The overall impact of conflict on the economy has been well documented in the literature and confirms the aforementioned findings of depressed SES in conflict settings as determined by the DHS. Recent estimates of annual income per capita for Eritrea and Mozambique are 150 and 100 USD, respectively, both well below the SSA average of 350 USD (Cousin 1997). In fact, levels of absolute poverty in Mozambique rose from 15% in 1980 to over 60% in 1995, making it the poorest country in the world by the time of its peace agreement in 1992. It is estimated that 33-57% of the urban population and nearly 70% of the rural population currently live in absolute poverty in Mozambique (WFP 1997).

According to Machel (1996), great risks to education are typical during conflict. Schools are often targeted during war, especially in rural areas, in part, due to their high profiles. For example, it was estimated that 45% of the primary school networks in Mozambique were destroyed over the course of the war. Teachers may also be targeted, as they are viewed as important community members and are often politicized. During the Rwanda crisis, it was estimated that more than two-thirds of teachers either fled or were killed. Additionally, formal education relies on consistent funding, which is difficult to sustain during periods of conflict. The impact of the destruction of the educational infrastructure and resources represents one of the greatest developmental setbacks for countries affected by conflict. The years of education and training lost will take equivalent years to replace, and, in the meantime, their absence imposes greater vulnerability upon the affected population.

Female literacy differentials are apparent between conflict and non-conflict countries, as presented in Table 3. Among countries affected by conflict, female literacy levels are lower than in neighboring countries. This is particularly true among the countries that have experienced high intensity conflict, such as Eritrea and Mozambique, where literacy rates fell below 35%. This level is nearly half that in neighboring countries such as Kenya and Zambia. The cases of Uganda and Sudan show that the maintenance of educational services during low level conflict, or in areas with reasonable security, may be less difficult.In the Sahel, literacy rates are the lowest in SSA, and may in fact contribute to instability.

It is generally believed by the international community that CEs and conflict give rise to an increase in female-headed households as men are often involved in war, killed or separated from their families, or migrate in search of employment in the aftermath of war. However, the DHS data suggest that female-headed households may not be more prevalent in conflict areas, on average, than regional averages. Great variability does exist in the distribution of female-headed households among and between conflict-affected countries. A unique pattern is seen in Eritrea, where 45% of the urban sample is composed of female-headed households, while only a quarter of the rural households are headed by women. On the other hand, in Mozambique, female-headed households are somewhat more prevalent in rural than in urban areas. Among the Sahelian countries, Chad has particularly high levels of female-headed households, but other Sahelian countries have low percentages of households headed by women (Figure 6). This is consistent with the fact that Chad has been the most significantly affected by war in the past few decades. Thus, the effect of conflict on household structure appears to vary among conflict affected countries. Although not explore in this analysis, it has been suggested that the lack of young men has prompted a major redistribution of the gender division of labor, placing a much greater burden on women, and creating a labor shortage that remains a major constraint facing vulnerable households, especially those that are female-headed (Cliffe 1994).

Table 3 reflects a large degree of population movement within Africa as a whole. Regardless of conflict status, respondents indicated a high degree of transience within a five-year period prior to the survey. Although the motivation for this migration is not available from DHS data, migrants are likely to be a heterogeneous group. Possible factors in the decision to move may include marriage and pursuit of economic opportunity in addition to flight from insecurity and post-conflict resettlement. Given the levels of poverty, it is not surprising to see a high degree of in-migration from the rural areas to urban centers, whether in search of food, health services, or employment. According to WFP, the urban population of Mozambique was estimated at 13% in 1980, but increased to over 30% by 1997. When migration is due to insecurity or abject poverty, households become extremely vulnerable, often losing all possessions relevant to their livelihoods in the process (Cliffe 1994). Unfortunately, DHS instruments do not solicit reasons for migration, which is particularly important in the case of conflict-affected countries.

Figure 7 suggests the fluidity of migratory patterns in SSA during the 1990s. Mozambique, Zambia, and Cameroon reflect substantial inter-regional variability in recent migration. In contrast, Uganda’s high levels of recent migration appeared throughout the country. The display also implies cross border population flows. For example, in border regions of Tanzania and Zambia, migration levels are elevated and similar to those in adjacent regions in Mozambique.

Two of the more unique cases included in the study are Sudan and Uganda. The Sudanese profile indicates somewhat better socioeconomic conditions for this population than for others among the conflict-affected group, although female literacy rates remain lower than those of countries not affected by conflict. In part, this finding might be explained by the fact that northern Sudan is transitional between African and Near Eastern/North African cultural and ethnic characteristics.

Despite its difficult political history, Uganda has relatively high levels of female literacy in relation to the other countries affected by conflict. It is characterized by the highest level of recent migration of all samples included in this study, and its dependency ratios also are highest. The moderate level of literacy (compared to other conflict-affected countries), among other measures of improved socioeconomic status, may be explained by the relatively quick and dramatic improvements brought about by the elected government of President Museveni, highlighting that even the perception of impending stability can improve sociodemographic standing.

Maternal and child nutrition status are also indicators of health and development. The nutritional status of women and children are a function of morbidity, behavioral factors and household food access. Again, armed conflict tends to undermine service availability and utilization, and through the loss of land and increased destitution, household food access is generally diminished, and would be expected to persist in post-conflict settings.

According to Green (1994), the loss of life in Tanzania has resulted primarily from the economic and fiscal burden of its war with Uganda setting back food security, and, more especially, hampering the delivery and expansion of effective primary health services (Green 1994).

Thus, the retrospective pregnancy history data appear to capture patterns of change in IMR over periods of conflict. Differential infant mortality rates over time are consistent with estimates from other sources, e.g. census data, and in general suggest that IMRs were elevated during periods of active conflict, and have improved during the post-conflict period. However, in Mozambique, infant mortality rates remain extremely elevated.

Table 4 also demonstrates the nutritional profiles of countries by conflict status. Surprisingly, the levels of stunting (short stature) among children are similar in both country sets, while the prevalence of wasting (thinness) among children and chronic energy deficiency (CED) among non-pregnant mothers, defined by a BMI < 18.5, is elevated in countries affected by conflict. Eritrea, for example, has levels of wasting nearly double those of neighboring countries. Nearly 40% of mothers are classified with CED, perhaps indicating a severe depletion of assets and low levels of food security. Among the Sahelian countries, maternal nutritional status in Chad is somewhat worse, possibly reflecting its more war-torn past. Levels of wasting and CED are also high in Mali, Niger, and Senegal, although these may merely typify the harsh ecology of the Sahelian environment.

High levels of malnutrition have been well documented during conflict situations and with the assistance of food aid and development programs tend to be corrected relatively quickly. In Mozambique, ACC/SCN (1994) reported wasting levels for IDPs and returning refugees in excess of 20% at war’s end in 1992-1993. With large-scale feeding, health, water, and sanitation programs established by 1993, the nutrition and health situation in Mozambique was brought quickly under control. Subsequent surveys by NGOs in 1993 showed a much-improved situation, with wasting levels generally below 10%. In general, the highest levels of malnutrition occurred in areas with the largest numbers of returnees, where poor water and sanitation had led to periodic outbreaks of dysentery and cholera. In the next section of this paper we demonstrate that indeed intrastate variation is quite substantial among conflict-affected countries.

Eritrea, on the other hand, sustains high levels of malnutrition. This may be due to the fact that efforts by Ethiopia to disrupt agricultural production in Eritrea seem to have succeeded. In 1997, only 20% of the national farms remained irrigated. The government estimated that at best, 60% of the nation’s food needs could be met, and nearly two-thirds of the population relied on some form of food aid (Cousin 1997). It has been suggested that few died from starvation though, because of food aid and an efficient distribution system developed by the ELPF during the conflict, and continued into the postwar era. However, some evidence suggests that available rations are insufficient to meet minimum requirements, resulting in the sale of available assets, thereby limiting future coping ability (Cliffe 1994). It is unfortunate that time series data are not available regarding nutritional status, so that trends could be assessed.

In addition to increased mortality, the factors that cause malnutrition among children during peacetime (e.g., household food security, poor diet, insufficient health services, unsanitary environment, inadequate maternal/child care practices) are often accentuated and exacerbated during armed conflict. Due to varying levels of exposure to these factors amongst the regions, differentials in nutritional status also appear in post-conflict settings. As stated earlier, the prevalence of stunting (chronic malnutrition) was similar among both country sets; yet, intrastate regional comparisons showed stunting level svarying by up to 350 percent in Mozambique and Eritrea, in comparison to 180 percent among regions of non-conflict countries (Figure 14). Larger differentials were found for wasting (acute malnutrition), as well as underweight status (Figure 15, below), where differences of up to 500 percent were found between regions in conflict-affected countries (also see Figures 16 & 17).

From cross-sectional data it is not possible to correlate nutrition and mortality at the individual level. Rather, analysis was performed at the cluster level, examining the relationship between the prevalence of malnutrition and the proportion of children having died among births during the five years preceding the survey. Rank correlation was used to determine associations between the three forms of malnutrition (stunting, wasting, underweight) and the proportions of children dying within clusters in the five years prior to the survey (see Table 5).

As indicated above, strong associations are present between malnutrition and mortality in relatively stable countries, but not among conflict-affected countries. Stunting (and underweight) shows the expected positive correlation with mortality in non-conflict areas, as it is often indicative of repeated bouts of infectious disease, as well as chronic deficiencies of food intake. The lack of correlation between mortality and malnutrition among the conflict-affected countries is puzzling, and yet not easily explainable with the data at hand. In certain cases, like Eritrea, where stunting is high but childhood mortality is low, the provision of food aid and basic health services through relief efforts may be serving to protect child survival. In other instances, such as in Mozambique, children may not be living long enough to exhibit the signs of chronic malnutrition. Regardless, this suggests some uniqueness to conflict-affected settings, and implies that mechanisms of malnutrition and mortality must be better understood before appropriate policies and programs can be developed.

F.3 Health care utilization

It is well documented that large scale violence has multiple direct and indirect effects on health and disease, often through an overall disruption of health service delivery, at times accompanied by reductions in health care expenditures (Kalipeni 1998; Cliff 1993; Sabo 1989). Health personnel, supplies, and facilities are often targeted by combating forces, which when combined with security threats, leave routine under-five clinics, immunization programs, and prenatal care virtually unavailable. Although limited information is available regarding the impact of conflict upon the Eritrean health care system, it has been estimated that more than 40% of Mozambique’s health centers were destroyed during the war (Machel 1996) and immunization programs stopped in many areas (Garenne et al. 1997).

Table 8 illustrates the general level of health care utilization among the comparison groups by conflict status. It details coverage indicators, such as immunization, antenatal care (ANC), and the utilization of health care professionals during the birthing process. Immunization of children and pregnant women is promoted as a particularly effective intervention for improving child survival and reducing childhood morbidity. However, the percentage of children 12-23 months of age who were fully vaccinated in conflict areas was very low, relative to non-conflict countries (Figure 18). In the Sahelian region, full vaccination coverage is in the teens for Chad and Niger, versus nearly 40% in the Central African Republic and Cameroon. In the Eastern and Southern regions, the proportion of children fully vaccinated improved among conflict-affected countries, ranging from 41-51%. Yet by comparison, non-conflict countries in the same geographic areas demonstrated much better coverage rates, ranging from 71-80%.

Table 8 also reflects substantial differences in tetanus coverage according to conflict status. Among non-conflict countries, coverage of pregnant women ranges from 70-90%, while in most conflict-affected countries (except for Senegal and Uganda) coverage is at or below 50%. Together with lower levels of assisted births and ANC, these data suggest that access to care is a significant constraint to health in conflict and post-conflict settings.

Numerous studies suggest that post-conflict societies may vary from those not affected by conflict in several important ways, and these differences may be illuminated by a multivariable analysis. Based on the geographic patterns of risk identified above (that is, the fact that interregional variability appears to be particularly great among conflict affected countries), we would expect that location variables such as region of residence or urban-rural status might be more significantly correlated with risk among conflict-affected countries. The patterns of and motivations for recent migration are likely to be different in post-conflict as opposed to non-conflict countries as more migrants might have been forced to move due to security reasons or abject destitution. Female household headship might be a significant marker for poverty in post-conflict as compared with non-conflict settings, because more women may be widowed, or permanently separated from their spouses and because single parent families in post-conflict settings face greater obstacles to rebuilding household livelihoods. It also is likely that access to health and environmental services might be more important in the post-conflict environment, given the greater vulnerability of post-conflict populations.

In order to explore potential hypotheses regarding differences between conflict-affected and comparison countries we built the multivariable regression models in stages to 1) explore associations between study variables (migration, household headship, service coverage measures, and place of residence), and 2) to assess potential mechanisms through which these may be operating (see Appendix 2). The final regression models presented in Tables 9-12 are the full models, which include all variables that were identified through the literature or bivariate analyses.

F.4.a Correlates of childhood mortality

Many environmental, socioeconomic, and health-related factors have been linked to child mortality. Such factors, including low socioeconomic status, urban-rural residence, female-household headship, low maternal education, poor access to water/sanitation, poor access to healthcare, and migration have been associated with an increased risk of child morbidity and child mortality (Conteh et al. 1990; Cutts 1996; Robinson et al. 1999).

In this study, risk factors from the literature, as mentioned above, were selected for inclusion in the analysis. Bivariate analysis was performed to determine the relationship between major risk factors and child mortality, and the results are presented in Appendix 1. Using multivariate logistic regression, the adjusted effects of socioeconomic, environmental, household, and access-related factors on child survival were determined. The adjusted odds ratios can be interpreted as indicating the risk of a child death relative to the reference category, when all other factors in the model are held constant, and are presented in Table 9.

1. Results of Bivariate Analysis

As presented in Appendix 1, childhood mortality, as measured by proportions dead of births in the five year period preceding the survey, varies according to urban-rural residence, and illustrates wide interregional variability among the conflict-affected group. Although no distinct pattern emerges according to residence, rural residence is associated with elevated child mortality in Niger and CAR, and urban residence is associated with greater risk in the cases of Eritrea and Tanzania, while other countries indicate no significant urban-rural differentials. Additionally, among the two countries having experienced the most widespread and high-intensity conflict, Eritrea and Mozambique, regional levels of childhood mortality differ substantially, with risk increasing 3 to 4 times depending upon the region of residence.

At least one of the socioeconomic factors, as measured by household possessions, dwelling characteristics, and maternal education (literacy), are associated with child mortality in most countries, with the exceptions of Eritrea and Uganda. The proportions of children dead decreases with increasing numbers of possessions and in households with finished floors, indicating that child mortality occurs disproportionately among poorer households. Among countries showing an association between literacy and mortality, children of illiterate mothers are at a greater risk of dying.

Improved environmental conditions are associated with reduced levels of child mortality in many of the countries being studied. Risk of child mortality is increased in households using surface water as compared to a piped water supply in the countries of CAR, Kenya, Niger, and Zambia. The use of well water is also associated with greater levels of mortality in Mozambique, Niger, Cameroon, and CAR. Sanitation reflects differentials in child mortality primarily among countries affected by conflict. The lack of sanitation is associated with increased risk in Eritrea, Mozambique, Niger, and Uganda, while among the non-conflict group, significant associations are found only in CAR and Kenya.

Maternal characteristics such as age and parity had variable associations across countries, while service utilization/health behavior variables were consistently associated with child mortality. Mortality by mothers’ age varied somewhat by conflict status. Among non-conflict countries, increased levels of child mortality are consistently found among younger women (15-24), although insignificant in Zimbabwe. Yet among the conflict-affected countries there is no discernable pattern. Uganda and Niger reveal patterns similar to non-conflict countries, while Mozambique indicates higher mortality among older women (35-49); Chad and Eritrea do not reveal mortality differentials by mother’s age. Mortality differentials are also evident by service utilization. Among all countries but Chad, receiving antenatal care from a professional health worker significantly reduced the risk of child mortality. With regard to the pace of childbearing, as measured by preceding birth interval, all countries (except Zimbabwe) indicated greater risk of child mortality among births with intervals shorter than 24 months.

Male children typically experience higher mortality than female children in the absence of discriminatory allocation of resources and care (Waldron 1987). Therefore, child sex was included as a risk factor of childhood mortality in our study. Among eight of the twelve countries, female children were found to experience significantly lower levels of mortality as compared to males, while in others, the difference was insignificant.

Levels of child mortality vary by household headship and conflict status; migration patterns, on the other hand, are associated with child mortality levels primarily among the conflict-affected group. Children of women who had recently migrated in Eritrea, Mozambique and Uganda had an elevated risk of death. In countries such as Mozambique, Niger, CAR, and Kenya, children of women who had never migrated were also at increased risk of mortality, as compared to those who had migrated sometime in the past (6-35 years prior to the survey). These findings suggest that past migration may in fact have improved chances of child survival, while more recent migration induces the opposite effect, and implies that the reasons behind migration decisions may have been different, i.e. elective vs. forced migration.

2. Results of Multivariate Analysis

Table 9 presents the results of the multivariate analysis of factors associated with the likelihood of child death among births during the five-year period prior to the survey. Reflected are some similarities, as well as differences, in the patterns of correlates among countries grouped by conflict status. Consistent correlates of childhood mortality in both country sets include low socioeconomic status and short preceding birth intervals. Other important factors associated with increased risk of death, albeit less consistent, include poor service utilization (ANC), child sex (male), and surprisingly, urban residence. Environmental factors were relatively insignificant after controlling for the effects of socioeconomic status. Important differences by conflict status include factors such as migration patterns and household headship. Among certain cases of the conflict-affected countries, children of mothers who have recently migrated or in those in female-headed households were at increased risk of mortality.

Low socioeconomic status remains an important determinant of child death in all countries, thus confirming the results of the descriptive analysis; on the other hand, environmental predictors became insignificant in virtually all countries, with the exceptions of Eritrea and Kenya. In these two countries, the presence of any type of toilet facility was associated with a decreased risk of child mortality.

Urban-rural residence is typically used as a proxy for socioeconomic status in a descriptive analysis. In the multivariate model, it becomes clear that wealth and environment are the mechanisms accounting for urban-rural differentials in childhood mortality (see Appendix 2). In some cases, the relationship between urban-rural residence and mortality is strengthened, while in others it is reversed, favoring rural residence in many countries when these factors are held constant.

Of the maternal-related factors, mother’s age did not remain an important determinant of childhood mortality, although younger maternal age (15-24) was associated with higher proportions of child death in Chad, Niger, Kenya and Tanzania. On the other hand, health behavior as measured by the pace of childbearing (birth interval) and service utilization (proxied by antenatal care) remained significant covariants of mortality. Among all countries but Zimbabwe, a child born within 24 months of a previous birth was at greater risk of mortality. Alternatively, children born to a mother receiving antenatal care from a trained healthcare professional had significantly improved chances of child survival among eight of the twelve countries in our study. The variable for mother’s parity was found to be collinear with maternal age in the multivariate model, and was dropped from the analysis.

Child sex also remained an important correlate of childhood mortality in six of the twelve cases. Multivariate analysis confirmed the findings of the descriptive study, indicating that male children remain at greater risk of early childhood death as compared to females.

Important differences in determinants of childhood mortality by conflict status were found among the migration and household headship variables. Among the conflict-affected group, children of women who had migrated within five years of the survey in Eritrea, Mozambique and Uganda had an elevated risk of death. It has been suggested that heightened odds of childhood mortality from migration may occur from the disruptive effects of the move (for instance, premature termination of breast-feeding, temporary isolation from health facilities, or greater physical demands during pregnancy/early postpartum), or from maladjustment to the new surroundings after migration (for instance, an inability to quickly obtain housing, health services, and income-earning opportunities (Brockerhoff and Hewett 2000). Children in female-headed households also remained at increased risk of mortality in Eritrea, while the bivariate association between female headship and mortality in Mozambique became insignificant in the multivariate model. Among the non-conflict group, Zambia and Tanzania also reflect positive associations between female headship and child mortality.

When region of residence was included in the full model, the above effects did not change considerably. However, several countries had regional variables that were significantly associated with child mortality. This indicates that regional differences were not completely accounted for by the previous factors in the model, and something unique to particular regions further influenced child survival. Regional differentials in child mortality were most notable among conflict-affected countries, particularly Eritrea and Mozambique, where the odds of child death increased by three to four times depending upon region of residence.

F.4.b Correlates of childhood stunting

Nutritional status is often used as an indicator of children’s overall health and nutrition. Undernutrition (stunting, wasting, and underweight) often results from the synergistic effects of repeated, improperly treated illnesses, especially diarrhea, and inadequate food intake. It has been established that undernourished children are at greater risk of mortality than well-nourished children (Pelletier et al. 1993), and therefore the assessment of the nutritional status of children in a given population, combined with knowledge of the mortality rate among this age group, can reveal changes in general dynamics of health and nutrition.

Biodemographic, socioeconomic, environmental and service utilization factors have been identified in the literature as having pronounced effects upon children’s nutritional status. In this section, we will look at the associations between these factors and stunting, indicated by height-for-age more than two standard deviations (-2SD) below the NCHS reference median. Stunting reflects past or chronic undernutrition and results from inadequate food intake over a long period, and/or repeated episodes of illness. Bivariate analysis of stunting by biodemographic, socioeconomic, environmental, service utilization/behavior, and expected conflict-related risk factors (household headship, migration, and residence) was performed, and the results are presented in Appendix 1. Multivariate analysis was also performed in order to explore important determinants/mechanisms of stunting (see Table 10) among both conflict-affected and non-conflict countries.

1. Results of the Bivariate Analysis

As presented in Appendix 1, there are consistent differences in the levels of stunting among children age 6-35 months between urban and rural areas. Stunting is strongly associated with rural residence in all countries of this study. When stunting levels are examined by region within each country, there appears to be great variability among regions within both country sets regardless of conflict status.

Socioeconomic status of children born in the five years preceding the survey was determined from variables related to household possessions, dwelling characteristics (floor), and mother’s education (literacy). Against these socioeconomic factors in all countries, stunting was consistently higher among households where household possessions were few in number, where there was an unfinished (dirt/dung) floor, and where the mother was illiterate, reflecting the overall association between malnutrition and poverty.

Unfavorable environmental conditions as measured by type of water source and absence of a toilet facility were consistently associated with higher levels of stunting. In most countries, an increased risk of childhood stunting was associated with the use of wells and surface water, as compared to piped water sources. Stunting levels were also significantly increased in households where there was no toilet facility. Water source and type of toilet facility, although related to household socioeconomic status, tend to have more of an impact on nutritional status through their relationships with disease. Poor water/sanitation puts infants and young children at risk of increased incidence and duration of illness, particularly diarrhea (Haggerty et al. 1994).

Maternal and child biodemographics are associated with differential levels of childhood stunting. Stunting levels by mother’s age vary according to conflict status. Among the conflict-affected countries, specifically Mozambique, Niger, and Uganda, stunting is highest among children of the youngest mothers (15-24), which may indicate these mothers lack the resources for and/or knowledge about proper child care and child feeding. There is no clear pattern by mother’s age among the non-conflict countries. The pace of childbearing (birth interval) also reflects an association with stunting, where preceding birth intervals of less than 24 months put children at greater risk in five of the countries. In addition, children of mothers of higher parity (2+) are at increased risk of stunting. Short birth intervals may lead to shortened duration of breast-feeding and/or less maternal time available for child care. Similarly, mothers of high parity have other children to care for, and may lack the time or resources to ensure proper nutrition and adequate health care. Differentials in stunting levels by child’s sex occur primarily among the non-conflict countries, where being male and more stunted is reflective of the tendency seen elsewhere for boys to have higher rates of infectious diseases than girls in early childhood (Arnold 1996).

When stunting is examined by service utilization factors, more stunting occurred when the child’s mother received antenatal care by non-medically trained personnel or no one at all. While antenatal care can provide mothers with valuable information about child care, health, and nutrition, these associations probably reflect a mother’s overall propensity to use modern health care services, which would impact the severity and frequency of childhood illness and consequent health and nutritional status.

Of the hypothesized conflict-related risk factors, household headship was associated with stunting in the two countries most affected by conflict in our study, but migration status showed no clear pattern. In both Eritrea and Mozambique, children in a female-headed household were at greater risk of stunting as compared to those in male-headed households. Among the non-conflict cases, Zambia also indicates increased risk by female headship.

2. Results of the Multivariate Analysis

Table 10 presents the results of the multivariate analysis of factors associated with the likelihood of stunting among children 6-35 months. Factors associated with stunting across all countries include those variables related to low socioeconomic status. Other factors, such as environment and maternal/child bio-demographics remain variable. Correlates of stunting particular to conflict-affected countries include female headship, as in the cases of Eritrea and Mozambique.

In the multivariate model, socioeconomic factors remain perhaps the most important determinant of stunting in most countries. At least one of the socioeconomic variables related to household income – higher number of possessions, finished floor, and maternal literacy – was associated with a reduction in the levels of stunting, except in the cases of Niger and CAR. Upon controlling for socioeconomic status, the urban-rural differential among stunting levels disappears in all countries but Chad.

Differentials in stunting levels by environmental factors (found in the bivariate analysis) became virtually insignificant among the conflict-affected countries, while associations were variable across the non-conflict group. Poor water and/or sanitation conditions remained positively associated with stunting in Cameroon, Central African Republic, Kenya, and Tanzania. Among the conflict-affected countries, only in Mozambique was water source (well versus piped) associated with an increased risk of childhood stunting.

Among the biodemographic and service utilization factors, the multivariate analysis confirms the findings of the descriptive analysis with regard to mother’s age and child spacing (birth interval). Mother’s age remains an important covariant of stunting among several of the study countries, with children who are born to young mothers put at heightened risk. Most notably in Mozambique, children born to mothers age 15-24 are two to three times more likely to be stunted as compared to the reference group (25-29). Children born within 24 months of a previous birth are also at increased risk in several of the study countries. Among service utilization factors, antenatal care by a trained health care professional, which was found to be highly significant in the bivariate analysis, became relatively insignificant in the full model.

One important difference among correlates of stunting by conflict status was found when looking at stunting levels by household headship. Even after controlling for socioeconomic factors, female headship remained an important determinant of stunting only in Eritrea and Mozambique, the two post-conflict countries in our study. The magnitude of these associations was, in fact, strengthened in the full multivariate model, and the stepwise analysis suggests socioeconomic mechanisms. When region of residence was added to the full model, regional differences continue to account for some of the variability in stunting levels among both conflict and non-conflict countries.

A similar analysis was performed to look at risk factors of wasting among children 6-35 months. Wasting is indicated by weight-for-height more than two standard deviations (-2SD) below the median of the NCHS reference median, and reflects recent or acute malnutrition. Wasting results from a recent shortage of adequate nutrition and/or recent or current illness. Bivariate analysis of wasting by biodemographic, socioeconomic, environmental, service utilization/behavior, and expected conflict-related risk factors (household headship, migration, and residence) is presented in Appendix 1. Multivariate analysis was also performed to further explore important determinants/mechanisms of wasting (see Table 11).

1. Results of the Bivariate Analysis

For six of the eleven countries where child anthropometry is available, wasting among children is more prevalent in rural than urban areas; among the others, no urban-rural differential is present (see Appendix 1). Within countries, wasting levels exhibit regional variability, with children living within certain regions of Mozambique and Kenya at ten times the risk of children in the reference region.

A relationship between socioeconomic status and childhood wasting is seen among all countries. Wasting levels consistently decrease as the number of household possessions increase, the only exception being in Tanzania. A finished floor was also associated with lower levels of wasting among most countries. Among illiterate mothers the prevalence of wasting is highest in all countries but CAR and Zimbabwe.

Unfavorable environmental conditions as measured by type of water source and lack of a toilet facility were generally associated with higher levels of wasting. Children from houses with piped water are less likely to be wasted than other children, with a few exceptions. Among the conflict affected countries, specifically Chad, Mozambique and Niger, both well and surface water sources (rivers. lakes, streams, canals, etc.) reflect an increased risk of wasting, as compared to piped water. Uganda showed similar but insignificant results. Among the non-conflict countries, two countries indicate an association between wasting and the use of well water, while two others reveal an association with surface water. With regard to sanitation, the proportion of children classified as wasted tends to be highest among children whose family reported no toilet facility, although this pattern is not observed in all countries.

Wasting levels by maternal biodemographics are variable. Wasting by mother’s age and parity reveal no consistent pattern with the proportion of children who are wasted. Nor is there a consistent pattern in the level of wasting by length of the preceding birth interval.

Our service utilization variable, antenatal care by a trained health professional, indicated some association with wasting levels. Six of the eleven countries indicate lower levels of wasting among children whose mothers received antenatal care by a trained health worker, as compared to non-trained personnel or no one at all. As stated earlier, this variable probably reflects a mother’s health seeking behavior, and her overall propensity to use modern health care facilities in the management of childhood illness.

Of the conflict-related variables, migration status and household headship, female-headship was associated with an increased risk of childhood wasting in three of the study countries. In Eritrea and Mozambique, children in female-headed households were at 1.5 times the risk of wasting as compared to children in male-headed households. Among the non-conflict countries, Tanzania also indicated increased risk of wasting by female-headship. Zimbabwe was the only country to show reduced levels of wasting among children in households with female heads. Wasting by migration status revealed no consistent pattern.

2. Results of the Multivariate Analysis

Table 11 presents the results of the multivariate analysis of factors associated with the likelihood of wasting among children 6-35 months. Similar correlates of wasting across countries include low socioeconomic status and poor environmental conditions, primarily via household water source. Factors particular to conflict-affected countries are recent migration status and female headship. In Mozambique, child spacing (< 24 months) was also associated with an increased of wasting.

As with stunting, socioeconomic factors remain an important determinant of wasting in most countries. At least one of the socioeconomic variables related to household income – higher number of possessions, finished floor, and maternal literacy – was associated with a reduction in the levels of wasting, except in the cases of Uganda and Tanzania.

Upon controlling for socioeconomic status, the urban-rural differential among wasting levels disappears in all countries but Uganda and Cameroon. In these two countries, an increased risk of wasting among children is associated with urban residence.

Poor environment, as measured by water source and sanitation type, remained a determinant of childhood wasting in six of eleven countries. Among the conflict-affected countries of Mozambique and Uganda, the use of wells (also surface water in Uganda) was associated with an increased risk of wasting, while among the non-conflict countries, the use of surface water was a risk factor in 3 of the 6 countries. This finding may indicate that wells in conflict-affected countries may be in poor condition, due to inadequate servicing of pumps or greater risk of contamination. After controlling for water source, the effect of poor sanitation was a significant factor only in the cases of Cameroon and Tanzania.

Maternal biodemographics remained variable among countries in the multivariate analysis. Wasting levels by mother’s age and preceding birth interval revealed no consistent patterns, although short birth intervals (< 24 months) were associated with an increased risk of childhood wasting in Mozambique.

Service utilization (ANC), which was associated with reduced levels of wasting in the bivariate analysis, reflected less importance in the full model. The use of antenatal care by trained health personnel remained associated with reduced risk of wasting in Mozambique, Niger, and Cameroon.

Among the conflict-affected countries, variables such as migration status, household headship, and region of residence remained important factors associated with childhood wasting. Female headship is associated with increased levels of wasting in Eritrea and Mozambique, while recent migration (past five years) is also a risk factor in Eritrea, Mozambique and Niger. When region of residence was included in the model, regional differences were found to account for some of the remaining variability among wasting levels. This indicates that something unique to particular regions further influenced child nutritional status. Regional variability in wasting was most notable among conflict-affected countries, particularly in Mozambique, where the risk of wasting in certain regions was eleven times greater than in the reference region.

A woman’s nutritional status is an important determinant of pregnancy outcome for both mother and child, and contributes to morbidity and mortality that may follow a birth. Maternal height is commonly associated with birth weight, child survival, maternal mortality, and pregnancy/birth complications, while prepregnancy weight determines the characteristics of pregnancy, e.g. weight gain, premature birth, etc. (Loaiza 1997). A composite measure of height and weight, the body mass index (BMI), can be used to characterize the nutritional status of women. A BMI < 18.5 indicates chronic energy deficiency (CED), and puts a women at high risk of unfavorable maternal outcome upon becoming pregnant.

The literature identifies several factors associated with maternal nutritional status, including socioeconomic, biodemographic and environmental characteristics. Variables from the DHS capturing these characteristics along with expected conflict-related risk factors (FHH, migration, and residence) were included in the bivariate (see Appendix 1) and multivariate analyses (Table 12).

1. Results of Bivariate Analysis

There are also consistent differences in the level of maternal malnutrition between urban and rural residence, while regional differences within countries are somewhat more variable. CED is more common in rural areas in all countries except Chad, where CED is an urban phenomenon. Differentials between high-risk regions and low-risk regions (base) were found to vary according to conflict status. In conflict-affected countries such as Chad, Eritrea and Mozambique, certain regions are more susceptible to CED, with up to 6-9 times more CED than that of the base region. By comparison, non-conflict countries reflect smaller risk differentials, with risk increasing up to three-fold in the worst regions.

Socioeconomic status of women with children born in the five years preceding the survey was determined from variables related to household possessions, dwelling characteristics (floor), and mother’s education (literacy). In all countries, the higher the level of socioeconomic status as determined by a possession scale, the lower the risk of mothers with CED. Similarly, improved dwelling characteristics show a similar pattern, with the exception of Chad. In Chad, a "finished floor" is actually associated with an increased risk of CED, as may be expected in the only country where urban residence is associated with CED. Mother’s education, which is proxied here by female literacy, also is associated with lower levels of CED in most study countries.

Biodemographic variables, or fertility-related variables, include mother’s age, parity, and pace of childbearing (birth intervals). Generally BMI increases with age, mainly due to weight gain throughout the reproductive years. Among the conflict-affected countries this pattern holds true, with a greater proportion of younger women exhibiting CED. In contrast, an increased risk of CED appears among older women (35-49) in the non-conflict countries, with the exception of Zimbabwe. Zambia reflects a U-shaped distribution of CED by mother’s age, which may be indicative of long-term maternal depletion. As parity is associated with mother’s age, a similar pattern of lower CED at higher parity is found across the sample, with a few exceptions. Only in Mozambique was CED associated with a short birth interval.

In general, unfavorable environmental conditions were associated with higher levels of CED. In most countries, an increased risk of CED was associated with the use of wells of surface water, as compared to piped water sources. The presence of any toilet facility, regardless of type, was found to be negatively associated with CED in most cases as well.

Migration patterns and household-headship are generally affected by conflict, and were evaluated for associations with CED. The bivariate analysis does not reveal consistent differences in CED levels by headship, although female-headship is a protective factor in Uganda. Migration status, on the other hand, does indicate an association with CED in certain cases. Among the conflict-affected countries, there is an increased risk of CED among those who have never migrated in Eritrea and Mozambique; in Eritrea, mothers having recently migrated (past 5 years) were also at an increased risk of CED.

2. Results of Multivariate Analysis

The primary objective of the multivariate analysis is to examine the relationship of each variable with maternal chronic energy malnutrition when the effects of other variables are held constant, and to compare the results across countries according to conflict status (Table 12). To further explore mechanisms, variables were entered into the models in a stepwise fashion (see Appendix 2). As with child nutritional status, low socioeconomic status is correlated with poor maternal nutritional status in most countries. Among the conflict-affected countries, other factors such as environment, migration status, household headship, and maternal characteristics are also determinants of CED. Mozambique is a unique case, where additional factors entered into the model such as birth spacing and residence also remained associated with poor maternal nutrition outcomes.

Of the socioeconomic variables, including the possession scale, finished floor, and maternal literacy, all countries had at least one variable significantly explaining the variations in CED, thus confirming the results of the bivariate analysis. In fact, for the non-conflict countries, socioeconomic status is the primary factor determining CED, indicating that poor maternal nutritional status is most prevalent among poorer groups. By contrast, among conflict-affected countries, socioeconomic factors only partially account for the variability in CED, and other factors continue to play a role.

After controlling for socioeconomic status, urban-rural differentials found in the bivariate analysis become relatively unimportant, except in a few cases. Residence remains an important determinant of CED in Mozambique, CAR, and Tanzania, though in Mozambique, urban residence rather than rural is associated with greater risk. Stepwise analysis indicates that the rural correlation found during the bivariate analysis operated primarily through socioeconomic and environmental mechanisms.

Environmental conditions remain important determinants of CED only among the conflict-affected countries. In both Eritrea and Mozambique, the use of surface water is associated with an increased risk of CED as compared to a piped water source. In Mozambique, the use of wells is also associated with increased risk, and further bolsters our hypothesis that wells may be in disrepair or contaminated in this country. Sanitation is also a determinant of CED among conflict-affected countries, with CED most prevalent among mothers in households lacking any type of toilet facility.

Among biodemographic variables, mother’s age remains an important determinant of CED among both conflict and non-conflict countries, although among different age groups. Among the conflict-affected countries of Chad, Eritrea, Mozambique, and Uganda, younger mothers (15-24) are most at risk. In the non-conflict countries, either there is no differential by age group, or the older age group (35-49) is most at risk, as in the cases of CAR and Zambia. In terms of CED by previous birth interval, Mozambique was the only country to exhibit significant associations. In Mozambique, mothers with birth intervals shorter than 24 months are at higher risk of CED. Parity was found to be collinear with maternal age, and was dropped from the full model.

Of those variables that we hypothesize to be most affected by conflict – migration status and household headship – both are associated with CED among the conflict-affected group. Interestingly, those having never migrated in Eritrea and Mozambique, and those who have recently migrated in Eritrea, experience higher levels of CED. Household.34 headship was also a determinant of CED among the conflict-affected countries, where female-headship was in fact a protective factor in the cases of Chad, Mozambique and Uganda.

Although inherent biases in DFR (Kirk and Pillet 1998) exist in countries where infant and child survival rates are low, desired births may in fact exceed total fertility rates. Indeed, where mortality is a salient threat, fertility levels are typically high, as families seek to ensure adequate numbers of surviving children. In addition, high mortality affects fertility through the loss of the contraceptive effect of breast-feeding and the traditional period of postpartum abstinence typical of many SSA populations.

However, modern contraceptive use also is extremely low in countries that have been recently affected by conflict, reflecting, in part, poor access to family planning services. In none of the conflict-affected countries does the modern contraceptive prevalence rate (CPR) exceed eight percent. Even in Uganda, a country that is making a rapid economic recovery, the CPR was only 7.8. Throughout SSA, where CPR rates are low and where DFRs remain relatively high, the CPRs found within the conflict-affected countries are strikingly low.

In general, these findings suggest that an increased emphasis should be placed on family planning programs in the post-conflict context. In the case of Eritrea, infant and child mortality rates are not extraordinary (though still unfortunate) in the SSA context. In these countries, means to improve birth spacing may well be welcomed by women. In addition, the high variability in mortality rates among women living in different regions of Mozambique and in the urban areas of Sahelian countries suggests that geographically targeted programs should also be considered.

As might be expected, post-conflict countries or those being affected by regional or local conflict are quite diverse in terms of their health, population, and nutrition profiles, as well as health-related behaviors. Even in countries that have experienced wide-scale and protracted conflicts, such as Eritrea and Mozambique, childhood mortality and nutritional experiences vary. Mozambique is characterized by very early childhood mortality rates but relatively unremarkable levels of undernutrition, while Eritrea seems to show the opposite tendency. These findings suggest the relative importance of attention to health care and food security needs in these two countries, respectively, and point to the need for varying HPN strategies in post-conflict environments. The Eritrean example also demonstrates the importance of examining population coping mechanisms that often evolve in the face of adversity and that may have implications for long-term programming, such as features of the primary health care system that evolved under the EPLF.

Some similarities among conflict-affected countries, particularly those affected by generalized and intense conflict, are noteworthy. Of particular importance is the high level of interregional variability in health and nutrition status found in these countries. This finding is not surprising, given the spatial and temporal variability in the nature of conflict and its sequelae, as well as cross-border interactions between populations from stable and conflict-affected countries. This high variability suggests the need for greater attention to targeting considerations in conflict-affected countries. It also means that DHS surveys should be designed to take into account the importance of the need to disaggregate data.

Another important differential apparent from these contrasts is the lack of correlation among mortality and nutritional indices, suggesting that the ecology of mortality and malnutrition might be somewhat different in conflict-affected as opposed to stable settings. The general nature of the DHS does not permit detailed analysis of the differential determinants of mortality and nutrition.

These findings also make clear the consistent need to rebuild basic health and population services in conflict-affected countries. Indeed, although the absence of an infrastructure can appear to be daunting, the opportunity for profound health sector reform is considerable. When countries have undergone loss of the physical and human infrastructure, it is often possible to promote the rebuilding of those sectors consistent with current best practices. In this way, many of the bureaucratic constraints of stable countries are avoided.

Some of the most striking differences between the conflict-affected countries appear in the variables associated with migration and female-headed households. Migrants and members of female-headed households should be of special interest in the design of interventions in conflict-affected countries, particularly those affected by generalized conflict and complex emergencies. Although this finding might have been expected, it is still significant for planning interventions in post-conflict societies. Among the study countries included in this analysis, women who had recently changed residence are somewhat better off socioeconomically than those who had not moved in the five years prior to the survey. Yet, in the two countries most affected by recent conflict, Eritrea and Mozambique, migration within the past five years was associated with a significant increase in the probability of child death in both urban and rural environments. In the multiple variable analysis, migration status increased in significance when socioeconomic factors were held constant. The analysis also shows that residential fluidity appears to be more common in countries that have experienced widespread conflict, signaling the particular need to address the health problems of these populations. Migration among the destitute appears to be of particular concern.

As the migrant group appears to be diverse in terms of its socioeconomic and health characteristics, it would be helpful to better understand the migration status of households, including reasons for migration. The literature suggests that those who migrate due to loss of land, resettlement after forced migration, or forced resettlement in designated areas (Robinson et al. 1999; Cliffe 1994) are at particular risk of poor health outcomes.

Female household headship is also of particular concern to planning in the post-conflict environment. Although households headed by women were no more likely to experience poor health outcomes in relatively stable African countries, the dynamics of conflict and post-conflict environments appear to differentially affect female-headed households. Our analyses suggest that some of the excess risk is associated with socioeconomic factors, as female-headed households in both urban and rural settings are socioeconomically disadvantaged as compared with households not headed by women. However, even when socioeconomic and environmental factors are controlled in a multivariable regression model, the female-headed household remains significant. Again, the DHS does not collect key information that may explain the nature of the mechanisms for this relationship. Other literature suggests that women may not be eligible for land tenure, for example in the case of Mozambique (Carver 1995). In the post-conflict environment, increased vulnerability among female-headed households may also be due to permanent loss of husbands to the household economy (because they were killed or separated from their families) than is the case in non-conflict-affected countries.

Another finding of interest for program planning and future DHS surveys in conflict-affected countries is that access to social services is particularly important, but may not be well measured by the DHS protocol as it currently stands. In the case of Mozambique, for example, wells seem to remain in disrepair and access to basic health, educational and environmental services is typically variable as recovery interventions are put into place. Currently, DHS protocol includes an optional service availability module, which does not include information that may be particularly important in post-conflict settings. Important service information at the community level includes repair of water systems, schools, roads, existence of functioning markets, and other factors.

Finally, the high levels of "destitution" in conflict-affected countries point to the great need for investment in programs to improve the livelihood of households. The countries in this analysis that have experienced the greatest levels of conflict, Eritrea, Mozambique and Chad, have extreme levels of poverty, as reflected by housing characteristics and a simple possessions scale. In turn, socioeconomic status remains a strong predictor of health outcomes, even in the presence of health services. Safety net programs together with programs aimed at rebuilding household livelihoods are particularly important in the post-conflict environment.

However, the measurement of socioeconomic status used by DHS is also likely to be missing important information. Debt burden, access to land and animals, and the extent of recovery or transition programs are factors that may have a significant impact on the household capacity to rebuild their livelihoods and to create good health but are not covered by DHS. Implications of these findings for implementing DHS in conflict-affected countries are discussed in the next section of the paper.

Improving the utility of the DHS program in conflict-affected settings

DHS is an important data collection vehicle that is being widely used in SSA for a variety of applications. Traditionally, DHS has been used as a national planning and monitoring tool. Health, population, and nutrition information together with basic risk factors are often analyzed in combination with other information to develop sectoral and intersectoral strategies and programs. This application is particularly critical in the post-conflict environment, where population representative data are sparse. Many countries now have multi-year DHS survey rounds planned so that they can track the progress of these indicators. In the post-conflict context, DHS also is important for planning and monitoring health-related programs. With some modification, DHS also can be used to monitor the coverage of programs specifically aimed at improving the recovery of nutrition and health, for example, food aid, livelihood, and mental health programs.

DHS also is being used, at the regional level, for comparative analysis of demographic and health factors, in order to identify countries of particular need and as a basis for health, population, and nutrition sector policy analysis. Given its methodological standardization and wide-scale application in SSA, DHS is a unique program in this respect. In relation to conflict, DHS is one of the only available vehicles for gathering comparative data on the demographic impact of conflict and complex emergencies in SSA.

Additionally, DHS is increasingly being used as an instrument of impact evaluation. Together with advanced statistical techniques, repetitive surveys are being combined with community-level information to isolate the impact of health, population, and nutrition interventions on demographic and health outcomes. This application might be used in two different situations in post-conflict countries. In the case that DHS might be used to evaluate the impact of a family-planning program or other traditional health program, measuring contextual variables unique to conflict settings may be important to "control" for extraneous variables in a statistical analysis. For example, in the post-conflict setting, community-level factors affecting access might include neighborhood safety. Evaluating the effect of programs on other outcomes, such as childhood mortality, will require community level information on infrastructures, including for example the availability of safe water. DHS also might be used to evaluate the impact of specific transition programs, for example food aid programs that are large-scale recovery programs. The use of food aid for transition programming is an important area in need of solid evaluation research, and the DHS could serve as a vehicle for implementing this research in settings where food aid programs constitute key strategies for recovery, for example in the Horn of Africa.

The regional nature of conflict and its pervasiveness in SSA has implications for all of these applications of DHS. The results of this study reflect the general conclusion that few African countries have been untouched by conflict, either by direct experience or indirect economic and cross-border effects. This situation deserves consideration in the development of strategies to apply the DHS and other cross-national survey programs in the African context. Several suggestions can be made to improve the utility of DHS as a program planning and evaluation instrument. These include:

1. Regional coordination of country or partial country surveys in areas such as the "Great Lakes" region, parts of West Africa, and the Horn.

2. Increased emphasis on geo-referencing DHS data collection and spatial analysis of DHS data.

3. Improved coordination among survey programs, such as the Living Standards Measurement Surveys, the UNICEF Multi-Indicator Cluster Surveys, and other agricultural/socioeconomic surveys to address the information requirements of post-conflict settings.

4. Modifications to the sample design to include temporary settlements in some countries, and to extend criteria for inclusion of households to those headed by children, a common occurrence in some conflict-affected populations.

5. Enhancement of the information content of DHS data collection protocols for optional application in conflict-affected countries.

6. Conducting research to assess the effects of conflict on the accuracy of DHS methodology when applied to post-conflict settings.

 

It is increasingly apparent that even in SSA, where the communications and transportation infrastructure is poor, many health-related problems have a cross-border dimension to them. This is particularly true in the case of conflict-affected areas, where cross-border population movement is a significant issue, most obviously manifested in the formation of refugee camps and displaced population groups, but it often also affects the flow of economic resources and even the use of health and social services. In the case of conflict, regionally coordinated surveys could greatly enhance the ability to better understand migration patterns associated with conflict and to better predict and monitor the cross-border dimensions to population coping patterns.

Consistent with this recommendation is to design DHS to permit spatial analysis of DHS data. This includes ensuring adequate sample size and geo-referencing cluster level data. In the case of conflict, administrative boundaries appear to be more limiting than in more stable contexts, given the high intra- country variability in health-related outcomes and correlates, together with the cross-border nature of conflict and its consequences. Our simple thematic mapping exercise demonstrates the potential utility of geographic analysis of DHS data in the post-conflict environment. Cluster mapping and the applications of GIS techniques would greatly enhance the use of DHS for planning and evaluation in the post-conflict context. An example of the feasibility of spatial analysis of DHS data already has been demonstrated for the West African region (Croft et al. 1997). Cluster mapping will also be increasingly important as DHS is used as an instrument of program impact evaluation.

Better coordination among large-scale household survey programs also could result in improved utility of DHS data for post-conflict applications. Three survey programs, sponsored by international donors, are particularly useful for comparative policy analysis and social programming applications in SSA (Table 14). The Living Standards Measurement Surveys (LSMS) is an in-depth longitudinal socioeconomic survey program, supported by the World Bank, that contains a core module that is relatively standard across all participating countries and has additional country-specific information. The LSMS is particularly useful for analyzing the socioeconomic determinants of health status, and the longitudinal component of the program also provides the opportunity to investigate change in these factors. In addition, the sampling frame is not limited to households with women between 15 and 50 years, therefore, making it more representative of the general population. Limitations of the LSMS for planning and evaluation relating to health outcomes are several. The LSMS has only been fielded in a limited number of African countries, including Tanzania, Ivory Coast, Zimbabwe, and Kenya. The survey program is relatively expensive. Health-related information and information related to the proximate determinants of health status are limited. Standardization in the application of survey methodology is somewhat variable across survey programs, and the collection, processing, analysis and dissemination of information is slow, often taking years. To date, LSMS has not been implemented in a post-conflict country, though great need exists for longitudinal information on socioeconomic recovery in post-conflict settings.

On the other hand, the DHS program is well recognized as the most complete and standardized field survey program addressing health, population, and nutrition sector information needs. Information is collected across a wide variety of health-related outcomes, behaviors, and programmatic areas. Increasingly, the service availability module is being implemented to gather information on community-level characteristics. In so doing, the effects of community-level variables, such as the nature and availability of social services and accessibility of populations can be assessed. Limitations of the DHS include relatively limited measures of socioeconomic characteristics of households and the absence of information on migration of household members. Other information important to the post-conflict setting is discussed further below.

Because of its popularity as a survey vehicle, countries often add numerous modules on to the DHS. Because of the length of the survey in these cases, respondent fatigue may result in survey error. This may be a more important consideration in post-conflict settings if respondents are reticent to participate in survey programs anyway.

All three of these survey programs, however, can provide valuable information for programming and evaluation in the post-conflict context. Closer collaboration among these programs can lead to considerable gains in information and efficiency. A commonly missed opportunity for coordination is sample overlap, which provides the opportunity to link information from two or more surveys at the individual level. In the case of LSMS, for example, socioeconomic and consumption data are extensive in comparison with that in DHS, which contains more extensive data on individual health, nutrition, and fertility status. The ability to link data at the household and individual level across these surveys would result in a great enhancement of socioeconomic information without further burdening the DHS survey protocol. The LSMS also can be used to explore the dynamics of intervention strategies, particularly those that address food security and livelihood, at the household level. Together with DHS, which tends to measure more basic markers of socioeconomic status but more in-depth information on health outcomes, the two survey programs may monitor changes in outcomes while having some confidence in the mechanisms through which interventions are resulting in these changes.

Similarly, DHS and MICS can be coordinated to better facilitate planning and evaluation activities. For example, while DHS might be used to identify regions of the country at particular risk as well as the risk factors to health and nutrition, MICS could be oversampled in these areas and repetitive survey rounds could be planned even within a one or two year time frame. In this way, MICS could be applied as a monitoring tool to track recovery of population groups particularly affected by conflict. MICS also carry some potential to implement surveys in unstable contexts, which DHS cannot. MICS is sponsored and can be identified with UNICEF, which may, in some cases, be viewed as a well-received organization by most. UNICEF typically maintains field presence during periods of conflict and instability and for this reason is typically well appreciated by local populations. In these cases, MICS and DHS might be coordinated such that MICS are undertaken early during the transition from relief to development activities and DHS implemented somewhat later.

In some countries, such as Uganda, parts of the Democratic Republic of Congo, and Guinea, refugees living in temporary settlements comprise a large population group. Past experience suggests that many "temporary settlements" often persist for more than five years. From a planning perspective, areas that have large refugee populations are important from several standpoints. First, as they often draw upon local social services, it is important to have quantitative data on their health characteristics and behaviors for planning purposes. From an evaluation perspective, international humanitarian assistance standards require that refugees have standards of care and health that are comparable to that of their hosts. In fact, in many cases, refugees have enjoyed higher standards, which should signal a readjustment of resources to ensure that local host populations enjoy similar access to resources. Another important reason for the inclusion of refugees is that no systematic information is currently collected on the demographic and health status of refugees with the exception of ad hoc rapid nutrition, and less often, mortality surveys. As such, little is known about the migration, mortality, health, and nutritional status of these populations, nor is much known about their socioeconomic status. As these population groups represent a substantial segment of the population in some provinces/states in a few countries, their inclusion in the DHS program should be considered.

Child-headed households also represent a considerable population group in some countries, although their extent is not known due to lack of data. UNICEF estimates that more than 12% of households were recently headed by children in Rwanda (Machel 1996). Households headed by children are at particular risk of poor health outcomes and may pose unique challenges to the range of social programs aimed at promoting recovery and social development. Yet this segment of the population is currently excluded from the DHS sampling frame.

Perhaps the most important limitation of DHS in the SSA context is that the information content of the survey does not address the unique character and needs of conflict-affected populations. Conflict affects population health, population, and nutrition through changes in household structure, destitution, and physical and mental trauma. Traditional problems, such as infectious disease threats and poor access to services, are compounded by the new problems associated with violence, trauma, lack of access to land, animals, and basic agricultural inputs to meet basic needs. If violence or distrust, or both of these persist, the propensity to use health and social services may be less than is typically the case in more stable environments. The physical and social infrastructure may be partially or fully destroyed. Services may be functioning, but only at limited capacity. Poor transportation routes and systems also may render population access particularly challenging.

Although current DHS protocol provides useful information for planning and evaluation in conflict-affected settings, much important information is not provided. Table 15 summarizes suggested modifications to DHS information content in post-conflict settings together with the rationale for these modifications. Some changes require only small adjustments to the DHS core survey or existing modules, but others would require either a "post-conflict module" to be added to the field protocol or a follow-on survey to be conducted, using the DHS sample. An illustrative set of instruments is provided under separate cover.

Finally, little information exists on the nature and magnitude of survey errors resulting from conflict. At least two sources of error exist, including sampling error that may result from faulty sampling frames and reporting bias that may arise from respondent suspicion or other conditioning related to the conflict experience. No published research is available on the nature or magnitude of either of these sources of bias. Both of these are potentially important and should be assessed.

ACC/SCN. 1993. Second Report on the World Nutrition Situation, Volume II, Country Trends,Methods and Statistics. Geneva: ACC/SCN.

Wallensteen, Peter and Margareta Sollenberg. 1998. Armed Conflict and Conflict Complexes, 1989-97. Journal of Peace Research; 35(5): 621-634.