Purpose: The current popularity of neighborhood-based child welfare services is not without evidence that certain neighborhood social characteristics are linked to child maltreatment. System reform efforts seek to target neighborhoods at high risk for child welfare involvement in hopes of strengthening families and communities and, in turn, protecting children from harm. This strategy has the added benefit of fiscal efficiency; if rates of maltreatment and foster care entry vary widely across neighborhoods, then directing resources to the areas with the greatest need is an appropriate response. However, the problem of how to define neighborhoods limits the usefulness of this approach for neighborhood-based policy design and implementation. This study addresses this problem by using spatial data analysis to investigate the relationship between neighborhood social structure and foster care entry at three spatial scales (zip codes, census tracts and block groups) and asks whether this relationship changes depending on how planners may classify neighborhoods.

Method: First-time entries to foster care from a California County between 2000 and 2003 are geocoded by their removal home addresses and assigned to neighborhoods at each of the three scales. Next, homogeneity is explored among foster care entry risk and the social structural factors that make up the neighborhoods at each scale. Finally, Ordinary Least Squares models are compared with spatial regression models that detect and control spatial autocorrelation among these administrative units across geographic scales to examine the relationships between foster care entry risk and neighborhood social structure.

Findings: 1) While ZIP code areas are probably too large to represent homogeneous neighborhoods, they produce the most stable and robust model. Block groups may be better proxies for neighborhoods from a practical perspective because they are much smaller but they create a host of small area analysis problems that make them less than practical for estimation. Census tracts are somewhere in between. 2) Generally, the OLS and spatial lag models are very similar within and across scales. Impoverishment and Hispanic immigrant concentration are strong predictive factors of high foster care entry rates, no matter how neighborhoods are delineated. Residential instability was positively related to entry rates, but not at the census tract level, once spatial autocorrelation was controlled.

Implications: This study supports previous research showing relationships between foster care entry risk, certain neighborhood characteristics and their geographic locations. However, some effects change depending on how neighborhoods are defined. This suggests the importance of considering spatial scale and spatial autocorrelation in ecological studies before settling on any one unit of analysis to represent neighborhoods. Explicitly accounting for the role of space can offer more accurate estimates of neighborhood effects and point to the areas at highest risk for foster care entry. In turn, this information can guide policy initiatives that seek to concentrate services and resources in high-risk neighborhoods.