Unveiling the Causal Mechanisms within Multidimensional Poverty

Abstract This paper combines machine learning and econometrics to explore the relationship between multidimensional poverty components and violence in Colombia. First, I create a directed acyclic graph (DAG) using Bayesian networks and census data to predict how multidimensional poverty components are interrelated. I find that minimum living standards— measured in terms of access to water, connection to the sewage system, and the quality of walls and floors—are strong predictors of the education and health dimensions of poverty. Second, I use the DAG output to identify potential instrumental variables (IV) that may be used to test the effect of multidimensional poverty on a household’s likelihood to be a victim of violence. Illiteracy is predicted by a complex set of poverty indicators in the DAG, yet it seems to be the only connecting point between multidimensional poverty and violence showing potential to meet the validity assumptions in an IV approach. Using 2SLS, I show that having an illiterate person within a household increases by 0.4% the household’s likelihood to be a victim of violence.