Three Essays on Government Health Expenditure

Abstract

In 2010, governments spent over $4.05 trillion on health. This is 7% of the world economy. This dissertation measures the health outcomes achieved by this spending, how this spending changes upon the receipt of health aid, and the methods used to analyze this spending. The first chapter assesses the effect of government health expenditure on health outcomes. I use a large cross-country panel and a new instrument to test how domestic government health expenditure affects under-five mortality. I find that the average spending-to-outcome elasticity is -0.34, although it ranges between -0.04 and -0.61 depending on country-level characteristics. Countries with larger GDP per capita and more civil liberties, political rights, and democracy have the elasticities furthest from zero, although countries with the largest under-five mortality rates have the largest effects when measured in deaths averted. The second chapter assesses the crowding out of domestic government health expenditure upon the receipt of development assistance for health channeled to the government. Using a cross-country dynamic panel, I use Difference GMM to simultaneously measure the displacement and replacement rates. I find that increases in development assistance lead to the displacement of government resources, but the reduction in development assistance does not lead to the replacement of resources. Combined with the estimates from the first chapter, these findings suggest that net effect of health aid is less than previously measured. The third chapter is an assessment three linear regression estimation methods used to control for unobserved heterogeneity that exists in most clustered data. This type of heterogeneity exists in most analyses of government health expenditure. This chapter explains the standard clustered data model, the traditional random- and fixed-effects estimators, and the `within-between' estimator - a variant of a model proposed by Mundlak in 1978. Simulation is used to illustrate when each estimator is optimal. The chapter ends with recommendations for when each estimator is preferred.