union for the Scientific Study of Population: Ordina Editions, pp. 371-381.
• Population Reference Bureau. 1995. 1995 World Population Data Sheet, Population Reference Bureau, Washington, D.C.
• Schultz, T. P. 1979. Interpretation of Relations among Mortality, Economics of the Household and the Health Environment. In Proceedings of the Meeting on Socio-economic Determinants and Consequences of Mortality, Mexico City, June 19-25. Geneva: World Health Organization.
• Swaziland Government. 1986. Swaziland National Nutrition Survey: 1983. Atlanta, GA: Centers for Disease Control.
• United Nations, 1985. Socio-economic Differentials in Child Mortality in Developing Countries, ST/ESA/SER.A/97. New York, NY: United Nations, Department of International Economic and Social Affairs.
• Visaria, L. 1988. "Levels, Trends and Determinants of Infant Mortality in India." In A. Jain and L. Visaria (eds.), Infant Mortality in India: Differentials and Determinants Sage Publications, pp. 67-126.

Copyright 1997 - Union for African Population Studies.