摘要：Objectives. We investigated whether within-county racial segregation was associated with increased odds of violent injury beyond individual risk. Methods. In a cross-sectional study, data on 75 310 patients admitted with an injury to Pennsylvania hospitals from 1997 to 1999 were analyzed to determine the association between county-level racial segregation and violent injury. We used multilevel analysis to adjust for individual- and county-level factors. Principal components analysis allowed us to separate the effect of segregation from other county-level variables. Results. After adjustment, greater segregation was associated with increased odds of violent injury among Whites (odds ratio [OR] = 1.20; 95% confidence interval [CI] = 1.11, 1.30) and non-Whites (OR = 1.45; 95% CI = 1.28, 1.64). The association was stronger for non-Whites. Conclusions. Our results suggested that living in a county with high levels of racial segregation was associated with increased odds of violence not explained by an individual's own risk. These findings represent an important step in understanding the nature of observed links between race and violence. Future work should develop prevention strategies that simultaneously target community and individual risks. Research has found that racial segregation is associated with a variety of public health outcomes, including violence. 1 – 9 Although recent data suggest that racial segregation continues to decline, many communities remain highly segregated. In 2000, 25% of the metropolitan statistical areas in the United States were hypersegregated (dissimilarity index [a measure of segregation in which a higher number indicates a greater degree of segregation 10 ] > 0.60), and another 50% were partially segregated (dissimilarity index = 0.40–0.60). 11 This high level of exposure and the variety of associated public health outcomes highlight the importance of understanding the association between racial segregation and individual risk. However, much previous research concentrated either on macrolevel relationships between community segregation and aggregate violence rates 1 – 6 or on individual-level analyses. 7 – 9 These single-level approaches cannot identify how much of the association between segregation and violence represents an independent contextual effect on top of simple population compositional effects. Previous work had 2 major limitations First, no studies adequately looked at the association of segregation and violence by incorporating the individual- and community-level effects of race on individual risk. To address this association, we used generalized estimating equations to adjust for the correlation structure inherent in multilevel data, which allowed us to tease apart the independent associations of individual- and community-level race. Second, the few existing multilevel studies employed racial compositional measures (i.e., percentage of individuals of a specific race in a community) as indicators of segregation. 12 – 14 Such measures have been shown to be highly correlated with other neighborhood-level risk factors. 15 Therefore, it is not known how much of an observed association between compositional measures of racial segregation and violence is attributable directly to racial segregation or to other factors such as socioeconomic status. To isolate the effects of racial segregation, we used principal components analysis to calculate an independent and separate factor for racial segregation. Principal components analysis reduces the number of correlated variables into a smaller number of uncorrelated variables. Previous work has shown that high levels of racial segregation within Pennsylvania counties were associated with high county-level rates of violent injury after other known county-level risk factors were controlled for. 6 We updated those analyses by including individual-level data and conducting multilevel analysis. Our purpose was to investigate whether the contextual effects of racial segregation are associated with increased risk of violent injury beyond an individual's risk. We used multilevel analytic techniques to simultaneously account for individual- and county-level risk factors. The model consisted of 2 levels, individual and county. Individual-level data were collected from Pennsylvania hospital discharge data, and county-level data were collected from the 1990 US Census. 16
