Methods/Statistics
Much ado about nothing? Do we need harmonized census tract data to study longitudinal associations between neighborhood conditions and the health of individuals? Tongtan Chantarat* Tongtan Chantarat David C. Van Riper Jenny L. Wagner Michelle J. Ko
Epidemiologists use population data from the Decennial Census to describe various neighborhood conditions and their associations with health. The Census Bureau redraws census tract boundaries before each Decennial Census to reflect neighborhood and population changes, making it challenging to characterize neighborhood conditions consistently across time. To account for the potential “noise” introduced by measuring conditions using changing neighborhood boundaries, epidemiologists can use data from different censuses harmonized onto a single set of census tract boundaries. However, boundary harmonization is a resource-intensive, time-consuming process. We used a case study of exposure to residential segregation in mid-life and later-life cognitive decline to examine differences in model estimates using census year-specific vs. harmonized boundaries. Using survey data from Black participants in the Health and Retirement Study who lived in urban areas at least once between 1990-2010, we characterized exposure to county-level residential segregation by calculating indices of dissimilarity and isolation in two ways, using: (1) population estimates based on the 1990, 2000, and 2010 boundaries, and (2) estimates based on 2010-harmonized boundaries from the National Historical Geographic Information System, the Longitudinal Tract Database, and the Neighborhood Change Database. Using linear mixed-effects regressions, we found no significant differences in the model estimates when using census year-specific vs. harmonized data. Boundary harmonization may not be needed to accurately capture the relationship between life-course exposure to neighborhood conditions and health outcomes in later life, opening up opportunities for epidemiologists to characterize the exposure from much earlier in life.