Neighborhood disadvantage and adolescent behavior problems: how do inferences differ across definitions of disadvantage?

Background: Various definitions of neighborhood disadvantage (ND) exist, yet evidence is lacking on how ND operationalization may impact scientific inference. Methods: We used data from wave 6 of the Fragile Families study, excluding those without census tract or behavior problem data (n = 2363). Outcomes included five scales from the caregiver-reported Child Behavior Checklist (CBCL) and a self-reported delinquency scale. ND was defined in four ways: (1) a modified Sampson definition which included four neighborhood poverty variables; (2) a poverty-only definition which only included percent households below poverty; (3) an overextended definition which added a mediator between the ND-behavior relationship, and (4) an expanded definition which added six additional ND variables to the modified Sampson definition. Using effect estimates from generalized linear models, differences were calculated using percent change-in-estimate, with the modified Sampson as the referent. Results: Effect estimates were similar for the modified Sampson and expanded definitions (< 5% difference). The poverty-only definition differed from the modified Sampson unsystematically. Estimates for the overextended definition were consistently larger compared to modified Sampson (10–37% greater). The expanded and modified Sampson definitions produced similar results. Conclusion: Poverty-only and overextended ND definitions should be interpreted with caution.