Constructing built environment indices using alternative standardizations and spatial scales

Objectives: Existing approaches to neighborhood-level index development have often ignored the implications of standardization choice and relied on administrative boundaries to define neighborhoods. Using objective survey data on the built environment (BE), we constructed BE indices using alternative standardizations with varying geographic scales to represent neighborhood context. We assessed the impact of alternative standardizations on index reliability and validity.  

Methods: In 2008 and 2011, the Children’s Environmental Health Initiative conducted parcel-level assessments of 57 BE variables in Durham, NC (parcel N=30,293). Based on a priori defined variable groupings, we constructed 7 mutually exclusive BE domains (crime, tenancy, nuisances, housing damage, property disorder, vacancy, and security). Domain-based indices were developed by standardizing component variables according to count per parcel, count per parcel area, and proportion of census block area with a given variable. A raw count was used for comparison. Component variables were aggregated to the census block and primary and secondary adjacency levels. Preservation of rank was calculated (Spearman’s rho) within each index (across standardizations) and among indices (within standardization). To assess validity by standardization, we examined correlations among indices and census variables.

Results: Security was impacted by standardization (0.68<rho<0.91) while ranking of other indices was preserved. The relationship between tenancy and security was less preserved across standardizations (-0.16<rho<0.58) than the relationships among other indices. For each BE index, the correlation with census variables depended on standardization (0.27<rho<0.71). Indices constructed at varying spatial scales that smooth over administrative boundaries resulted in less variability within index.

Conclusion: Standardization choice substantively altered ranking within selected indices, relationships among indices, and relationships among indices and census variables. This has implications for measuring aspects of BE, both singly and jointly, particularly when linking BE indices to health outcomes. Associations with health outcomes should be evaluated in terms of alternative standardizations and spatial scales.