Economic inequality and child mortality

Despite significant declines in child mortality rates since 1950, socio-economic disparities in child mortality across levels of socio-economic disadvantage have not only persisted [18], but they have increased [19]. During the past four decades, income inequality in the Chicago region, for example, increased by nearly 36%, surpassing the 19% rate of increase for the United States [20]. Socioeconomic inequality is a reflection of housing insecurity, lack of job opportunities, quality education, healthcare, and childcare, and remains a critical social determinant of mortality among youth [16,19]. Spatial concentrations of child mortality risk have been attributed to variations in material wealth across communities [21]. Past research has demonstrated socioeconomic gradients in all-cause child mortality and specific-cause mortality in vehicular accidents [22], pedestrian injury [23], fires [24, 25], drowning [26], sudden infant death syndrome [27], child maltreatment [28], and congenital conditions [29]. The survival advantage of children who live in the least deprived neighborhoods has been attributed to the differences in quality education, employment, and healthcare opportunities found in wealthier areas [18,30]. Not all socio-economically disadvantaged neighborhoods within the same geographic region experience the same level of inequality, however. In metropolitan areas, the observed spatial heterogeneity in child mortality outcomes is influenced by a diverse range of social, economic, and environmental determinants that vary significantly depending on the region being considered [19,31–33]^. Neighborhood structural disadvantage has been reinforced by punitive justice-related and discriminatory housing policies, as well as antipoverty policies that aim to supervise and adjudicate family life in impoverished areas, serving as pathways by which structural determinants shape health outcomes for children [34, 35].

Neighborhood boundaries as social frontiers

Neighborhood boundaries may be defined as homogenous areas, such as census tracts or counties, or as spatial zones of rapid change often referred to as social frontiers [35–37]. Even when neighborhoods exhibit internal similarity based on specified characteristics, they may differ significantly in terms of their sociospatial positioning, as measured by how they compare to adjacent areas. Social frontiers, which represent spatial divisions in racial, ethnic, religious, cultural, or social characteristics that separate neighboring communities, act as “cliff edges” between adjacent neighborhoods [35]. These boundaries are scientifically important because they mark areas where social or economic conditions shift abruptly, distinguishing characteristics on one side from those on the other [37]. Unlike conventional boundaries, social frontiers are open, meaning they appear as line segments (i.e., not polygons) and do not enclose discrete areas. Because the processes that give rise to boundaries are not always associated with homogeneous areas, social frontiers may matter more on “one side where they divide Black [sic] city residents from surrounding White [sic] suburban areas, but they may not matter on another side (p. 20)” [38].

Social frontiers typically coincide with the edges of highly segregated regions where neighboring areas are more homogenous [38–41]. Social frontiers may be caused by a “strong aversion [among residents] to living at the interface of communities in conflict (pg. 274)” [35,42] or because of “benign forces” [35,43] and unintended consequences of micro-level (i.e., individual) decisions. Alternatively, social frontiers may emerge from explicit and intentional policies that have prevented people in lower-income communities of color from accumulating wealth and investing in home ownership rather than being an “accident of economic circumstance, demographic trends, personal preference, and private discrimination” [44]. Regardless of their origins, social frontiers are places of stark spatial divisions with complex social consequences. Such variation has important implications for understanding the impact of socio-spatial segregation and its disparate effects on health and well-being. These boundaries demarcate areas with differences in level of service and resource distribution, cultural and place-based identity, and social interactions, all of which can impact the social dynamics of these areas [45]. Residents’ social interactions and demographic characteristics shape the meaning and perception of the places where people live, even within the same geographic area [46].

How do social frontiers affect child mortality risk?

To fully understand the impact of spatial opportunity structure on health and well-being, it is essential to look beyond the characteristics and interactions within individual neighborhoods [45,47]. More specifically, it is essential also to consider spillover effects—how conditions in one neighborhood impact the health and well-being of neighboring areas [45]. Externality space theory provides an organizing framework for conceptualizing differences across bordering areas. According to the theory, places “where individuals do not share externalities with proximate others is axiomatically a boundary between … neighborhoods … that identify schisms in the social meaning or attributes of space at a given point … (pg. 83).” Boundary areas significantly alter the benefits and detriments that individuals derive from a particular location [45].

Externality space theory identifies three core dimensions that help explain how and why boundaries shape neighborhood-level outcomes: congruence, generality, and accordance. Congruence refers to the level of concordance between the externality space and a defined geographic boundary. In the present case, the congruence between EHI and child mortality risk is high when confined to specific neighborhoods. The generality of an externality space refers to how consistently an individual’s well-being is mapped to the same area. As applied here, if externalities related to neighborhood economic inequality (e.g., healthcare access) are associated with increased child mortality risk, then the externality space has high generality. Accordance refers to whether all individuals share a common understanding about where the boundaries of EHI begin and end (i.e., the externality space). The externality space theory of neighborhood boundaries introduces a novel approach to understanding urban dynamics [45] because it emphasizes important linkages across different sociospatial contexts— including the interrelatedness between space, sociability, and spatial conditions [34]. In short, this perspective highlights residents’ social marginalization due to the imposed nature of neighborhood physical boundaries.

At social frontiers, spatial proximity may decrease social interaction and the sharing of social institutions across groups [48]. Social frontiers may erect barriers to increased social interactions between neighbors, leading to greater mistrust and misunderstanding between groups, lower levels of collective efficacy, and increased child mortality [35,49]. This is supported by Legewie (2018), who found that even after considering traditional indicators of social disorganization, including residential instability, immigrant concentration, ethnic heterogeneity, and concentrated disadvantage, racial neighborhood boundaries continued to predict violent crime and homicide risk in Chicago. According to Legewie, boundary areas create opportunities for criminal behavior due to the lack of social control and cohesion that characterizes adjacent homogeneous regions, as well as the potential for intergroup conflict. Legewie (2018) used a technique known as ‘areal wombling’ to detect boundaries based on sharp changes in racial-ethnic composition. Drawing on multiple theories, he claims that “...boundaries lack the social control and cohesion of adjacent homogeneous areas; are contested between groups, provoking intergroup conflict; and create opportunities for criminal behavior” (Legewie, 2018, p 1958). In this study, violent crime was higher at boundary locations, representing sharp delineations in racial-ethnic composition, net of neighborhood characteristics, measures of spatial interdependence, and other physical boundaries. Kim and Hipp (2018) similarly found that proximity to the city boundary, as a distance decay, predicted increased motor vehicle theft [50]. Boundaries between two areas have been shown to increase crime opportunities, partly because of the possibility of fewer guardians, or the transitory nature of these locations, making escape quicker and easier for perpetrators of violent crime [51].

Regarding child mortality specifically, the perception of relative resource deprivation can lead to social disenfranchisement, mental distress, and poor health outcomes, particularly if the deprivation is attributed to social stigma and racism [48]. Research shows that the overlapping and reinforcing effects of residential segregation are associated with higher rates of infant mortality and preterm birth among Black infants [52,53]. One study found that the excess risk of mortality among preterm infants born in neighborhoods characterized by high relative concentrations of Black and poor residents is explained in part by the hospital in which they receive care [52]. More specifically, infants from predominantly Black neighborhoods experienced worse outcomes when born in hospitals located in similarly disadvantaged areas than when born in hospitals located in neighborhoods with a higher proportion of White residents.

In contradistinction, studies have shown that spatial proximity can increase, rather than decrease, social interaction and the sharing of institutional resources across regions [48]. For example, ethnic diversity and multiculturalism have been viewed as positive externalities because they promote adherence to positive health behaviors consistent with child well-being through disease prevention and information about healthcare service locations [54–57]. Cross-cultural work has demonstrated locational advantages associated with proximity to affluent neighbors, such as greater accessibility to labor market opportunities and economic integration, but only when public spaces are shared effectively by socially dissimilar groups [33]. Moreover, increased social interaction in heterogeneous spaces can mitigate the impact of negative stereotypes and group threats and increase trust across groups [58,59]. In the context of child mortality, behavioral changes resulting from social interaction across diverse environments may be crucial for improving children’s survival. For instance, one study found that infant mortality is lower among rural-urban migrants than rural non-migrants, highlighting the role of social interaction in exposing individuals to new ideas, which can lead to changes in attitudes, lifestyles, and motivations [60].

Current study

Past research confirms the importance of neighborhood boundaries as defining aspects of highly segregated urban areas; however, the precise nature of this association remains challenging to predict. Living in proximity to a social frontier may impact child mortality by increasing access to buffering resources that can mitigate poor child health outcomes for those living in highly disadvantaged areas [61]. On the other hand, living near a social frontier may lead to social isolation and lower levels of collective efficacy, which can result in an increased mortality risk. To fully examine the effect of spatial inequality on child mortality, it is critical to consider multiple measures of economic inequality across and within regions. Boundary analysis captures the transition from low to high-class neighborhoods by detecting statistically significant latent discontinuities over a geographic surface [49,62]. In the present study, a novel form of spatial analysis called boundary detection is used to detect social frontiers of EHI – neighborhood boundaries based on extreme differences in economic inequality – and their association with child fatality in Cook County, Chicago. We hypothesized that living near a social frontier would be associated with decreased mortality due to the potential flow of resources from areas of less disadvantage.

Data

We extracted all-cause mortality data for children under 18 from 2015–2023 from the Cook County Medical Examiner’s Office’s (CCMEO) publicly available database [63]. The CCMEO investigates deaths due to criminal violence, accidents, suicides, and other suspicious or unusual circumstances. The data include detailed records of deaths occurring in Cook County, Illinois, from August 2014 to the present and are updated daily. Records include cause and manner of death, demographic characteristics, and other relevant information regarding the death.

To measure neighborhood-level socioeconomic conditions, we used the Economic Hardship and Inequality (EHI) Index, a tract-level subcomponent of the Area Deprivation Index (ADI) [64]. The ADI is a composite measure of disadvantage derived from the American Community Survey (ACS) 5-year estimates (2015–2019) [65].

Dependent variable

Child Fatalities. Observed counts associated with each child death location, were aggregated within census tracts. The number of children under 18 in each census tract in the county was downloaded from the American Community Survey (ACS) based on 5-year estimates for 2011–2015 and 2015–2019. Based on population size, the expected child fatality count, for each census tract was computed using the SpatialEpi [66] package in R.

Independent variable

Economic Hardship and Inequality (EHI) Index The ADI is a composite measure of disadvantage derived from the American Community Survey (ACS) 5-year estimates (2015–2019), incorporating indicators related to income, education, employment, and housing quality. The EHI subindex specifically reflects spatial patterns of economic deprivation, integrating measures such as the ratio of households earning under $10,000 to those earning over $50,000, the percentage of owner-occupied housing, unemployment rates, and vehicle access. EHI scores are standardized to facilitate within- and between-state comparisons. We downloaded tract-level EHI data using the get_adi function from the sociome package in R (version 4.3.2). For additional detail on scale construction, see [24,64,67].

Covariates Our models used multiple measures of social disorganization by creating indices corresponding to concentrated disadvantage, residential instability, immigrant concentration, and residential segregation similar to past research [49,68–70]. Following Legewie [49]: (1) concentrated disadvantage is a five-item index (eigenvalue 2.962; 59.2% variance) comprised of the following measures: poverty rate (factor loading: 0.847), unemployment rate (0.832), percentage of professional and management jobs (−0.681), the share of high-school graduates (0.616), and share single mother families (0.842); (2) residential instability is a three-item index (eigenvalue 1.85, 61.5% variance) comprised of the percentage of renter-occupied units (0.998), share of residents who moved to another dwelling since 2005 (0.490), and housing unit rental vacancy rate (0.781); (3) immigrant concentration is a three-item index (eigenvalue 2.54, 84.7% variance) comprised of the share of foreign-born residents (0.913), the share of residents who speak English less than “very well” (0.998), and the share of Spanish-speaking residents (0.844). Notably, the positive loading on the variable measuring educational attainment on the factor we labeled concentrated disadvantage may appear counterintuitive at first. However, this variable is defined as the number of individuals whose highest level of educational attainment is a high school diploma or equivalent. In this context, a higher proportion reflects limited access to higher education, which aligns with greater socioeconomic disadvantages. To measure the level of racial or ethnic segregation, or the concentration of different racial or ethnic groups within each neighborhood, we computed the Hirschman-Herfindahl Index (HHI). The HHI is calculated by squaring the proportion of the population that belongs to each racial or ethnic group and then summing the squared values. This index ranges from 0 to 1, where a value closer to 0 indicates complete diversity (i.e., a more even distribution of racial or ethnic groups), and a value closer to 1 indicates a highly segregated neighborhood. All social disorganization variables were derived from the American Community Survey (ACS) 5-year estimates for 2015–2019 [65].

Statistical analysis.

Overview of boundary detection approach. Our analysis proceeded as follows. We first calculated the decile-based EHI score for each census tract and computed the absolute difference in EHI deciles between geographically adjacent areas. Geographic adjacency was defined using a binary spatial weights matrix W, constructed from first-order contiguity via the poly2nb and nb2mat functions in R. We then generated a normalized dissimilarity matrix Z, where each element z_ij represents the scaled absolute difference in EHI deciles between adjacent census tracts i and j, such that We examined the distribution of EHI differences between neighboring areas to identify geographic borders where these differences are statistically significant. To model the relationship between child deaths and neighborhood disadvantage, we employed a dissimilarity-based conditional autoregressive model proposed by Lee [70] and Lee & Mitchel [71]. This approach incorporates spatial smoothing while allowing for spatial discontinuities across structural boundaries. The modified spatial neighborhood matrix, W*, is calculated as where is 1 if tracts i and j are adjacent and 0 otherwise. This specification ensures that spatial dependence is preserved among similar neighbors and gradually attenuated across areas with greater EHI dissimilarity, rather than enforcing an arbitrary threshold. All contiguous pairs contributed to the spatial structure in the final model, with greater smoothing across socioeconomically similar tracts and reduced smoothing across potential “social frontiers.” To visualize high-contrast boundaries, differences in EHI across adjacent tracts were visually mapped to show areas of greater structural inequality. Fig 1 illustrates a region characterized by a significant difference in EHI across boundaries.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Illustration of the boundary detection method.

The figure shows a census tract that neighbors two other census tracts labeled by lines A and B (left). The figure to the right shows that this tract was categorized as one of the least deprived 20% of neighborhoods by the EHI index, as shown in yellow. However, it borders two census tracts classified as the most deprived 20% of census tracts by the EHI.

https://doi.org/10.1371/journal.pone.0330449.g001

Modeling Following Lee and Mitchell (2012), a Bayesian spatial CAR model was specified as

(1)

(2)

where the risk of mortality in area k is denoted by , is the matrix of covariates representing indicators of social disorganization, and the random effects are specified as The random effects allow for overdispersion and spatial correlation in the data, even after controlling for the effects of social disorganization. Following Lee [71] and Lee and Mitchell [72] the S.CARdissimilarity function in R was used to run the localized CAR model, including the K x K dissimilarity matrix. As applied to the present analysis, incorporating the matrix into the modeling scheme allowed us to identify the magnitude of the difference in EHI between two areas. It was used to inform the spatial adjacency matrix governing the random effects structure, rather than being entered as a fixed-effect predictor. In this model, the random effects in neighborhoods that share a border are modeled as correlated or conditionally independent, depending on whether EHI in the two bordering areas is similar (correlated random effects) or very different (conditionally independent) [73]. Spatial correlation is induced using a CAR prior for the random effects via the K x K neighborhood matrix W*. We fit three separate boundary detection models: (1) a model without covariates; (2) a model with concentrated disadvantage, residential mobility, racial/ethnic diversity, and immigrant concentration; and (3) a final model that included spatially lagged versions of each covariate. Spatially lagged variables were calculated as spatially weighted averages of each variable across neighboring census tracts, using a predefined spatial weights matrix. We evaluated the presence of multicollinearity using Variance Inflation Factors (VIFs). In the base model without lags, all VIFs were well below 5. After adding lags, values increased moderately (e.g., concentrated disadvantage lag = 6.10; immigrant concentration lag = 5.00), but remained within an interpretable range and were lower than those reported by Legewie (2018) in a similar analysis. To further assess collinearity, we examined posterior correlations among regression coefficients. Strong negative correlations were observed between the variable concentrated disadvantage and its lag = –0.70, suggesting that these effects should be interpreted jointly rather than independently.

Inference for all models was based on 10,000 post-burn-in and thinned Markov Chain Monte Carlo (MCMC) samples and model convergence was based on Geweke diagnostics [74,75]. The residuals were assessed for spatial correlation using Moran’s I permutation test. Moran’s I measures the degree of spatial autocorrelation, indicating whether similar values are clustered or dispersed across a geographic area. A permutation test is used to assess whether the observed Moran’s I value is statistically significant, comparing it to a distribution of Moran’s I values generated under random spatial arrangements [76,77].

This is a retrospective study using publicly available archived records, and all data were fully anonymized before being accessed on March 26, 2024. According to the U.S. Department of Health and Human Services Policy for Protection of Human Research Subjects, research involving deceased persons is not considered human subjects research. Therefore, the study was deemed exempt from Institutional Review Board review.

Implications for policy and practice

The results of this study hold several public policy implications relevant to child health outcomes. Considering our results, policies aimed at reducing economic inequality rather than poverty or increasing income can play a vital role in mitigating child mortality risks. For example, cash transfer programs that reduce poverty, improve economic autonomy, and enhance child nutrition and health service utilization are associated with significant reductions in mortality among children under five years of age and women [84]. However, while such programs reduce poverty, they have been shown to have a more limited impact in reducing inequality [85]. On the other hand, mixed-income housing developments may effectively address inequality with other interventions that reduce prejudice and intergroup competition, facilitate civic participation, and support social well-being in built environments [86]. There are other effective ways to address economic inequality by focusing on neighborhood transformation through place-based initiatives [87], such as the Dudley Street Initiatives Promise Neighborhoods Program in Boston [88]. Place-based strategies such as this deliver comprehensive social services, healthcare, education, and child welfare, and can help families living in more deprived areas receive the same level of support as those in more affluent areas. By addressing neighborhood inequalities directly, these programs provide children with spatial access to quality education, healthcare providers, and safer environments, thereby indirectly reducing the risk of child mortality [89]. By focusing on EHI, these interventions not only reduce child mortality, but they also foster long-term resilience in the most disadvantaged neighborhoods.

Social frontiers should be public health priority areas where interventions are focused on bridging the gap in EHI between affluent and deprived areas through infrastructure and community development initiatives. These initiatives include affordable community centers and healthcare facilities that offer multifaceted services conditioned by local needs. To address spatial discontinuities in EHI, interventions to reduce child mortality risk should focus on localized opportunity structures rather than administratively defined areas [90]. This includes services targeted to areas of relative resource deprivation, i.e., “social frontiers” that serve both sides of the divide but are specifically designed to reduce the stark inequalities. Implementing mobile health clinics or social services that specifically target neighborhoods at social frontiers can improve access to healthcare in disadvantaged areas. Targeting areas of concentrated disadvantage directly confronts the socioeconomic divides that contribute to higher child mortality rates.

Limitations

This study highlights the crucial role of neighborhood environments in shaping health and socioeconomic disparities. Our conceptual and methodological approach emphasizes the significance of spatial factors in shaping policies aimed at enhancing child well-being, highlighting neighborhood boundaries and economic disparities as crucial elements in reducing child mortality, particularly in urban areas. Nevertheless, our study is not without limitations. First, the use of area-level data, such as census tracts, may obscure finer-scale variations within neighborhoods, limiting the ability to capture small-scale dynamics that influence child mortality. Additionally, the model assumes linear relationships between socioeconomic inequality (EHI) and child mortality, which may not fully capture more complex interactions. There is also the potential for unmeasured confounding factors that we did not consider that are known to influence both socioeconomic inequality and child mortality rates. In particular, future studies would benefit from incorporating additional covariates, such as availability of health care, natural environmental exposures, and aspects of the built environment to more fully capture the contextual factors influencing child mortality risk across neighborhood boundaries. Additionally, the use of longitudinal designs would allow researchers to explore how temporal shocks, such as the COVID-19 pandemic, interact with spatial risks over time. Additionally, moderate levels of multicollinearity in our spatially lagged model may contribute to coefficient suppression or unreliability, warranting caution when interpreting the results to avoid overinterpreting the direction or magnitude of the coefficients. The cross-sectional nature of the data limits the ability to account for temporal dynamics, and the findings may not generalize beyond Cook County, Illinois. Although our results were robust to alternative specifications for the EHI dissimilarity metric, using other indicators of socioeconomic deprivation may result in different results. These limitations should be considered when interpreting the findings, and future research would benefit from replicating our results to provide more insight into the impact of broadly defined inequality on child mortality.