Investigating the Causal Effect of Deforestation on Infant Health Through Soil Characteristics: A Comparison of Traditional and Machine Learning Mediation Analysis Using Simulated and Real Data

Deforestation is a global threat that impacts the environment and biodiversity in several ways. Recently, some scholars started investigating how deforestation affects human health. This paper aims to assess the causal effect of deforestation on infant health in Nigeria, evaluating the mediating role of soil characteristics. We analyse data relative to thousands of Nigerian children collected in 2008 and 2013 using two mediation analysis approaches: a traditional, regression-based approach and a machine learning one. Given the novelty of the latter, the first part of the paper is devoted to illustrating the two approaches and comparing their performances in terms of estimates’ bias and coverage rates through a simulation study. The second part of the work focuses on the analysis of deforestation data: in particular, we analysed the effect of deforestation on Nigerian children’s probability of suffering from cough, diarrhoea and malaria, through the pH, the organic carbon and cation levels of soil. The results show mixed evidence of the effect of deforestation on infant health, either because the results differ across the two approaches mainly in terms of significance, and because the significant ones show different signs. This leaves room for further analyses on this topic. Supplementary materials accompanying this paper appear online.