Application of Data Science in Health Economics

Jugal Marfatia

doi:10.7273/000002470

This dissertation consists of three papers which explore data science techniques to address issues in the field of health economics. The first paper, provides a novel investigation into the effectiveness of new and scalable machine learning models, LightGBM and XGBoost, to predict individual health care expenditures. In this paper, I find LightGBM and XGBoost outperform linear models, in terms of out-of-sample R2, by 5% to 8%. The paper also explores welfare implications in the insurance market resulting from widespread use of big data and scalable machine learning models by insurance providers. In paper 2, I study the heterogeneous effects of having a physical limitation on mental health across individuals in different income brackets. Results indicate that the negative effect of having a physical limitation on metal health is highest among individuals in the bottom 20% of the income distribution. Results from panel data analysis indicate that an increase in income has a greater positive impact on mental health among individuals with a physical limitation compared to the effect observed among individuals without a physical limitation. In paper 3, I estimate the causal effects of being enrolled in medicaid on mental and financial well being across individuals with and without pre-existing conditions such as asthma, diabetes, heart condition, high cholesterol and cancer. In other words, I compare the bene- fits observed among individuals with pre-existing to the benefits observed among individuals without pre-existing condition using data from Oregon Medicaid Lottery. My results indicate that on average, across different measures of mental and financial health, benefits from medicaid observed among individuals with pre-existing conditions were lower or similar to the benefits observed among individuals without pre-existing conditions. On two key measures, undiagnosed depression and catastrophic health expenditures, the benefit from enrolling in medicaid was lower, statistically significant at the 90% confidence level, among individuals with a pre-existing condition compared to the benefits observed among individuals without a pre-existing condition.

Application of Data Science in Health Economics

Files and links (1)

Abstract

Metrics

Details