The influence of maternal agency on severe child undernutrition in conflict-ridden Nigeria: Modeling heterogeneous treatment effects with machine learning.

Nigeria is one of the fastest growing African economies, yet struggles with armed conflict, poverty, and morbidity. An area of high concern is how this situation affects vulnerable families and their children. A key pathway in improving the situation for children in times of conflict is to reinforce maternal agency, for instance, through education. However, the state of the art of research lacks a clear understanding of how many years of education is needed before children benefit. Due to mother's differing social context and ability, the effect of maternal education varies. We study the heterogeneous treatment effects of maternal agency, here operationalized as length of education, on severe child undernutrition in the context of armed conflict. We deploy a repeated cross-sectional study design, using the Nigeria 2008 and 2013 Demographic and Health Survey (DHS). The sample covers 25,917 children and their respective mothers. A key methodological challenge is to estimate this heterogeneity inductively. The causal inference literature proposes a machine learning approach, Bayesian Additive Regression Trees (BART), as a promising avenue to overcome this challenge. Based on BART-estimation of the Conditional Average Treatment Effect (CATE) this study confirms earlier findings in that maternal education decreases severe child undernutrition, but only when mothers acquire an education that lasts more than the country's compulsory 9 years; that is 10 years of education and higher. This protective effect remains even during the exposure of armed conflict.

Déclaration de conflit d'intérêts

The authors have declared that no competing interests exist.