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

When people skip HIV tests, the resulting gaps in data can make infection rates look lower than they really are. This project builds statistical tools to correct for that kind of hidden bias. The problem is that health surveys often suffer from missing data—people opt out, refuse tests, or drop out of studies. If those missing people differ systematically from those who stay, the final numbers are skewed. This is especially acute in HIV prevalence studies, where low participation can distort national estimates and mislead policymakers. Current statistical methods exist to correct for this, but they rely on identifying "exclusion restriction variables"—factors that affect participation but not the outcome itself—which is notoriously difficult. The researchers aim to develop stable, reliable techniques that do not require perfect knowledge of those variables. They will determine the sample sizes and effect sizes needed for robust corrections, handle missing data in the covariates themselves, and build accessible software so that non-specialists can use the methods. If successful, the work will give epidemiologists and public health officials more accurate estimates of disease burden, particularly for HIV, and could be applied to other health surveys where non-ignorable missing data threatens the validity of conclusions.