2.2. Spatial Racial/Ethnic Segregation Measures

Given the spatial character of our outcome, we chose to use two measures of spatial (instead of aspatial) segregation recommended by Reardon and O’Sullivan [24] as our focal predictors. Specifically, the current study uses the Spatial Information Theory index (H) for spatial evenness, and the Spatial Isolation Index (P*) for spatial exposure. The key difference between spatial and aspatial segregation measures is whether an indicator considers the spatial arrangement of population. Most of the conventional segregation measures, such as the dissimilarity index, are aspatial [24], which mismatches with our spatial poor health clustering measure. We follow the suggestions of existing research to concentrate on the evenness and exposure dimension of segregation [8,17,47]. H can be understood as a measure of how high residential segregation is between two groups, 1 indicating maximum segregation and 0 representing complete integration. While P* generates both spatial exposure and spatial isolation components, we focus on the latter in light of its well-documented effect on health [7,27]. P* can be interpreted as the probability of two randomly selected individuals being racial/ethnic minorities. For simplicity when discussing P*, we reference the Spatial Isolation Index in place of ‘exposure.’

More importantly, the following features make H and P* outperform other commonly used spatial segregation measures [24]. First, both H and P* can be decomposed with the change in the boundaries of subareas and the decomposed values are additive. Second, H and P* can be applied to both aggregated population counts (zone-based) or continuous population density (surface-based). The latter helps to minimize the well-known modifiable area unit problem [91,92]. For a fuller visualization of the spatial character of isolation and evenness, we recommend Iceland et al. [93].

Data for spatial residential segregation were collected from the 2010−2014 American Community Survey (ACS). The measures were calculated with Census tracts with the ‘seg’ package in R for each of the 498 cities separately [92] (as a reminder, Honolulu and Las Cruces, New Mexico were excluded from the analyses). While the option existed to conduct a multigroup measure, we chose to conduct our analysis on just two groups at a time, Blacks and Whites, Hispanics and Whites, and Asians and Whites for two reasons: First, two group measures are more readily interpretable than multi-group measures. Second, the two group measure places primacy on segregation from Whites as all the measures directly compare Whites from non-Whites. This is important given how much of an effect the separation from Whites is thought to have on non-White health disparities [17].