Measuring racial disparity in small areas: The problem of aggregation bias in foster care admission disparities

Background: Disparity is a place-based construct. Relative disparity, measured as the ratio of incidence rates between two groups, is necessarily geographically bounded. For instance, disparity is often described among states or counties, but never individuals. The choice of geographic boundary may reveal an aggregation bias such that relative disparity changes depending on spatial scale.

Methods: Administrative foster care records containing removal home addresses of a cohort of 2,508 children who entered foster care for the first time in one Tennessee county between 2000 and 2006 are geocoded and assigned to census tracts. Admission rates for African American and white children and a relative disparity ratio are constructed and compared to rates at two larger spatial scales – state and county.

Results: The underlying populations of children in each area impact the calculation of African American to white admission disparity. The statewide disparity ratio is 1.2 while disparity for the region of Shelby is 3.5, and disparity within Shelby County ranges from 0.0 to 27.4. Furthermore, disparity cannot be described in 25% of the census tracts because they are highly segregated and contain nearly no white children upon which to make a disparity calculation.

Conclusion: Disparity is an artifact of data aggregation. As such, relative disparity can be a misleading metric by which to understand local system dynamics. A thorough analysis of disparity should include consideration of absolute disparity and its component parts (in this case, African American and white admission rates), which are not vulnerable to the same pitfalls as disparity.