Global Multidimensional Poverty Prediction using World Development Indicators

Effective implementation, monitoring ,and evaluation of targeted poverty reduction programs require accurate measurements of poverty levels and their changes overtime. The Multidimensional Poverty Index(MPI) offers a more comprehensive measure compared to traditional income-based assessments. However, for many countries, MPI data are either unavailable or limited to a few years due to the high cost of conducting relevant surveys. This paper presents alternative methodologies to predict the Global MPI across different countries and time periods using the World Bank’s World Development Indicators as predictor variables. Given that MPI construction involves proportions bounded within the unit interval, we tailor statistical learning methods accordingly. In a high-dimensional context, where the number of predictors exceeds the number of training observations, we evaluate methodologies such as dimension reduction, regularized models, and ensemble learning. We conduct cross-validation experiments to assess model performance, incorporating both measured and non-measured countries in the testing dataset.