“Is Health Insurance Always the Best Choice? A Machine Learning vs. Traditional Models Approach to Understanding Coverage Determinants” (link coming soon)
Abstract: This study examines the determinants of health insurance uptake using MEPS data for individuals aged 18 to 64 who make their own insurance decisions, excluding those with employer-sponsored plans and Medicare. Using logistic regression, random forest, and XGBoost with data balancing (SMOTE) and feature engineering (Gram-Schmidt transformation), the analysis finds that XGBoost with data balancing and feature engineering outperforms all models, achieving the highest precision, recall, and F1 scores. Shapley value analysis highlights income per person in family, healthcare utilization, and demographic factors as key predictors, with recent doctor visits as the strongest determinant. Findings suggest that targeted financial aid, preventive care incentives, and improved accessibility for non-English speakers could enhance insurance enrollment and reduce coverage disparities.
“Why do Immigrants Have a Lower Flu Vaccination Rate than Americans?" (link coming soon)
Abstract: Vaccination rates among immigrants in the U.S. are lower than those of native-born individuals, raising important public health concerns. This study examines whether this gap stems from differences in healthcare access or cultural influences from immigrants' countries of origin. Using data from the National Health Interview Survey (NHIS) and Blinder-Oaxaca decomposition, I find that health insurance coverage explains approximately 70% of the vaccination gap, underscoring access barriers as a primary driver. Incorporating the Vaccine Confidence Index (VCI) reveals that while home-country beliefs about vaccine efficacy impact immigrant vaccination behavior, access remains the dominant factor. These findings suggest that policies enhancing healthcare access and addressing cultural beliefs could effectively improve vaccination rates among immigrant populations, benefiting individual and public health.
“The Measurement of Labor Market Dynamics with Misclassification” (with Daniel Millimet)
Abstract: Labor market dynamics are a key indicator for researchers and policymakers, but they are often affected by misclassification issues. To improve accuracy, we extend recent work on partial identification of transition matrices, deriving bounds for economic flows under minimal assumptions about misclassification and mobility processes. Applying this method to the Survey of Income and Program Participation (SIPP) data, we confirm that misclassification has a significant impact on measuring labor market dynamics, and our approach substantially narrows the range of possible true values, providing a more reliable estimate.