Where people live matters. There is strong and consistent evidence showing variations in health across geography (Roux et al., 2010; Larson et al., 2009; Milliren et al., 2023; Mair et al., 2008; Rundle et al., 2007; Doyle et al., 2006; Morland et al., 2002; Saelens et al., 2003; Sellström and Bremberg, 2006). There is a 20-year gap between counties with the lowest and highest life expectancy in the U.S., with disparities across place widening over time (Dwyer-Lindgren et al., 2017; Schnake-Mahl et al., 2022; Boing et al., 2020; Wami et al., 2021). Yet it remains a challenge to isolate singular causes of that inequity; historic social stratification at the neighborhood-level and differential patterning of place-based resources ensures that households with fewer resources tend to cluster in neighborhoods with fewer resources, and the interaction of household-level and neighborhood-level deprivation likely contributes to population health disparities (Roux et al., 2010; Williams and Collins, 2001).

Recent research has prioritized isolating individual neighborhood characteristics (e.g., neighborhood socioeconomic status or physical attributes of the built environment) that can, in principle, be modified to improve health. This approach has largely ignored how neighborhoods function as complex systems consisting of interrelated factors from which population health is an emergent effect (Roux et al., 2010). For example, while increasing neighborhood walkability might be thought to increase physical activity among neighborhood residents, increasing or changing neighborhood characteristics that enhance walkability might also increase property values, potentially displacing current residents with more well-resourced individuals who otherwise already have more opportunities to be physically active (Immergluck and Balan, 2018). To date, most studies have examined whether and which individual neighborhood characteristics differ between neighborhoods and are related to health outcomes. However, few studies have isolated a single modifiable characteristic that clearly can improve neighborhood population health. This observation is likely because multiple neighborhood factors operate in complex ways to influence health outcomes.

An alternate approach to studying neighborhood effects on health, as exemplified by the landmark Moving to Opportunity study, (Leventhal and Brooks-Gunn, 2003) considers the individualized effect of moving between neighborhoods. This approach treats neighborhood features as fixed; that is, the study estimates the effect on individuals of moving from neighborhood A to neighborhood B but does not consider the effect of changing specific features of neighborhood A to be more like neighborhood B. This design better incorporates the complex systems that are unobserved but create synergies between neighborhood features (e.g., accumulations of political power may result in neighborhoods with wealthier residents having more greenspace, but the effect of wealth and the effect of greenspace do not need to be disentangled for the effect to be estimated).

However, it remains unclear how to consider and characterize these complex neighborhood systems succinctly. For example, a researcher examining the effects of moving needs to know whether a study participant moved between similar or different neighborhoods, but the idea of whether two neighborhoods are similar is not adequately defined by any single measurable neighborhood characteristic or index score. For example, two or more neighborhoods with the same retail food index or walkability index score could be represented by different patterns in their component scores. In this light, a typology of neighborhoods that incorporates the complex patterns of neighborhood characteristics yet also captures the diversity of conditions within and across neighborhoods, offers a more holistic alternative to attempting (and failing) to isolate particular features.

Latent modeling is a common technique for constructing such typologies (Hagenaars And Halman, 1989; Spurk et al., 2020). For example, Weden and collegues (Weden et al., 2011) used latent class analysis, a categorical latent modeling approach, to derive neighborhood archetypes in the U.S. using 1990 and 2000 national decennial census data (Weden et al., 2011). Their approach produced a holistic characterization of space that appeared stable over time, suggesting that these approaches could be used by public health researchers to more comprehensively and consistently approximate neighborhoods across various geographic contexts in neighborhood-health studies.

However, a latent class approach requires categorical indicators to derive its latent structure – that is, each measure must be cut into groups or dichotomized (e.g. population density equals 0 if density is less than the national median, else 1). An alternate approach, latent profile analysis (LPA), similarly constructs latent classes underlying observed indicators, but retains indicators in a continuous state. Identifying classes by maximizing between-group variance and minimizing within-group variance in a set of indicators, as LPA does, may efficiently capture the complex relationships present between several neighborhood indicators to identify neighborhood types. This LPA approach is distinct from other efforts to measure neighborhoods, such as the construction of neighborhood indices, (Kind et al., 2014; Messer et al., 2006; Mujahid et al., 2008) because it preserves the multidimensional nature of neighborhoods that are simplified in the use of indices (Palumbo et al., 2016). LPA-derived types could then be used more readily to examine relationships with health and behavior outcomes.

Recently, population health researchers have applied LPA and similar latent modeling approaches to explore neighborhood features within specific geographic locations, and their relation to health outcomes, (Lekkas et al., 2019) including excess weight, (Adams et al., 2011; Wall et al., 2012; Martinez et al., 2014; Jones and Huh, 2014) physical activity, (Adams et al., 2011, 2012, 2013; Kurka et al., 2015; Norman et al., 2010; McDonald et al., 2012; Mooney et al., 2015; Alves et al., 2013) breast cancer, (Palumbo et al., 2016; Shariff-Marco et al., 2021; DeRouen et al., 2019) delinquent behaviors, (Anderson et al., 2015) and alcohol consumption (Cronley et al., 2012). We build on this, and other existing work, to demonstrate the continued utility of latent approaches for large-scale characterization of place that is meaningful for population health outcomes. In this analysis, we use recent census data to derive a nationwide neighborhood typology using LPA at the census tract-level, and test for differences in self-reported health across the derived typology. We hypothesize that LPA-derived neighborhood profiles will reflect historical and contemporary processes that drive residential patterning and that these neighborhood profiles will be associated with health.