The impact of civil conflict on infant and child malnutrition, Nigeria, 2013

Abstract The new millennium brought renewed attention to improving the health of women and children. In this same period, direct deaths from conflicts have declined worldwide, but civilian deaths associated with conflicts have increased. Nigeria is among the most conflict‐prone countries in Sub‐Saharan Africa, especially recently with the Boko Haram insurgency in the north. This paper uses two data sources, the 2013 Demographic and Health Survey for Nigeria and the Social Conflict Analysis Database, linked by geocode, to study the effect of these conflicts on infant and young child acute malnutrition (or wasting). We show a strong association in 2013 between living close to a conflict zone and acute malnutrition in Nigerian children, with larger effects for rural children than urban children. This is related to the severity of the conflict, measured both in terms of the number of conflict deaths and the length of time the child was exposed to conflict. Undoubtedly, civil conflict is limiting the future prospects of Nigerian children and the country's economic growth. In Nigeria, conflicts in the north are expected to continue with sporadic attacks and continued damaged infrastructure. Thus, Nigerian children, innocent victims of the conflict, will continue to suffer the consequences documented in this study.


| INTRODUCTION
The new millennium brought renewed attention to improving the health of women and children, resulting in the adoption of the Millennium Development Goals for 2015 and, more recently, the Sustainable Development Goals for 2030 (World Health Organization, 2015). In establishing these goals, the nations of the world came together to attempt to reduce hunger, infant mortality, and child mortality, with specific targets worldwide. Developing countries, especially Sub-Saharan African countries, face particular challenges meeting these ambitious goals.
In this same period, direct deaths from conflicts have declined worldwide, but civilian deaths associated with conflicts have increased Kruk, Freedman, Anglin, & Waldman, 2010;World Health Organization, 2015). One study estimates that in 2016, 59% of the world's children lived in conflict-affected countries (Bahgat, Dupuy, Ostby, Rustad, Strand, & Wig, 2018). Armed conflicts are more likely to occur in poor countries with weak states, particularly in sparsely populated rural areas and in places where there is competition for land and water (Blattman & Miguel, 2010).
Numerous studies have shown the negative health impact of such conflicts on civilians, particularly children. Table 1 summarizes findings from 17 empirical studies, including one pan-African study of 35 countries, 11 African country-specific studies, and five non-African country-specific studies. All concern the impact of armed conflict on maternal and child health. Although no study is completely free from methodological limitations, the combined results suggest what is intuitively obvious that-across many countries-there is a substantial negative effect on maternal and child health from war and conflict.
The most recent study, and most methodologically rigorous, shows a clear relationship between African conflicts and infant mortality in 35 countries over a substantial time period (Wagner, HeftNeal, Bhutta, Black, Burke, & Bendavid, 2018). However, caution is needed because even the most rigorous studies show a correlation that may or may not be causal. For example, one study in Ethiopia concluded that although there was a correlation between conflict and child health, after controlling for drought conflict events did not have a significant effect on child health (Delbiso, Rodriguez-Llanes, Donneau, Speybroeck, & Guha-Sapir, 2017). In addition, the size of effects cannot be generalized across the studies because of substantial methodological differences. For example, the child health outcomes in these studies are quite varied, including various measures of poor nutritional status, such as small weight-for-age, height-for-age, or weight-forheight ("wasting" or "acute malnutrition"). Counting the number of conflicts and the characteristics of those conflicts is challenging and different data sources tally such events in different ways, each with potential biases (Johnson, Spagat, Gourley, Onnela, & Reinert, 2008).
The most common limitation among the studies cited in Table 1 is that conflict exposure was generally measured by residence in a country, region, province, or district where conflict was underway for a period of time, with the exception being the Wagner et al. study cited previously and the Kinyoki et al. (2017) study. A particular mother's or child's exposure to conflict and the intensity of that person's exposure, as distinct from any other resident of the same geographic area, generally was not measured. Other limitations of one or more studies include use of aggregate data from a variety of surveys, small sample sizes, and a lack of adjustment for maternal, child, or household characteristics.
Armed conflict is prevalent in Sub-Saharan Africa. Nigeria is among the most conflict-prone countries in the region, experiencing the highest number of conflict-related deaths of all Sub-Saharan  (Salehyan et al., 2012). Most (77.7%) conflicts were armed conflicts. and injured another 12. They also razed some 300 homes, stole motorbikes, cars, livestock and money." (Jackson, 2013). Much of the violence in 2013 was concentrated in the northeast, near Maiduguri, the capital of the Borno state.

| Conceptual framework
The purpose of this study is to add to existing literature by documenting the effect of conflict on child malnutrition in Nigeria by improved empirical methods. Figure 1 provides a conceptual framework for how civil conflict may affect child nutrition. Child malnutrition can result from loss of parental or family support, including the loss of farms or herds. Civil conflict can lead to many other factors that affect a child's nutritional status, such as displacement from food, shelter, and safe water (leading to malnutrition, diarrheal disease, and reduced resistance to other diseases) and disruption in the food supply chain and destruction or disruption of health and social services that may prevent access to life-saving treatments. In addition, the

Key messages
• The suffering of small children is often an unrecognized outgrowth of civil conflict.
• The effect of conflict on child malnutrition in Nigeria has not been studied, even though the country is one of the most conflict-prone in the world.
• We find that infant and young children exposed to conflict in Nigeria are much more likely to suffer from acute malnutrition.
• The effects are more extreme when they live close to the conflict or if the conflict persists over time.
• More should be done to address the nutritional needs of young children in conflict zones.
T A B L E 1 Summary of findings from empirical studies of the effect of conflict on mothers and children

2000-2004
In all countries, the conflict zones experienced higher child mortality and acute malnutrition than nationally.
Used aggregate data from a variety of surveys and populations; not nationally representative.

Kinyoki et al. (2017)
Somalia 2007-2010 Children exposed to recent conflict had a higher risk of wasting (OR = 1.38). Risk of stunting was similar. Children exposed to longer conflicts had higher risk.
No adjustment for intensity of conflict (e.g., number of deaths).

1988-2011
Women in the conflict zone had lower rates of contraception use and institutional delivery but higher rates of skilled delivery.
Conflict exposure measured by living in conflict zone; no measure of conflict intensity.
Verwimp (2012) Burundi 1998-2007 Children exposed to civil war in their area of residence had a 10% increase in the probability of dying.
Small sample size (N = 283, only 75 exposed to civil war).

| Data sources
We used two data sources, linked together by geocode.

| Outcome and control measures
All data on outcomes derive from the DHS survey conducted in 2013.
The DHS is an in-person interview with mothers. It asks her questions about her health and fertility and that of members of her household.
In particular, for this study, she reports the birth dates for each of her children, and children under 5 are measured by the interviewer for height and weight. The survey responses are used to construct the outcome measure (acute malnutrition) for each child under age 5 in the survey, as follows: The z-score for weight for height is the number of standard deviations above or below the norm weight for height for children, as defined by the World Health Organization, multiplied by 100. Acute malnutrition, also called "wasting," is defined as weight for height more than 2 standard deviations below the norm. When a height or weight was missing or out of the expected range (as set by the DHS), the z-score was not available and those children were left out of the analysis. Demographic variables, used as control in the regression models, are defined as follows: • Child characteristics: birth order, year of birth, gender, and number of months since birth of previous sibling (with the overall average used for first order births) • Maternal characteristics: age at each child's birth, marital status at time of interview, length of education, and ethnicity (tribe) • Household characteristics: region, wealth index quintile (constructed from numerous DHS indicators such as housing quality), and household size • Urban/rural residence: According to the Nigeria 2013 DHS, urban residence is defined as living in a large city (over 1 million people), a small city (over 50,000 people), or a town (not defined by population but by the Nigerian government).

| Conflict exposure measures
From the SCAD, we developed two indices of child exposure to conflict, with the assumption that a conflict with a higher number of deaths or that lasted longer would have more serious consequences for the child's nutritional status. Thus, conflict exposure is measured by all conflict-related deaths and by length of exposure during the year preceding the interview. In addition, we assumed that a conflict that was further away would have less consequence for a child's health. The number of conflict deaths is weighted inversely by distance of the conflict to the mother's home at the time of the interview. For example, a death within 1 km of the home has a weight of 1 and a death within 2 km has a weight of.5.
Deaths to conflicts outside a 10-km radius for urban children and a 25-km radius for rural children are given a weight of zero. A child can be exposed to more than one conflict in a given period, and if so, all deaths are included in the exposure measure. Similarly, a measure of length of exposure in days, also weighted by distance in a similar manner to deaths, is used. In this measure, a child does not necessarily experience the entire duration of the conflict; only those days with one or more conflicts within the child's exposure period are counted.

| Data limitations
The DHS has a major limitation for the purposes of this study, in that location data (which are merged with the SCAD) are collected at the time of the survey. However, thousands of people have been displaced from their homes during the Nigerian civil conflicts of the study period (Project Ploughshares, n.d.). Consequently, when a mother is interviewed, she may not reside at or near the place where she gave birth or where a particular child lived during the exposure period. This inserts a substantial measurement bias in the conflict exposure data. It is not clear whether the bias is in a favourable or unfavourable direction. For example, people who are better off (and with better outcomes) may have a better ability to move, but those who are worse off (and with poorer outcomes) may be more motivated to move, especially persecuted ethnic groups. In the first circumstance, the bias would be in the direction of finding an effect of the impact of conflict and vice versa for the second circumstance. This problem with using DHS data for the analysis of conflict effects was noted in the first DHS study of this topic (Savitz, Thang, Swenson, & Stone, 1993) and continues today. A second limitation of these data is that distance to a conflict is measured relative to the geographic centroid of a survey cluster rather than to the precise location of the mother's home (clusters are generally small geographic areas.) This introduces random measurement error in the key independent variable, which tends to bias coefficient estimates towards zero, reducing the likelihood of finding a significant effect.
There are several large databases that tabulate the number and location of conflicts in Africa and other countries, as well as the type and length of the conflict and the number of deaths associated with it. One of the most often used is the ACLED (Raleigh, Linke, Hegre, & Karlsen, 2010). Both the SCAD and the ACLED document conflicts in Nigeria over the study period. However, one major difference between the data sets is the sources for reports of conflicts, locations, and conflict deaths. For the SCAD, these come from only two sources, the Associated Press and Agence France Presse news wires. In contrast, the ACLED uses up to 50 local, regional, national, and continental media sources. Not surprisingly, the ACLED documents about twice as many conflict deaths as the SCAD. We chose the more conservative data source in terms of numbers of conflicts and conflict deaths, which likely underreports small localized conflicts but is likely more consistent over time.
3 | RESULTS Table 2 shows the characteristics of the study group, separately for urban and rural children. The study group is characterized by children in large families (average family size of 6.6 in urban areas and 7.4 in rural areas). In addition, rural residents are much poorer than urban residents. Only 3.0% of urban residents are in the poorest quintile compared with 33.5% of rural residents. Rural residents are concentrated in the northern regions.
Using the DHS Statcompiler (n.d.), we found that unadjusted rates of wasting in 2013 for urban and rural Nigerian children in the sample were very high (data not shown). The child wasting rate was 19.2%, with 19.0% for urban children and 19.3% for rural children. The mean z-score was −0.6 (SD, 1.5). Table 3 shows the average exposure of children in the sample to conflicts and conflict deaths. The table shows both absolute numbers of conflicts and conflict deaths as well as the number of deaths weighted by distance from conflict. Keeping in mind the potential location bias discussed above, the data suggest that Nigerian children were exposed to many conflicts during the study period, which took place during the Boko Haram insurgency. An urban child was exposed to an average of 4.32 conflicts in the year prior to the survey, whereas a rural child was exposed to an average of 1.04 conflicts. The average annual number of conflict deaths was 8.94 deaths (urban children) and 2.43 deaths (rural children). After weighting for distance, those numbers become 1.83 deaths for urban children and only 0.14 deaths for rural children. The average length of conflicts was 11.85 days (urban) versus 3.08 days (rural) during the year before the survey.
After weighting for distance, this conflict measure is 2.78 days (urban) and 0.16 days (rural). models are presented for urban and rural children, as well as separate models for the two conflict measures, conflict deaths and conflict duration. All four regressions show consistently that conflict is positively and significantly associated with child wasting (acute malnutrition). For example, the odds of experiencing wasting are 5% higher for every unit increase in the conflict death index (deaths/km, e.g., one death within 1 km and two deaths within 2 km) in urban areas and 17% higher in rural areas (borderline significance). This is a substantial increase in the odds of wasting for the average child, especially for urban children who are exposed to more conflicts.
It is possible that a lack of access to good nutrition affects all children throughout the weight for height spectrum. For example, children who have an excellent weight for height may lose weight, even though they do not become acutely malnourished (wasted). To examine whether the relationship of conflict to child nutritional status is continuous throughout the weight for height spectrum, we also modelled the continuous z-score. There was no clear cut point, suggesting a gradually declining nutritional status below the norm.
This regression also showed a strong and significant relationship between both the number of conflict deaths weighted by distance and the duration of the conflict weighted by distance. For example, using the conflict death measure, for urban areas, a conflict death within 1 km led to a decrease of.059 SD from the WHO norm of weight for height for the child's age. These regression results can be found in a separate document (Howell, Waidmann, Birdsall, Holla, & Jiang, 2018).
We conducted several sensitivity analyses. First, to understand the sensitivity of the findings to the treatment of distance, we recalculated the conflict metrics using the squared distance rather than the linear distance. Although coefficient values change (as expected given the change in the scale), all significant coefficients shown in the paper remain statistically significant. Similarly, to control for long duration events, we added dummy variables indicating conflicts lasting between 3 and 12 months and conflicts lasting more than 12 months. In both sets of analyses using different distance weighting methodologies, the values of the coefficients on the conflict indicators changed only slightly and remained significant. Other sensitivity analyses are described in the previously cited document (Howell et al., 2018).

| DISCUSSION
This article has shown a strong association between living close to a conflict zone and acute malnutrition (wasting) in Nigerian children in 2013. This is related to the severity of the conflict, as measured both in terms of the number of conflict deaths and the length of time the child was exposed to conflict. For example, urban children have about 5% greater odds of being wasted for each conflict day or death they are exposed to. Findings for rural children are even stronger with odds ratios of 1.13 or 1.17 depending on the conflict measure. Primary prevention would consist of addressing the underlying causes of conflict (e.g., corrupt governments, lack of education and jobs, the flow of small arms into countries, and increasing drought).
Secondary prevention consists of numerous targeted interventions, such as food assistance and "therapeutic feeding" (e.g., using enriched foods) for malnourished children (Maternal and Child Undernutrition Study Group, 2008;Young, Borrel, Holland, & Salama, 2004). Tertiary prevention includes peace-keeping after the conflict to avoid recurrence as well as rehabilitation of education and health facilities. A recent World Bank report suggests several ways that the Bank could improve its assistance to postconflict countries to address children's health, such as institutional support to Ministries of Health and to local governments, more flexible financing, and more explicit efforts to address the health needs of children postconflict (Bustreo, Genovese, Omobono, Axelsson, & Bannon, 2005). International aid and development projects can provide needed assistance but are often not targeted to the most vulnerable populations (Kotsadam et al., 2018).

| CONCLUSION
Undoubtedly, civil conflict is limiting the future prospects of Nigerian children and the country's economic growth. As pointed out in the recent Global Nutrition Report, malnutrition "is a problem of staggering size-large enough to threaten the world's sustainable

CONFLICTS OF INTEREST
The authors declare that they have no conflicts of interest.

CONTRIBUTIONS
EH oversaw the design and implementation of the study and drafted the initial manuscript. TW oversaw the statistical design and data analyses and contributed to the manuscript. NB provided a critical review of the study design and the manuscript. NH and KJ analysed the data and contributed to the manuscript. All authors reviewed and approved the manuscript.