Projects

Master’s Thesis

“Investigating Determinants of Birth Weight Using Bayesian Tree-Based Nonparametric Modeling”

Advisor: Dr. P. Richard Hahn (Arizona State University)

Defense Date: May 22nd, 2025

This work develops DM-CART (Dirichlet-Multinomial Classification and Regression Trees), a Bayesian nonparametric modeling framework for analyzing low birth-weight (LBW) risk using U.S. Natality data. The method extends CART by integrating a Dirichlet-Multinomial likelihood for categorical birth weight outcomes, informed by prior quantiles from the 2020 dataset. A two-tier bootstrap procedure yields stable ensemble predictions, allowing precise subgroup-level risk estimates.

Key risk determinants include maternal race, smoking status, and marital status. When focusing only on LBW cases, infant sex and maternal age emerge as primary discriminators. The approach delivers interpretable trees with robust uncertainty quantification, supporting targeted intervention and health equity analysis in maternal-infant care.

Learn More

• 📖 Thesis Publication
• 🔬 GitHub Repo
• 📑 Full Thesis PDF

Papers

– Zheng, Y., Reiser, M., & Kurth, A. (in preparation), A Monte Carlo Comparison of the Efficacy of Mplus, flexMIRT, PROC IRT, ltm, and mirt in IRT Model Estimation

This study assesses the relative estimation accuracy of leading IRT software packages, informing best practices in psychometric modeling.