Unveiling the Causal Mechanisms Within Multidimensional Poverty.

Area of Science:

• International Development
• Causal Inference
• Econometrics

Background:

• Sustainable Development Goals (SDGs) highlight the need for evaluating complex interventions.
• Current evaluation methods struggle with multiple, interrelated development outcomes.
• Assessing the impact of interventions on interconnected factors like poverty and violence remains challenging.

Purpose of the Study:

• To propose a novel methodological framework for complex causal inference in international development.
• To combine machine learning and econometric designs for evaluating multifaceted development interventions.
• To analyze the relationship between multidimensional poverty and violence in Colombia.

Main Methods:

• Utilized Bayesian networks (BN) to construct a directed acyclic graph (DAG) modeling poverty-poverty and poverty-violence interrelationships.
• Employed the DAG output to identify instrumental variables (IV) for causal analysis.
• Applied 2SLS (Two-Stage Least Squares) regression to test the causal effect of poverty on violence victimization.

Main Results:

• Minimum living standards (water, sewage, housing quality) significantly predict education and health dimensions of poverty.
• A 0.4% increase in household violence victimization likelihood was associated with having an illiterate member.
• Bayesian networks effectively predicted complex causal patterns in multidimensional poverty.

Conclusions:

• The proposed framework integrates machine learning (BN) and econometrics (IV, 2SLS) for robust causal inference in development studies.
• This approach can enhance the evaluation of interventions targeting complex, interrelated outcomes like poverty and violence.
• Findings underscore the link between poverty dimensions and vulnerability to violence, informing policy and intervention design.