The Built Environment and Walking Behavior: Addressing Residential Sorting through Matching in the Cardia Study

Recent research has shown neighborhood walkability to be positively associated with walking behavior. However, accounting for selective migration and socio-spatial segregation is an ongoing challenge. These forms of residential sorting may result in individuals who are not comparable across levels of neighborhood walkability and may therefore lead to biased inferences about the built environment. We explored the implications of residential sorting using coarsened exact matching (CEM), a method that compares individuals who have similar characteristics but are exposed to different “treatments” (here, different levels of neighborhood walkability). We used data from 2,840 participants in Year 20 (2005–2006) of the Coronary Artery Risk Development in Young Adults study. The exposure was a walkability index created from measures of population density, land use mix, street connectivity, and parks, validated with existing walkability measures. Outcomes included a binary measure of walking participation (yes/no) and a continuous measure of walking exercise units. Covariates included socio-demographics (age, gender, race, education, income, household size, marital and employment status), general health status, and neighborhood selection factors (selected residential neighborhood for reasons related to its built environment). We matched participants who had the same covariate values but lived in different neighborhood walkability types (low, medium, high); unmatched participants were dropped. We used two-part models for mixed discrete-continuous outcomes to estimate associations between walkability and walking before and after CEM. We found significant differences by neighborhood type for most covariates prior to CEM. After CEM, 77% of participants were dropped for not having matches in the other neighborhood types; non-white individuals and those with lower socioeconomic status (SES) were less likely to be matched.