Abstract
The past few years have witnessed an explosion of interest in neighborhood or area effects on health. Several types of empiric studies have been used to examine possible area or neighborhood effects, including ecologic studies relating area characteristics to morbidity and mortality rates, contextual and multilevel analyses relating area socioeconomic context to health outcomes, and studies comparing small numbers of well-defined neighborhoods.
Strengthening inferences regarding the presence and magnitude of neighborhood effects will require addressing a series of conceptual and methodological issues. Many of these issues relate to the need to develop theory and specific hypotheses on the processes through which neighborhood and individual factors may jointly influence specific health outcomes. Important challenges include defining neighborhoods or relevant geographic areas, identifying significant area or neighborhood characteristics, specifying the role of individual-level variables, incorporating life-course and longitudinal dimensions, combining a variety of research designs, and avoiding reductionism in the way in which “neighborhood” factors are incorporated into models of disease causation and quantitative analyses.analyses.
ALTHOUGH INTEREST IN geographical variations in health has a long history,1 the importance given to the examination of area differences in studying the causes of disease has varied over the years. With few exceptions (e.g., see references2–4), the focus on individual-level risk factors over the past few decades was generally associated with little interest in area characteristics as potential disease determinants. However, the notion that “place” may be important to health reemerged in a handful of publications in the 1980s and early 1990s,5–9 and interest has increased sharply in recent years. Numerous reports on area or neighborhood differences in health have recently appeared in epidemiology and public health journals (e.g., see references 10–28). In this issue of the Journal, Rauh et al.29 and Pearl et al.30 examine the role of neighborhood or community contexts in shaping the distribution of low birthweight.
Several factors have converged to stimulate a resurgence of interest in area or neighborhood health effects. Chief among these factors has been a rekindling of interest in the social determinants of health31,32 and the recognition that social influences on health operate through many different processes, one of which may be the types of areas or neighborhoods in which people live.33–35 Simultaneously, there has been a growing discussion of the use of ecologic variables in epidemiology.36–38 This discussion is related to a critique of the notion that all health determinants are best conceptualized as individual-level attributes. Research on neighborhood effects has fit into this emerging paradigm because it has postulated that neighborhood contexts may be related to health, independently of individual-level attributes. In addition, the emergence of new methodological approaches such as multilevel analysis39–42 has stimulated both thinking and empiric research in this field.
Sociologists and social geographers have long recognized the importance of neighborhood environments as structural conditions that shape individual lives and opportunities.43,44 In the United States, recent discussions in sociology on the causes and consequences of residential segregation and urban poverty45–47 have reinvigorated interest in the sociologic and ethnographic investigation of neighborhoods and the ways in which neighborhood contexts may affect individuals. Social scientists have examined how living in areas of concentrated poverty influences individual-level outcomes such as employment and single parenthood.43 Neighborhood context has been investigated in relation to violence,48 child development,49,50 and childbearing practices.51 In public health, it has been argued that the physical and social environments of neighborhoods may be important in understanding the distribution of health outcomes.33–35,52,53
Investigation of how neighborhood environments may be related to health is not only of academic interest. Documentation of neighborhood effects, as well as elucidation of the mechanisms through which they are mediated, could have important policy implications for health promotion and for the reduction of health disparities. Neighborhood differences may be especially relevant in the context of increasing spatial concentrations of poverty and increasing geographic clustering of poverty with other forms of disadvantage.54 The articles by Rauh et al.29 and Pearl et al.30 illustrate some of the key conceptual and methodological challenges faced by researchers as they investigate neighborhood health effects.
Basically, 3 empiric strategies have been used to investigate neighborhood or area effects: ecologic studies, contextual or multilevel studies, and comparisons of small numbers of well-defined neighborhoods. Ecologic studies have been used to examine variations in morbidity and mortality rates across areas and to relate this variability to area characteristics. The sizes of areas examined have varied greatly, ranging from relatively large areas not really analogous to neighborhoods at all (e.g., counties in the United States or district health authorities in the United Kingdom)55–58 to smaller areas (e.g., census tracts in the United States or wards in the United Kingdom).59–62 The most common area characteristics investigated have been aggregate measures of the socioeconomic characteristics of residents or indices of deprivation constructed by combining several aggregate measures based on theoretic or empiric considerations.63 These area characteristics have been found to be strongly related to mortality rates (e.g., see references 55565861 and62).
However, the use of such aggregate measures has sometimes generated ambiguity regarding whether these variables are conceptualized as measures of area-level properties or simply as summaries of individual-level variables, and hence whether the objective is to examine how area constructs are related to health outcomes or to document the area-level (or ecologic) expression of a well-known individual-level relation. Although ecologic studies may be useful to document and monitor inequalities in health, they cannot directly determine whether differences across areas are due to characteristics of the areas themselves or to differences between the types of individuals living in different areas. They cannot evaluate the role of individual-level factors as confounders, mediators, or modifiers of the area effect.
The recognition of the need to distinguish the effects of “context” (e.g., area or group properties) and “composition” (characteristics of individuals living in different areas) when examining area effects on health39 has led to a proliferation of reports involving contextual and multilevel analyses. The articles by Rauh et al.29 and Pearl et al.30 employ this analytic approach. Contextual and multilevel analyses require data sets including individuals nested within areas or neighborhoods. By simultaneously including both neighborhood- and individual-level predictors in regression equations with individuals as the units of analysis, these strategies allow examination of neighborhood or area effects after individual-level confounders have been controlled. They also permit examination of individual-level characteristics as modifiers of the area effect (and vice versa).
Furthermore, multilevel analysis allows the simultaneous examination of within- and between-neighborhood variability in outcomes64,65 and of the extent to which between-neighborhood variability is “explained” by individual- and neighborhood-level factors. Studies using these approaches have usually linked information on small-area characteristics available in censuses to individual-level covariate and outcome data from surveys, epidemiologic studies, or vital statistics (as in the 2 studies in this issue). For the most part, contextual and multilevel studies have been consistent in documenting an “independent” effect of neighborhood socioeconomic environment on individual-level outcomes after controlling for individual-level socioeconomic position indicators.10–28,52,53 However, the percentage of total variability in outcomes between areas has often been small,6,35,66–68 and the strength (and relative importance) of the neighborhood or area effect is still under debate.52,53
In contrast to the large-scale quantitative approaches just summarized, an alternative strategy has been to compare a small number of well-defined and purposely selected contrasting neighborhoods.33,69–72 These types of studies can incorporate knowledge on local history, sociology, and geography in defining neighborhoods. In addition, they can directly collect detailed information on neighborhood characteristics and health outcomes through combinations of quantitative and qualitative strategies.33 This approach has been used to document differences across neighborhoods in resources and services and to relate these differences to differences in health behaviors.33,69–72 However, it is limited in the number and range of neighborhoods investigated and possibly in the generalizability of results. Its strength resides in the use of locally based definitions of neighborhoods (rather than administrative proxies) and in the direct collection of data on neighborhood characteristics, which may help provide an understanding of the processes through which neighborhood environments can affect health.
As in other fields,49 strengthening inferences regarding the presence and relative importance of neighborhood or area effects will require addressing a series of conceptual and methodological issues. Many of these issues stem from the need to develop theories and more specific hypotheses on the dynamic processes through which neighborhood and individual factors may jointly influence specific outcomes.73
A first issue is the definition of “neighborhoods” or, perhaps more precisely, definition of the geographic area whose characteristics may be relevant to the specific health outcome being studied. In health research, the terms neighborhood and community have often been used loosely to refer to a person's immediate residential environment, which is hypothesized to have both material and social characteristics potentially related to health. The more generic term area has also been used. Clear distinctions between the terms neighborhood, community, and area are usually not made. Administratively defined areas have been used as rough proxies for “neighborhoods” or “communities” in many studies, including the 2 included in this issue. Rauh et al.29 base their analyses of community-level variables on New York City–defined health areas (clusters of 4–6 census tracts). Pearl et al.30 base their analyses of neighborhood-level variables on the smaller census-defined block groups.
The concepts of “neighborhood” and “community” are not precise.54,73,74 Although both are geographically anchored (the term community is used to implicitly refer to geographically anchored communities in these types of analyses), there are several possible definitions, depending on the criteria used. Criteria can be historical, based on people's characteristics, based on administrative boundaries, or based on people's perceptions. Boundaries based on these different criteria will not necessarily overlap,54 and alternative definitions may be relevant for different research questions.73 For example, neighborhoods defined on the basis of people's perceptions may be relevant when the neighborhood characteristics of interest relate to social interactions or social cohesion, administratively defined neighborhoods may be relevant when the hypothesized processes involve policies, and geographically defined neighborhoods may be relevant when features of the chemical or physical environment (e.g., toxic exposures) are hypothesized to be important.
More generally, the size and definition of the relevant geographic area may vary according to the processes through which the area effect is hypothesized to operate and the outcome being studied.54 Areas ranging from small to large with varying geographic definitions may be important for different health outcomes or for different mediating mechanisms, and many of these “areas” may not be thought of as “neighborhoods” or “communities” in the common sense at all. For example, counties may be important geographic contexts for outcomes potentially related to county policies or economic structures. School districts may be relevant for child outcomes. For some purposes, the relevant area may be the block on which a person resides; for others, it may be the blocks around the residence; and for still others, it may be the geographic area in which services such as stores or other institutions are located.54 The size and definition of the area, the relevant processes, and the outcome being studied are linked. The development and testing of hypotheses regarding the precise geographic area that is relevant for a specific health outcome will help strengthen inferences regarding area effects.
Research also requires operationalizing the theoretically relevant “areas” so that data on individuals and area attributes can be linked to them. Investigators conducting large quantitative analyses spanning many areas often have no choice but to rely on existing administrative definitions for which standard data are available and to which individuals and other sources of area data can be easily linked. Although the use of these proxies is the only practical alternative in many cases, they are obviously limited in that they do not necessarily correspond to the theoretically relevant area. Another alternative involves defining relevant areas or neighborhoods on the basis of local historical, social, and geographic knowledge as well as information collected from residents. Development of these boundaries is an enormous task in itself and may be feasible only in studies focusing on a limited geographic region or city. The challenges involved in operationalizing relevant areas or neighborhoods are important but should not paralyze thinking or empiric investigation. Comparisons of alternative operationalizations with respect to their consequences for research and interdisciplinary work with geographers, urban planners, and social scientists will be important.
Most existing research, including the articles by Rauh et al.29 and Pearl et al.,30 has examined how aggregate measures of neighborhood socioeconomic context are related to health outcomes.52 These associations are compatible with a wide range of processes relating neighborhood environments to health. Both Rauh et al.29 and Pearl et al.30 discuss some of the processes that could mediate effects of neighborhood or community context on birthweight. A next step (and a way of establishing the extent to which neighborhoods or communities are or are not relevant) will be the more direct empiric examination of the specific features of areas (be they physical or social) that may be related to different outcomes.33
Examining the role of specific neighborhood or area characteristics is complex, because many of these dimensions may be interrelated (and thus difficult to tease apart)75 and may also influence each other.33 For example, features of the physical environments of neighborhoods may influence the types of social interactions, and vice versa. In addition, the processes involved and the relevant neighborhood attributes may differ from one outcome to another. For example, mechanisms involving resources and the physical environment may be more relevant for certain outcomes (e.g., physical activity), whereas those involving social norms or contagion processes may be more important for others (e.g., smoking). Living in a deprived area may itself be a source of stress that may be particularly relevant for stress-related outcomes.
Some hypothesized processes may predict linear effects of neighborhood characteristics, whereas others may predict threshold effects. Hypotheses involving social comparisons (or relative deprivation hypotheses) may predict different things, depending on the group within which the comparison occurs. For example, if a person's perception of his or her position relative to others in a given area is important, then being the “best-off” person living in a disadvantaged area could be associated with a better outcome than being the “worst-off” person residing in a better area. Other patterns might be predicted if relevant social comparisons are made across, rather than within, areas.
From the operational point of view, measurement of specific characteristics of neighborhoods is complex. Options for the collection of this type of information include surveys of residents (which may be aggregated up to the desired area level) on objective and subjective characteristics of their neighborhoods, direct observation or videotaping and ranking of neighborhoods on prespecified criteria by raters (systematic social observation), and linking databases with geographically linked information (e.g., from public agencies) and estimating density and distance measures.76–78 Assessment of neighborhoods or areas presents a series of methodological challenges related to the measurement of ecologic settings (termed “ecometrics”77) that have only recently begun to be addressed. The development of valid and reliable measures of relevant area characteristics that can be obtained in a systematic fashion across many areas is an important need in this field.79
A crucial issue in the examination of neighborhood effects is how individual-level variables should be incorporated into conceptual models and included in analyses. The most common criticism of postulated neighborhood effects is that they result from confounding by individual-level variables.80,81 The selection problem73,82 is a variant of this issue: people may be sorted into neighborhoods according to individual characteristics, and these individual characteristics may be related to outcomes. As a way to respond to these critiques, studies have attempted to control for individual-level variables, most commonly indicators of social position.52 Both Rauh et al.29 and Pearl et al.30 make a special point of controlling for individual-level social position indicators via stratification and multivariate adjustment. Short of randomization, these adjustment strategies are the best way to analytically demonstrate an effect of neighborhood or area context on health. Studies demonstrating this “independent” effect are therefore a necessary and valuable first step.
However, the relation between neighborhood characteristics and individual-level socioeconomic position is a complex one. To the extent that neighborhoods influence the life chances of individuals,43,45,82 neighborhood social and economic characteristics may be related to health through their effects on achieved income, education, and occupation, making these individual-level characteristics mediators (at least in part) rather than confounders. In addition, because socioeconomic position is one of the dimensions along which residential segregation occurs, living in disadvantaged neighborhoods may be one of the mechanisms leading to adverse health outcomes in persons of low socioeconomic status. For these reasons, although teasing apart the “independent” effects of both dimensions may be useful as part of the analytic process, it is also artificial. A related issue (which pertains to investigation of contextual effects generally) is that contexts are often underspecified (particularly relative to individual-level constructs, which are usually much easier to define and measure), making comparisons of relative magnitudes of effects problematic and often biased against contextual effects.
Of course, because disease is expressed at the level of the individual, neighborhood factors necessarily exert their effect through individual-level processes, including behaviors and biological precursors of disease. Whether an individual-level variable is conceptualized as a confounder or a mediator depends on the question being asked. Both studies in this issue adjust for individual-level variables (e.g., smoking and substance abuse in Rauh et al.29 and prenatal care initiation in Pearl et al.30) that could be conceptualized as confounders or mediators, depending on the hypothesis being tested. Further complexity results from the fact that, in some cases, neighborhood and individual characteristics may mutually influence each other. For example, the availability of healthy foods in a neighborhood may influence the dietary behaviors of individuals, and individual behaviors may in turn affect food availability. Understanding area or neighborhood effects may require the testing of hypotheses involving such dynamic and reciprocal relations. The methods commonly used in epidemiology today are not well suited to examination of these reciprocal and dynamic relations.
Area- and individual-level characteristics may also interact: the effects of individual-level variables may differ by contextual characteristics, and the effects of contextual characteristics may differ by individual-level variables. For example, gradients by individual-level income may be stronger in poor neighborhoods (where those with low incomes are unable to gain access to resources outside the neighborhood) than in rich neighborhoods (where the comparative advantage conferred by high income is not as great). Alternatively, if increased individual-level income confers little advantage in the presence of important neighborhood deprivation, the effects of individual-level social position may be stronger in richer neighborhoods. Although a few studies have investigated interactions between neighborhood socioeconomic characteristics and individual-level social class indicators,22,28,83–85 results have not been fully consistent regarding the types of interactions present.
Interactions between area characteristics and other individual-level factors have been less commonly investigated. The articles by Rauh et al.29 and Pearl et al.30 are of special interest because of the specific interaction hypotheses they investigate. For example, Rauh et al.29 examine whether the rise in reproductive risk with maternal age differs by community characteristics but find no clear evidence of an interaction. Pearl et al.30 examine whether the effects of neighborhood characteristics on birthweight differ by race/ethnicity. Their results suggest that the meanings, and implications for health, of different neighborhood environments may differ across ethnic groups. Development and testing of specific hypotheses regarding interactions may help enhance our understanding of the processes involved.
Investigating the relation between areas or neighborhoods and health may require using several different research designs. Quantitative studies need to include data on both individuals and the areas in which they live, as do the studies by Rauh et al.29 and Pearl et al.30 Ideally, they should include sufficient numbers of neighborhoods and sufficient numbers of individuals per neighborhood to allow examination of within- and between-neighborhood variability in the outcomes and in the factors associated with them.86,87 Ensuring sufficient range in the types of neighborhoods included is also important.49
People change neighborhoods over their life course, and neighborhoods themselves may also change over time.73,74 Although several longitudinal studies have investigated the relation between neighborhood characteristics and mortality or incidence of disease,7,10,11,20,22 most research has relied on measurement of neighborhood environments at one point in time. The cumulative or interacting effects of neighborhood environments measured at different times over the life course, the effects of duration of exposure to certain neighborhood conditions, the effects of changes over time in neighborhood characteristics, and the impact of moving from one neighborhood to another have not been systematically examined. Rauh et al.29 allude to the need for studies that explore residential history or patterns of exposure to various community conditions in examining community effects on low birthweight. Investigation of these longitudinal and life-course dimensions will require study designs that follow both individuals and neighborhoods over time.
Studies linking census data for areas to individual-level data on health outcomes and covariates will continue to be of use. The studies by Rauh et al.29 and Pearl et al.30 illustrate some of the possibilities of these data linkages. The systematic way in which area data are collected for the entire population makes census-based measures a valuable resource, despite limitations of the geographic areas available and the absence of direct measures of potentially important neighborhood-level processes. Studies using census-based data may be especially useful if they examine aspects that have been infrequently addressed in the past, such as the relation of neighborhood socioeconomic context to potential mediators of neighborhood effects, longitudinal and life course aspects, and theoretically driven interaction hypotheses. In some cases, it may be possible to link census data to other local data sources with information on area resources or other characteristics. However, a fuller understanding of the effects of neighborhood environments on health will require collection of new data on specific area or neighborhood attributes and studies specially designed to test hypotheses regarding the processes through which neighborhood or area effects may be mediated.
Combinations of quantitative and qualitative approaches may be especially useful.74,88 There is a long history of ethnographic studies of how neighborhoods influence individuals within them (e.g., see reference75). Qualitative studies may be helpful in elucidating the processes involved as well as the dynamic interactions between area and individual characteristics, which may be difficult or impossible to examine via purely quantitative approaches.88 The combination of smaller scale, in-depth approaches (qualitative and quantitative) focusing on a few contrasting neighborhoods and large-scale analyses of routinely available quantitative data on a large sample spanning a broader range of neighborhoods89 is a promising area.
It may also be possible to take advantage of intervention studies to answer questions regarding area or neighborhood effects on health. A fruitful area may be the evaluation of policies or interventions targeted at improving neighborhood environments generally (often with objectives unrelated to health) with respect to their impact on health outcomes, as suggested by child development researchers.76
Most people are keenly aware in their daily lives of the many tangible and intangible benefits of living in a “good” as opposed to a “bad” neighborhood. Investigation of whether and how neighborhood factors affect health may lead to more effective public health strategies. However, inclusion of “neighborhood-level” factors in epidemiology is also plagued with potential pitfalls. These pitfalls have to do with unintended reductionisms or simplifications in the ways in which “neighborhoods” are incorporated into epidemiologic models of disease causation and quantitative analyses.
First, in studying neighborhood effects on health, it is important to place “neighborhoods” themselves within their broader context. Differences across areas or neighborhoods are the result of macrostructural factors shaping residential segregation (e.g., economic restructuring, migration, discrimination, political decisions and public policies, institutional policies). Neighborhood differences thus need to be linked to their upstream determinants. Moreover, the degree to which neighborhoods are important to health may differ, depending on the broader social context. The importance of neighborhood or area differences, for example, may be small in comparison with the overwhelming health effects of societywide factors (such as trends in the economy, restructuring of work, and mass production of foods).
The effects of neighborhoods may also be small in comparison with the individual-level effect of being a member of a discriminated group, having a low income, being uneducated, or having an unskilled job in an unequal society. Neighborhoods have become the “contexts” of choice in epidemiology in part because of the availability of administrative data that can be used to characterize them. But many other “contexts” (e.g., families, peer groups, other social groups or communities that are not necessarily geographically bounded) may be more relevant than neighborhoods for some individuals or some outcomes.
Second, as noted earlier, “neighborhoods” are only one of a number of nested and overlapping geographic areas (or levels) that may have implications for health. Properties defined at these different levels (including physical, resource, social, and policy features) may be related to health outcomes. Moreover, just as individuals are interacting and interdependent parts of social groups,90 “neighborhoods” (as well as other geographically defined areas) are interdependent and interacting parts within larger wholes. For example, neighborhoods may play different roles within the social and economic structure of a city, and health-related differences across neighborhoods may be partly shaped by how neighborhoods relate to each other within the larger city structure. The presence of multiple levels, as well as the roles of dynamic interactions within and between levels, is a challenge in the investigation of “neighborhood” effects as it is for epidemiology generally.91
Third, use of the term neighborhoods may often unintentionally conjure up a somewhat idealized notion of independent and socially cohesive neighborhood communities. The conceptualization of neighborhoods as distinct and independent entities with properties that are somehow under the “neighborhood's” (or residents') control may lead to “blame the neighborhood” explanations analogous to “blame the victim” explanations for the causes of diseases in individuals. In addition, in emphasizing the social attributes of neighborhoods, we should not forget that neighborhoods are fundamentally places that exist in physical space and often differ markedly in environmental, service, and resource characteristics.
Finally, epidemiologists must guard against simplistic explanations that reduce “neighborhoods” or areas to yet another “variable” to be included in the web of causation. Examination of area or neighborhood effects provides a unique opportunity in epidemiology, because it highlights the need to develop models and analytic strategies that incorporate systems defined at multiple levels: regions, cities, neighborhoods, people. Investigating area or neighborhood effects challenges us to theorize and investigate how large-scale macro processes—processes determining residential segregation, for example—ultimately trickle down to people and their bodies without ignoring the interdependencies and mutual influences between people, between places, and between people and the places in which they live.
This work was supported by grant R29 HL59386 from the National Heart, Lung and Blood Institute of the National Institutes of Health.
Many thanks to Bruce Link for helpful comments on an earlier version and to Sharon Stein Merkin for assistance with references.
