Using Machine Learning to Improve Residential Segregation for Association Analysis

The Index of Concentration at the Extremes (ICE) is a widely used method to measure residential segregation, offering a powerful tool for public health and healthcare researchers to study social determinants of health. Its ability to incorporate race/ethnicity, income, or a combination of both allows for nuanced analyses that can reveal critical patterns of inequality across geographic regions, such as counties and ZIP codes. Unlike other measures, ICE does not require complex assumptions about comparison groups, making it both flexible and accessible for a variety of research contexts. Numerous studies have demonstrated a significant relationship between ICE-measured residential segregation and health outcomes. However, ICE has a notable limitation: it focuses solely on the extremes, capturing information only from the highest and lowest percentiles of a distribution. This narrow focus overlooks the middle of the distribution, which likely contains valuable insights for comprehensively understanding residential segregation within a neighborhood or region. To address this limitation, we propose a new methodology that utilizes machine learning (ML) techniques. Our approach organizes ICE values into distinct groups, ranging from the most deprived to the most privileged. By applying this methodology, we can more effectively represent both the two extreme groups and the middle of the distribution, offering a more comprehensive understanding of the relationship between residential segregation and health outcomes.