Census tracts aren’t neighborhoods: Issues in examining the impact of historical redlining in the United States on present-day environmental exposures and health outcomes

Abstract

BACKGROUND AND AIM: In the 1930s, the US Home Owners’ Loan Corporation (HOLC) developed residential security maps for 239 urban areas across the US, ranking the investment risk using a color-coded scale: "A”(green), "B”(blue), "C”(yellow), and "D”(red/hazardous). Areas in proximity to industrial facilities and other sources of pollution were downgraded and more likely to have Black and foreign-born inhabitants—resulting in disproportionate exposure of racial/ethnic minorities to environmental toxicants. HOLC-derived neighborhood boundaries do not align with Census tracts, making the examination of historical redlining on present-day environmental exposures challenging. Multiple methods exist to define redlining when using Census data, but how well they classify redlining exposure is unknown. METHODS: We examined agreement in redlining status when assigning exposure based on area-level geographic data compared to having complete address data (gold-standard). Our analysis included 5,042 addresses located in census tracts that overlapped with the HOLC map for Atlanta. We classified redlining status using complete address data and census tract using four methods: the Centroid, Majority, Weighted Score, and Highest HOLC methods. Each method was compared to the HOLC grades assigned when superimposing complete address data onto the HOLC map. RESULTS: Of the 5,042 addresses, 3% were A-graded areas, 9% B, 24% C, 12% D, and 52% ungraded. Overall percent agreement was highest for the Weighted Score method, which correctly classified 70% of addresses. The Majority, Centroid, and Highest HOLC methods had 68%, 65%, and 53% agreement, respectively. However, these methods only correctly identified 40%, 49%, and 57% of HOLC graded areas (excluding ungraded) compared to 76% using the Weighted Score method. CONCLUSIONS: Full address data are often unavailable in population-based studies examining environmental exposures on health outcomes, thus relying on census tract to classify redlining exposure. Investigators should carefully consider the method used to minimize misclassification. KEYWORDS: Redlining, exposure misclassification, validation study