A tale of many neighborhoods: latent profile analysis to derive nationally representative neighborhood typologies in the US

Abstract

BACKGROUND AND AIM: Neighborhoods are critical to health, yet challenging to model effectively and consistently across geographies. We used latent profile analysis (LPA) to derive a typology of residential neighborhood found across the US, and then explored how self-reported health varies between neighborhood types. METHODS: We defined neighborhoods as census tracts and compiled 2015-2019 ACS tract-level estimates for neighborhood indicators of commute mode, education, immigration, race, recent growth, and urban form. LPA models ranging from 1 to 6 classes were fit to these data. We the used a one-way ANOVA to compare tract-level 2019 prevalence estimates of poor physical health and poor mental health, provided by the US Centers for Disease Control and Prevention PLACES dataset, between neighborhoods grouped by latent profile. RESULTS: Model fit statistics identified a 6-profile model permitting varying variances and covariances as the optimal categorization of neighborhood profiles. Of the 73,057 census tracts classified, 55% of tracts fell into one of two profiles that were broadly characterized as rural typologies. The remaining 45% of tracts fell into four broadly urban profiles, differentiated by sociodemographic composition, education, and recent growth. Tracts characterized as rural and with a high racial/ethnic minority class had the highest poor health prevalence (MeanPoor Physical health(SD): 16.3 (3.96) , MeanPoor Mental health(SD): 18.0 (3.42), whereas tracts characterized as urban with high education had the lowest poor health prevalence (MeanPoor Physical health(SD): 10.5 (2.40), MeanPoor Mental health(SD): 12.4 (2.6); p<0.001) CONCLUSION: LPA can be used to derive meaningful and standardized partitioning of census tracts sensitive to the spatial patterning of health in the US. These neighborhood types can be used to quickly categorize tracts in the US, and there is potential to replicate these methods in non-US contexts to promote population health globally. KEYWORDS: neighborhood, built environment, social environment, latent profile analysis