Assessing the relationship between national poverty rates and health outcomes

Samuel Preston’s seminal work enumerated the relationship between national income, innovation and national life expectancy. Since then, many other researchers have broadened this evidence base, and it is now firmly established that mean income explains a meaningful portion of the variation in many population health outcomes.  Furthermore, these associations have also been shown to exist at the individual level, with a known gradient connecting income and health. One shortcoming of these analyses, however, is that at the population level, only mean income is routinely assessed as a determinant of health. The relationship between other likely population health predictors such equity and poverty have not been widely assessed in cross-national studies due to lack of comprehensive and comparable data. Furthermore, it is not clear how the effect of poverty on health changes across various thresholds of poverty. In this study, we construct 51 complete poverty series measured at unique income thresholds and test which threshold is most associated with gains in health.

National poverty estimates are reported in the World Bank’s POVCAL database.  We systematically consider 2,578 potential predictive covariates, all publically available data from the World Bank, Institute for Health Metrics and Evaluation, Freedom House, and the Quality of Government databases. A variable selection process based on linear regression and Bayesian Model Selection is used to derive a tractable set of predictors.  To estimate the poverty rate at 51 different thresholds, we employ twenty variants of three models, including multiple imputation, hierarchical mixed effects models, and Gaussian Process Regression. The best model is selected using out-of-sample validation. Finally, we use fixed effects regression to test if national poverty rates are associated with changes in adult and child mortality.

Our analysis shows that poverty is significantly associated with health outcomes, above and beyond the effect of gross domestic product.  More importantly, our analysis shows that the threshold at which the national poverty rate is defined is critical.  Escaping extreme poverty, as currently defined as living on or below $1.25 per day, is not sufficient to drastically improve health.  On the other hand, when poverty is redefined as living on $5 per day, moving out of poverty is a significant predictor of improved health outcomes. 

The World Bank has stated that one of its two operating goals is to “end extreme poverty by 2030.” Over the last several decades, there has been a great deal of progress made globally on reducing the number and share of people living in extreme poverty. When a higher income threshold is used to define poverty, poverty and health are closely related and gains in poverty alleviation are significantly associated with improved health outcomes.  Increasing the threshold at which poverty is measured makes achieving the World Bank’s goal much harder, but is a truer reflection of the income needed to improve population health.