RESSOLVE-HD: Regularized Estimation for Stable Solutions to Overcome Selection Bias in Health Data

In health research, missing data can skew results, especially when data are missing for reasons related to unobserved values (non-ignorable missing data). This bias can occur due to sample selection, where individuals opt out of participation, affecting the representativeness of the sample. For example, in HIV (Human Immunodeficiency Virus) prevalence studies, low participation in testing can distort our understanding of HIV rates in the population. Sample selection models (SSMs) help address this but identifying exclusion restriction variables (ERVs) – factors affecting study participation but not outcomes – is challenging. The aim of this project is to develop novel statistical techniques for overcoming this challenge. We will focus on establishing the necessary sample sizes and effect sizes for robust SSM applications, ensuring reliability for medical researchers and decision-makers. Our approach will involve developing stable variable selection methods for SSMs, handling missing data in covariates, and creating accessible software for wider adoption. Using both simulated and real-world datasets, including those from the Demographic and Health Surveys and nationally representative studies, we aim to provide practical solutions for addressing missing data in health research.