Armed Conflict and Early Childhood Development in 12 Low- and Middle-Income Countries

Using geospatial data on armed conflicts and development outcome measures of young children linked to global positioning system data, we evaluated the association between exposure to nearby conflict and the probability of a child being developmentally on track.

We obtained data on ECD and individual-, cluster-, and country-level characteristics from Demographic and Health Surveys (DHS), nationally representative population-based surveys routinely conducted in many low- and middle-income countries.²⁰  DHS contains information on the individual and his or her household, including an estimate of the living standard of each household called the “wealth index,” which is derived by using principal component analysis of data on a household's ownership of assets.²¹  Georeferenced surveys contain the latitude and longitude of each participant’s cluster (analogous to a city block in urban areas and a village in rural areas), identified using global positioning system devices.²²  We used all georeferenced surveys for which data on ECD were available (13 surveys collected between 2011 and 2018). Data on conflicts that took place between 2006 and 2018 were used in the analyses to track the association with ECD up to 5 years after exposure to conflict. The years and countries of the surveys used as well as the number of participants in each survey who met the inclusion criteria are available in Supplemental Table 6.

We acquired data on the time and location of conflicts from Uppsala Conflict Data Program (UCDP) Conflict Encyclopedia, a publicly available online database that provides high-quality data on conflicts.²³  The UCDP produces comparable data on conflicts from all parts of the world and is updated annually. The geographical and temporal data are acquired from multiple sources after UCDP systematically evaluates each source, adhering strictly to the definition of armed conflict: “incidence of the use of armed force by an organized actor against another organized actor, or against civilians, resulting in at least 1 direct death in either the best, low or high estimate categories at a specific location and for a specific temporal duration.”²⁴ 

Children 36 to 59 months of age who live with their mothers with ECD outcomes and geographic data available in the DHS were included in the analyses.

We used the Early Childhood Development Index (ECDI) as a measure of whether a child is developmentally on track.²⁵  The ECDI is a measure of ECD with the highest coverage developed by United Nations International Children’s Fund,^26,27  incorporated in widely-conducted surveys, including the DHS and Multiple Indicator Cluster Survey (MICS).^20,27  The ECDI uses a total of 10 questions on a child’s skills and behavior covering 4 domains of development, social-emotional, physical, approaches to learning, and language and cognitive domains, to measure a child’s development.²⁵  A child who is developmentally on track in at least 3 of 4 domains is considered developmentally on track.²⁵  The details of individual domains are shown in Supplemental Information 2. The psychometric properties of ECDI have been validated²⁵  using widely accepted indices of ECD, such as the Early Development Instrument²⁸  and Strengths and Difficulties Questionnaire.²⁹  The ECDI has also been shown to have external reliability²⁵  and has been used in multiple international studies on child development.^30–32 

We evaluated the association between exposure to conflicts and the probability of a child being developmentally on track after inverse probability of treatment weighting with the R package causalweight.³³  Inverse probability weights can be represented as W^Conflict = 1/f[Conflict|L], in which L represents the covariates and Conflict represents a binary indicator denoting whether the child was exposed to an conflict within 50 km in the past 1, 2, 3, 4, or 5 consecutive years. Weights were constructed from logistic regression models of exposure to conflicts as a function of the covariates: f[Conflict|L].^34,35  After weighting observations, the association between exposure to conflicts and the probability of a child being developmentally on track was calculated by taking the mean difference of the outcomes with and without exposure to conflicts.³⁵  The computed standard errors were based on independent and identically distributed bootstrap, a valid method for estimating treatment effects based on inverse probability of treatment weighting.³⁶  Missing data were imputed by using multiple imputations by chained equations (MICEs) by predictive mean matching. All analyses were conducted by using R statistical software, version 3.6.3, and the statistical code is available on request.

We first estimated the association between exposure to conflict and the probability of a child being developmentally on track after weighting with individual- and cluster-level covariates, which included the child’s age, sex, birth year, mother’s level of education, wealth index, presence of an older sibling, presence of the child’s father (or the mother’s partner), and whether the cluster is urban or rural, each of which has been shown to be associated with being developmentally on track.^{10,25,30,31,37}  Additionally, we adjusted for the birth weight of the child because low birth weight is associated with delayed social and behavioral development in young children.³⁸  We also included country-year fixed effects among the covariates, reflecting the political or economic situation in the country at the time of the survey.

We evaluated the associations between exposure to conflict and individual domains of ECD (socioemotional development, physical development, approaches to learning, and language and cognitive development). We superimposed the associations of the 4 domains to compare the associations. The raw results of the individual domains are shown in Supplemental Table 7.

We conducted mediation analyses using the variables listed in Supplemental Information 4.

As a sensitivity analysis, we estimated the associations between consecutive exposure to conflicts from 51 to 100 km and ECD to determine the significance of proximity to conflicts (Supplemental Table 9). We then estimated the associations after accounting for possible migration (Supplemental Table 10). Additionally, considering that our estimates could be biased because of highly influential weights, we estimated the associations after discarding observations with extreme probability weights (<0.01 or >0.99; Supplemental Table 11). Finally, to address the possibility that MICEs may have changed the results, we conducted the main analyses by using samples imputed by random forests and k-nearest neighbors (Supplemental Table 12).

All data used in this study were obtained from deidentified databases that were publicly available (UCDP) or available from ICF (which implements the DHS) on request, and no ethical approval was needed.

In our study sample, 16 311 (59.2%) children were developmentally on track, and 11 227 (40.8%) were not on track. Other characteristics of the participants are shown in Table 1. Adjusted for individual-level and cluster-level characteristics, exposure to conflict was associated with a 5.9% decrease (95% confidence interval −7.5% to −4.3%) in the probability of a child being developmentally on track from the first year of exposure (Table 2; Fig 1). The degree of delay appeared to be attenuated in the second consecutive year but became progressively greater after that, with 5 consecutive years of exposure associated with an estimated 10.4% decrease in the probability of a child being developmentally on track (95% confidence interval −13.7% to −7.2%). Of individual domains of ECD, socioemotional development had the most negative association and showed a similar trend as overall ECD with chronic exposure (Fig 2; Supplemental Table 7). This trend was not evident in individual domains other than the socioemotional domain (Fig 2; Supplemental Table 7).

In the mediation analyses, the total indirect effects were significant until the second consecutive year of exposure, with stunting, lack of early childhood education, inadequate child care, and mother’s experience of abuse having significant mediatory effects (Table 3). In the third, fourth, and fifth consecutive years of exposure, the total effects of conflict exposure on ECD were mainly through direct effects (Table 4-5), but lack of early childhood education exhibited significant mediatory effects even in the fifth year of exposure (−2.1% change in the probability of a child being developmentally on track; 95% confidence interval −3.1% to −1.1%). Mediation analyses with mother’s mental and physical health status did not exhibit significant mediatory effects (Supplemental Table 8).

The association between exposure to conflict from 51 to 100 km and ECD was small (Supplemental Table 9). Similar trends as Table 2 and Fig 1 were evident in the associations of exposure to conflict and ECD of participants in the same place of residence for ≥5 years (Supplemental Table 10), after trimming extreme weights (Supplemental Table 11), and with samples imputed by using alternative imputation methods (Supplemental Table 12).

This study demonstrated that exposure to nearby armed conflict is significantly associated with an increased probability of delayed ECD, especially with chronic exposure. Similar results were obtained after adjusting for possible migration as well as after trimming extreme weights but not when conflicts were farther away, supporting the robustness of our findings. The results of the mediation analyses showed that unavailability of early childhood education programs may be an important mediator of the association between exposure to conflict and delayed ECD. These new findings add to the body of knowledge on ECD.

In accordance with our hypothesis, we found that the association between exposure to conflicts and the socioemotional domain was the largest among the 4 domains and that the association became more negative with chronic exposure. In a longitudinal study on the effects of cumulative violence exposure on negative adverse mental health and behavioral outcomes, lower levels of chronicity and intensity of cumulative violence exposure appeared to have limited effect on the number of negative mental health outcomes, but the number of negative mental health outcomes increased as the cumulative exposure to violence reached higher levels,³⁹  a finding similar to the current study. Our results could be explained by a combination of 2 forms of adjustment to an adversity: minimal-impact resilience and emergent resilience.⁴⁰  Minimal-impact resilience is characterized by transient distress during or immediately after an adversity followed by a relatively rapid adjustment⁴⁰  and could explain the initial delay in socioemotional development in the first year of exposure to conflict followed by a transient recovery in the second year. As exposure becomes more chronic, adversities cause more enduring patterns of variability in psychological function, which is referred to as emergent resilience.⁴⁰  In emergent resilience, the distinction between resilience and maladjustment is not fully evident until later in the course of exposure to adversity. This could be the reason the negative consequences of conflicts on socioemotional development become increasingly pronounced with chronic exposure. Although further research is needed if these models are applicable in the context of early socioemotional development and ECD, they aid us in better interpreting our results.

Our study should be interpreted in light of several limitations. The variables considered in the mediation analyses were extracted from cross-sectional data in which participants recalled past experiences, and, thus, it is difficult to precisely establish the temporal precedence of the mediators to the outcome. Therefore, the mediation analyses should only be interpreted as proxies of actual mediatory effects. More robust mediation analyses would be possible with time-series data, which may yield stronger clinical and policy implications. Furthermore, because of the small sample size, our analyses with the mother’s physical and mental health as mediators were likely underpowered, and we were only able to estimate the effects for up to 3 consecutive years of exposure to nearby conflicts.

As with other studies in humanitarian settings, migration can have both overestimated (because migration itself can be a risk for a child’s mental health and, thereby, socioemotional development and ECD)⁴¹  and underestimated (because a child may have avoided the risk of being exposed to a conflict by moving away from it) our results. We accounted for this possibility in Supplemental Table 10 and showed that the trend revealed in Table 2 and Fig 1 was still observed.

Although the ECDI is a valid tool, the evaluation of a child’s ECD status is based primarily on reports by a parent.²⁵  A more precise evaluation would have been possible if a delay in ECD were directly confirmed by a clinician, and further studies evaluating targeted interventions to promote ECD of children in areas affected by conflict may benefit from a more thorough evaluation of ECD. However, in this study, clinical assessment of ECD would have made multinational data collection and analysis, a major strength of this study, impractical. Furthermore, tools based on parent-reported answers are widely used in child mental health research.²⁹ 

Notwithstanding these limitations, our study provides novel insight into child development in humanitarian settings and has multiple clinical and policy implications. Most importantly, conflicts should be avoided at all costs, but, if exposure is unavoidable, humanitarian aid to promote ECD should be implemented from an early stage, especially when children are chronically exposed to conflicts. Such aid could greatly improve outcomes, because in a previous study, researchers suggested that delayed development in young children may be reversible: children who were originally institutionalized but later moved to foster care showed improved cognitive development as early as 3 to 4 years old compared with children who remained institutionalized.⁶  Our study suggests that providing early childhood education, which has been shown in previous studies to be effective in preventing developmental delay,⁴²  may be a promising form of humanitarian aid, although this should be confirmed in future studies. Furthermore, given that the association between exposure to conflict and probability of a child being developmentally on track appeared to stem mainly from the socioemotional domain, humanitarian support should focus on providing psychosocial aid. Aiding young children’s socioemotional development can have significant impact later in their lives, from childhood to adolescence and adulthood: preschoolers’ socioemotional learning has been shown to predict early school success,⁴³  and greater socioemotional competence leads to better health, education, and social outcomes.^13,14 

Future studies should focus on identifying protective factors of ECD, especially the socioemotional domain, and whether psychosocial support interventions and early childhood education programs can help build these protective factors in humanitarian settings. Analyses should be performed on time-series data so that the effects of interventions can be assessed and mediators can be explored extensively. Currently, the best available measure of ECD used in multinational, population-based surveys like MICS²⁷  or DHS²⁰  is the ECDI, which consists of only 10 questions. Researchers can benefit from more extensive evaluations of ECD through questions on specific components of development (eg, expression of emotion, empathy, and social understanding for socioemotional development) or through direct evaluation by clinicians. Analyzing such surveys with geospatial methods would prove especially effective not only in regions affected by conflicts but also in natural disasters and epidemics (including the ongoing coronavirus disease 2019 pandemic) because the needs of vulnerable populations can be assessed remotely. Global health agencies should routinely collect and analyze such data, which will streamline the implementation and evaluation of humanitarian interventions.

Exposure to nearby conflicts is associated with an increased probability of delayed ECD in children 36 to 59 months old, especially if the exposure is chronic. Children in affected areas should be provided psychosocial support and early childhood education from an early stage.

We thank Anke Lux (Institute of Biometry and Medical Informatics, Otto-von-Guericke-University, Germany) and Stian Lydersen (Regional Centre for Children and Youth Mental Health and Child Welfare - Central Norway, IPH, Norwegian University of Science and Technology, Norway) for their input in the statistical analyses and interpretation. Additionally, we thank the TroNa partnership for making the collaboration between the authors possible.

DHS

Demographic and Health Survey

ECD

early childhood development

ECDI

Early Childhood Development Index

MICE

multiple imputations by chained equation

MICS

Multiple Indicator Cluster Survey

UCDP

Uppsala Conflict Data Program