Investigating Neighborhood and Area Effects on Health

American Journal of Public Health | November 2001, Vol 91, No. 11 Diez Roux | Peer Reviewed | Commentary | 1783

COMMENTARY

Investigating Neighborhood and Area Effects on HealthAna V. Diez Roux, MD, PhDThe past few years have witnessed

an explosion of interest in neighbor-hood or area effects on health. Sev-eral types of empiric studies havebeen used to examine possible areaor neighborhood effects, includingecologic studies relating area char-acteristics to morbidity and mortal-ity rates, contextual and multilevelanalyses relating area socioeconomiccontext to health outcomes, andstudies comparing small numbers ofwell-defined neighborhoods.

Strengthening inferences regard-ing the presence and magnitude ofneighborhood effects will require ad-dressing a series of conceptual andmethodological issues.Many of theseissues relate to the need to developtheory and specific hypotheses onthe processes through which neigh-borhood and individual factors mayjointly influence specific health out-comes. Important challenges includedefining neighborhoods or relevantgeographic areas, identifying signifi-cant area or neighborhood charac-teristics, specifying the role of indi-vidual-level variables, incorporatinglife-course and longitudinal dimen-sions,combining a variety of researchdesigns,and avoiding reductionism inthe way in which“neighborhood” fac-tors are incorporated into models ofdisease causation and quantitativeanalyses. (Am J Public Health. 2001;91:1783–1789)

ALTHOUGH INTEREST INgeographical variations in healthhas a long history,1 the impor-tance given to the examination ofarea differences in studying thecauses of disease has varied overthe years. With few exceptions(e.g., see references 2–4), thefocus on individual-level risk fac-tors over the past few decadeswas generally associated with lit-tle interest in area characteristicsas potential disease determinants.However, the notion that “place”may be important to healthreemerged in a handful of publi-cations in the 1980s and early1990s,5–9 and interest has in-creased sharply in recent years.Numerous reports on area orneighborhood differences inhealth have recently appeared inepidemiology and public healthjournals (e.g., see references10–28). In this issue of the Jour-nal, Rauh et al.29 and Pearl etal.30 examine the role of neigh-borhood or community contextsin shaping the distribution of lowbirthweight.

Several factors have convergedto stimulate a resurgence of in-terest in area or neighborhoodhealth effects. Chief among thesefactors has been a rekindling ofinterest in the social determi-nants of health31,32 and therecognition that social influenceson health operate through manydifferent processes, one of whichmay be the types of areas orneighborhoods in which peoplelive.33–35 Simultaneously, there

has been a growing discussion ofthe use of ecologic variables inepidemiology.36–38 This discus-sion is related to a critique of thenotion that all health determi-nants are best conceptualized asindividual-level attributes. Re-search on neighborhood effectshas fit into this emerging para-digm because it has postulatedthat neighborhood contexts maybe related to health, independ-ently of individual-level attrib-utes. In addition, the emergenceof new methodological ap-proaches such as multilevel anal-ysis39–42 has stimulated boththinking and empiric research inthis field.

Sociologists and social geogra-phers have long recognized theimportance of neighborhood en-vironments as structural condi-tions that shape individual livesand opportunities.43,44 In theUnited States, recent discussionsin sociology on the causes andconsequences of residential seg-regation and urban poverty45–47

have reinvigorated interest in thesociologic and ethnographic in-vestigation of neighborhoods andthe ways in which neighborhoodcontexts may affect individuals.Social scientists have examinedhow living in areas of concen-trated poverty influences individ-ual-level outcomes such asemployment and single parent-hood.43 Neighborhood contexthas been investigated in relationto violence,48 child develop-ment,49,50 and childbearing prac-

tices.51 In public health, it hasbeen argued that the physicaland social environments ofneighborhoods may be importantin understanding the distributionof health outcomes.33–35,52,53

Investigation of how neighbor-hood environments may be re-lated to health is not only of aca-demic interest. Documentation ofneighborhood effects, as well aselucidation of the mechanismsthrough which they are medi-ated, could have important policyimplications for health promotionand for the reduction of healthdisparities. Neighborhood differ-ences may be especially relevantin the context of increasing spa-tial concentrations of poverty andincreasing geographic clusteringof poverty with other forms ofdisadvantage.54 The articles byRauh et al.29 and Pearl et al.30 il-lustrate some of the key concep-tual and methodological chal-lenges faced by researchers asthey investigate neighborhoodhealth effects.

INVESTIGATINGNEIGHBORHOOD EFFECTSON HEALTH

Basically, 3 empiric strategieshave been used to investigateneighborhood or area effects:ecologic studies, contextual ormultilevel studies, and compar-isons of small numbers of well-defined neighborhoods. Ecologicstudies have been used to exam-ine variations in morbidity and

American Journal of Public Health | November 2001, Vol 91, No. 111784 | Commentary | Peer Reviewed | Diez Roux

COMMENTARY

mortality rates across areas andto relate this variability to areacharacteristics. The sizes of areasexamined have varied greatly,ranging from relatively largeareas not really analogous toneighborhoods at all (e.g., coun-ties in the United States or dis-trict health authorities in theUnited Kingdom)55–58 to smallerareas (e.g., census tracts in theUnited States or wards in theUnited Kingdom).59–62 The mostcommon area characteristics in-vestigated have been aggregatemeasures of the socioeconomiccharacteristics of residents or in-dices of deprivation constructedby combining several aggregatemeasures based on theoretic orempiric considerations.63 Thesearea characteristics have beenfound to be strongly related tomortality rates (e.g., see refer-ences 55, 56, 58, 61, and 62).

However, the use of such ag-gregate measures has sometimesgenerated ambiguity regardingwhether these variables are con-ceptualized as measures of area-level properties or simply as sum-maries of individual-levelvariables, and hence whether theobjective is to examine how areaconstructs are related to healthoutcomes or to document thearea-level (or ecologic) expres-sion of a well-known individual-level relation. Although ecologicstudies may be useful to docu-ment and monitor inequalities inhealth, they cannot directly de-termine whether differencesacross areas are due to character-istics of the areas themselves orto differences between the typesof individuals living in differentareas. They cannot evaluate therole of individual-level factors asconfounders, mediators, or modi-fiers of the area effect.

The recognition of the need todistinguish the effects of “con-

text” (e.g., area or group proper-ties) and “composition” (charac-teristics of individuals living indifferent areas) when examiningarea effects on health39 has ledto a proliferation of reports in-volving contextual and multilevelanalyses. The articles by Rauh etal.29 and Pearl et al.30 employthis analytic approach. Contex-tual and multilevel analyses re-quire data sets including individ-uals nested within areas orneighborhoods. By simultane-ously including both neighbor-hood- and individual-level pre-dictors in regression equationswith individuals as the units ofanalysis, these strategies allowexamination of neighborhood orarea effects after individual-levelconfounders have been con-trolled. They also permit exami-nation of individual-level charac-teristics as modifiers of the areaeffect (and vice versa).

Furthermore, multilevel analy-sis allows the simultaneous exam-ination of within- and between-neighborhood variability inoutcomes64,65 and of the extentto which between-neighborhoodvariability is “explained” by indi-vidual- and neighborhood-levelfactors. Studies using these ap-proaches have usually linked in-formation on small-area charac-teristics available in censuses toindividual-level covariate and out-come data from surveys, epidemi-ologic studies, or vital statistics (asin the 2 studies in this issue). Forthe most part, contextual andmultilevel studies have been con-sistent in documenting an “inde-pendent” effect of neighborhoodsocioeconomic environment onindividual-level outcomes aftercontrolling for individual-level so-cioeconomic position indica-tors.10–28,52,53 However, the per-centage of total variability inoutcomes between areas has

often been small,6,35,66–68 and thestrength (and relative importance)of the neighborhood or area ef-fect is still under debate.52,53

In contrast to the large-scalequantitative approaches justsummarized, an alternative strat-egy has been to compare a smallnumber of well-defined and pur-posely selected contrastingneighborhoods.33,69–72 Thesetypes of studies can incorporateknowledge on local history, soci-ology, and geography in definingneighborhoods. In addition, theycan directly collect detailed in-formation on neighborhoodcharacteristics and health out-comes through combinations ofquantitative and qualitativestrategies.33 This approach hasbeen used to document differ-ences across neighborhoods inresources and services and to re-late these differences to differ-ences in health behaviors.33,69–72

However, it is limited in thenumber and range of neighbor-hoods investigated and possiblyin the generalizability of results.Its strength resides in the use oflocally based definitions ofneighborhoods (rather than ad-ministrative proxies) and in thedirect collection of data onneighborhood characteristics,which may help provide an un-derstanding of the processesthrough which neighborhood en-vironments can affect health.

CONCEPTUAL ANDMETHODOLOGICALCHALLENGES

As in other fields,49 strength-ening inferences regarding thepresence and relative importanceof neighborhood or area effectswill require addressing a series ofconceptual and methodologicalissues. Many of these issues stemfrom the need to develop theo-

ries and more specific hypotheseson the dynamic processesthrough which neighborhoodand individual factors may jointlyinfluence specific outcomes.73

Defining “Neighborhoods” orRelevant Geographic Areas

A first issue is the definition of“neighborhoods” or, perhapsmore precisely, definition of thegeographic area whose character-istics may be relevant to the spe-cific health outcome being stud-ied. In health research, the termsneighborhood and communityhave often been used loosely torefer to a person’s immediate res-idential environment, which ishypothesized to have both mate-rial and social characteristics po-tentially related to health. Themore generic term area has alsobeen used. Clear distinctions be-tween the terms neighborhood,community, and area are usuallynot made. Administratively de-fined areas have been used asrough proxies for “neighbor-hoods” or “communities” inmany studies, including the 2 in-cluded in this issue. Rauh et al.29

base their analyses of commu-nity-level variables on New YorkCity–defined health areas (clus-ters of 4–6 census tracts). Pearlet al.30 base their analyses ofneighborhood-level variables onthe smaller census-defined blockgroups.

The concepts of “neighbor-hood” and “community” are notprecise.54,73,74 Although both aregeographically anchored (theterm community is used to implic-itly refer to geographically an-chored communities in thesetypes of analyses), there are sev-eral possible definitions, depend-ing on the criteria used. Criteriacan be historical, based on peo-ple’s characteristics, based on ad-ministrative boundaries, or based


COMMENTARY

on people’s perceptions. Bound-aries based on these different cri-teria will not necessarilyoverlap,54 and alternative defini-tions may be relevant for differ-ent research questions.73 For ex-ample, neighborhoods definedon the basis of people’s percep-tions may be relevant when theneighborhood characteristics ofinterest relate to social interac-tions or social cohesion, adminis-tratively defined neighborhoodsmay be relevant when the hy-pothesized processes involve poli-cies, and geographically definedneighborhoods may be relevantwhen features of the chemical orphysical environment (e.g., toxicexposures) are hypothesized tobe important.

More generally, the size anddefinition of the relevant geo-graphic area may vary accordingto the processes through whichthe area effect is hypothesized tooperate and the outcome beingstudied.54 Areas ranging fromsmall to large with varying geo-graphic definitions may be im-portant for different health out-comes or for different mediatingmechanisms, and many of these“areas” may not be thought of as“neighborhoods” or “communi-ties” in the common sense at all.For example, counties may beimportant geographic contextsfor outcomes potentially relatedto county policies or economicstructures. School districts maybe relevant for child outcomes.For some purposes, the relevantarea may be the block on whicha person resides; for others, itmay be the blocks around theresidence; and for still others, itmay be the geographic area inwhich services such as stores orother institutions are located.54

The size and definition of thearea, the relevant processes, andthe outcome being studied are

linked. The development andtesting of hypotheses regardingthe precise geographic area thatis relevant for a specific healthoutcome will help strengthen in-ferences regarding area effects.

Research also requires opera-tionalizing the theoretically rele-vant “areas” so that data on indi-viduals and area attributes canbe linked to them. Investigatorsconducting large quantitativeanalyses spanning many areasoften have no choice but to relyon existing administrative defini-tions for which standard data areavailable and to which individu-als and other sources of areadata can be easily linked. Al-though the use of these proxies isthe only practical alternative inmany cases, they are obviouslylimited in that they do not neces-sarily correspond to the theoreti-cally relevant area. Another al-ternative involves definingrelevant areas or neighborhoodson the basis of local historical,social, and geographic knowledgeas well as information collectedfrom residents. Development ofthese boundaries is an enormoustask in itself and may be feasibleonly in studies focusing on a lim-ited geographic region or city.The challenges involved in oper-ationalizing relevant areas orneighborhoods are important butshould not paralyze thinking orempiric investigation. Compar-isons of alternative operational-izations with respect to their con-sequences for research andinterdisciplinary work with geog-raphers, urban planners, and so-cial scientists will be important.

Specifying the RelevantNeighborhood or AreaCharacteristics

Most existing research, includ-ing the articles by Rauh et al.29

and Pearl et al.,30 has examined

how aggregate measures ofneighborhood socioeconomiccontext are related to health out-comes.52 These associations arecompatible with a wide range ofprocesses relating neighborhoodenvironments to health. BothRauh et al.29 and Pearl et al.30

discuss some of the processesthat could mediate effects ofneighborhood or communitycontext on birthweight. A nextstep (and a way of establishingthe extent to which neighbor-hoods or communities are or arenot relevant) will be the more di-rect empiric examination of thespecific features of areas (be theyphysical or social) that may berelated to different outcomes.33

Examining the role of specificneighborhood or area character-istics is complex, because manyof these dimensions may beinterrelated (and thus difficult totease apart)75 and may also influ-ence each other.33 For example,features of the physical environ-ments of neighborhoods may in-fluence the types of social inter-actions, and vice versa. Inaddition, the processes involvedand the relevant neighborhoodattributes may differ from oneoutcome to another. For exam-ple, mechanisms involving re-sources and the physical environ-ment may be more relevant forcertain outcomes (e.g., physicalactivity), whereas those involvingsocial norms or contagion pro-cesses may be more importantfor others (e.g., smoking). Livingin a deprived area may itself be asource of stress that may be par-ticularly relevant for stress-re-lated outcomes.

Some hypothesized processesmay predict linear effects ofneighborhood characteristics,whereas others may predictthreshold effects. Hypotheses in-volving social comparisons (or

relative deprivation hypotheses)may predict different things, de-pending on the group withinwhich the comparison occurs.For example, if a person’s per-ception of his or her position rel-ative to others in a given area isimportant, then being the “best-off” person living in a disadvan-taged area could be associatedwith a better outcome than beingthe “worst-off” person residing ina better area. Other patternsmight be predicted if relevant so-cial comparisons are madeacross, rather than within, areas.

From the operational point ofview, measurement of specificcharacteristics of neighborhoodsis complex. Options for the col-lection of this type of informationinclude surveys of residents(which may be aggregated up tothe desired area level) on objec-tive and subjective characteristicsof their neighborhoods, directobservation or videotaping andranking of neighborhoods onprespecified criteria by raters(systematic social observation),and linking databases with geo-graphically linked information(e.g., from public agencies) andestimating density and distancemeasures.76–78 Assessment ofneighborhoods or areas presentsa series of methodological chal-lenges related to the measure-ment of ecologic settings (termed“ecometrics”77) that have only re-cently begun to be addressed.The development of valid and re-liable measures of relevant areacharacteristics that can be ob-tained in a systematic fashionacross many areas is an impor-tant need in this field.79

Role of Individual-LevelVariables

A crucial issue in the examina-tion of neighborhood effects ishow individual-level variables


COMMENTARY

should be incorporated into con-ceptual models and included inanalyses. The most common crit-icism of postulated neighborhoodeffects is that they result fromconfounding by individual-levelvariables.80,81 The selection prob-lem73,82 is a variant of this issue:people may be sorted into neigh-borhoods according to individualcharacteristics, and these individ-ual characteristics may be relatedto outcomes. As a way to re-spond to these critiques, studieshave attempted to control for in-dividual-level variables, mostcommonly indicators of socialposition.52 Both Rauh et al.29 andPearl et al.30 make a special pointof controlling for individual-levelsocial position indicators viastratification and multivariate ad-justment. Short of randomization,these adjustment strategies arethe best way to analytically dem-onstrate an effect of neighbor-hood or area context on health.Studies demonstrating this “inde-pendent” effect are therefore anecessary and valuable first step.

However, the relation betweenneighborhood characteristics andindividual-level socioeconomicposition is a complex one. To theextent that neighborhoods influ-ence the life chances of individu-als,43,45,82 neighborhood socialand economic characteristicsmay be related to health throughtheir effects on achieved income,education, and occupation, mak-ing these individual-level charac-teristics mediators (at least inpart) rather than confounders. Inaddition, because socioeconomicposition is one of the dimensionsalong which residential segrega-tion occurs, living in disadvan-taged neighborhoods may be oneof the mechanisms leading to ad-verse health outcomes in personsof low socioeconomic status. Forthese reasons, although teasing

apart the “independent” effectsof both dimensions may be use-ful as part of the analytic process,it is also artificial. A related issue(which pertains to investigationof contextual effects generally) isthat contexts are often under-specified (particularly relative toindividual-level constructs, whichare usually much easier to defineand measure), making compar-isons of relative magnitudes ofeffects problematic and often bi-ased against contextual effects.

Of course, because disease isexpressed at the level of the indi-vidual, neighborhood factors nec-essarily exert their effect throughindividual-level processes, includ-ing behaviors and biological pre-cursors of disease. Whether anindividual-level variable is con-ceptualized as a confounder or amediator depends on the ques-tion being asked. Both studies inthis issue adjust for individual-level variables (e.g., smoking andsubstance abuse in Rauh et al.29

and prenatal care initiation inPearl et al.30) that could be con-ceptualized as confounders ormediators, depending on the hy-pothesis being tested. Furthercomplexity results from the factthat, in some cases, neighbor-hood and individual characteris-tics may mutually influence eachother. For example, the availabil-ity of healthy foods in a neigh-borhood may influence the di-etary behaviors of individuals,and individual behaviors may inturn affect food availability. Un-derstanding area or neighbor-hood effects may require the test-ing of hypotheses involving suchdynamic and reciprocal relations.The methods commonly used inepidemiology today are not wellsuited to examination of these re-ciprocal and dynamic relations.

Area- and individual-levelcharacteristics may also interact:

the effects of individual-levelvariables may differ by contex-tual characteristics, and the ef-fects of contextual characteristicsmay differ by individual-levelvariables. For example, gradientsby individual-level income maybe stronger in poor neighbor-hoods (where those with low in-comes are unable to gain accessto resources outside the neigh-borhood) than in rich neighbor-hoods (where the comparativeadvantage conferred by high in-come is not as great). Alterna-tively, if increased individual-level income confers littleadvantage in the presence of im-portant neighborhood depriva-tion, the effects of individual-level social position may bestronger in richer neighborhoods.Although a few studies have in-vestigated interactions betweenneighborhood socioeconomiccharacteristics and individual-level social class indica-tors,22,28,83–85 results have notbeen fully consistent regardingthe types of interactions present.

Interactions between areacharacteristics and other individ-ual-level factors have been lesscommonly investigated. The arti-cles by Rauh et al.29 and Pearl etal.30 are of special interest be-cause of the specific interactionhypotheses they investigate. Forexample, Rauh et al.29 examinewhether the rise in reproductiverisk with maternal age differs bycommunity characteristics butfind no clear evidence of an in-teraction. Pearl et al.30 examinewhether the effects of neighbor-hood characteristics on birth-weight differ by race/ethnicity.Their results suggest that themeanings, and implications forhealth, of different neighborhoodenvironments may differ acrossethnic groups. Development andtesting of specific hypotheses re-

garding interactions may help en-hance our understanding of theprocesses involved.

Appropriate Study DesignsInvestigating the relation be-

tween areas or neighborhoodsand health may require usingseveral different research de-signs. Quantitative studies needto include data on both individu-als and the areas in which theylive, as do the studies by Rauh etal.29 and Pearl et al.30 Ideally,they should include sufficientnumbers of neighborhoods andsufficient numbers of individualsper neighborhood to allow exam-ination of within- and between-neighborhood variability in theoutcomes and in the factors asso-ciated with them.86,87 Ensuringsufficient range in the types ofneighborhoods included is alsoimportant.49

People change neighborhoodsover their life course, and neigh-borhoods themselves may alsochange over time.73,74 Althoughseveral longitudinal studies haveinvestigated the relation betweenneighborhood characteristics andmortality or incidence ofdisease,7,10,11,20,22 most researchhas relied on measurement ofneighborhood environments atone point in time. The cumula-tive or interacting effects ofneighborhood environmentsmeasured at different times overthe life course, the effects of du-ration of exposure to certainneighborhood conditions, the ef-fects of changes over time inneighborhood characteristics,and the impact of moving fromone neighborhood to anotherhave not been systematically ex-amined. Rauh et al.29 allude tothe need for studies that exploreresidential history or patterns ofexposure to various communityconditions in examining commu-


COMMENTARY

nity effects on low birthweight.Investigation of these longitudi-nal and life-course dimensionswill require study designs thatfollow both individuals andneighborhoods over time.

Studies linking census data forareas to individual-level data onhealth outcomes and covariateswill continue to be of use. Thestudies by Rauh et al.29 and Pearlet al.30 illustrate some of the pos-sibilities of these data linkages.The systematic way in whicharea data are collected for theentire population makes census-based measures a valuable re-source, despite limitations of thegeographic areas available andthe absence of direct measures ofpotentially important neighbor-hood-level processes. Studiesusing census-based data may beespecially useful if they examineaspects that have been infre-quently addressed in the past,such as the relation of neighbor-hood socioeconomic context topotential mediators of neighbor-hood effects, longitudinal and lifecourse aspects, and theoreticallydriven interaction hypotheses. Insome cases, it may be possible tolink census data to other localdata sources with information onarea resources or other charac-teristics. However, a fuller under-standing of the effects of neigh-borhood environments on healthwill require collection of newdata on specific area or neigh-borhood attributes and studiesspecially designed to test hy-potheses regarding the processesthrough which neighborhood orarea effects may be mediated.

Combinations of quantitativeand qualitative approaches maybe especially useful.74,88 There isa long history of ethnographicstudies of how neighborhoods in-fluence individuals within them(e.g., see reference 75). Qualita-

tive studies may be helpful inelucidating the processes in-volved as well as the dynamic in-teractions between area and indi-vidual characteristics, which maybe difficult or impossible to ex-amine via purely quantitative ap-proaches.88 The combination ofsmaller scale, in-depth ap-proaches (qualitative and quanti-tative) focusing on a few con-trasting neighborhoods andlarge-scale analyses of routinelyavailable quantitative data on alarge sample spanning a broaderrange of neighborhoods89 is apromising area.

It may also be possible to takeadvantage of intervention studiesto answer questions regardingarea or neighborhood effects onhealth. A fruitful area may be theevaluation of policies or interven-tions targeted at improvingneighborhood environments gen-erally (often with objectives unre-lated to health) with respect totheir impact on health outcomes,as suggested by child develop-ment researchers.76

PITFALLS ANDPOTENTIALITIES

Most people are keenly awarein their daily lives of the manytangible and intangible benefitsof living in a “good” as opposedto a “bad” neighborhood. Investi-gation of whether and howneighborhood factors affecthealth may lead to more effec-tive public health strategies.However, inclusion of “neighbor-hood-level” factors in epidemiol-ogy is also plagued with potentialpitfalls. These pitfalls have to dowith unintended reductionismsor simplifications in the ways inwhich “neighborhoods” are incor-porated into epidemiologic mod-els of disease causation andquantitative analyses.

First, in studying neighbor-hood effects on health, it is im-portant to place “neighborhoods”themselves within their broadercontext. Differences across areasor neighborhoods are the resultof macrostructural factors shap-ing residential segregation (e.g.,economic restructuring, migra-tion, discrimination, political de-cisions and public policies, insti-tutional policies). Neighborhooddifferences thus need to belinked to their upstream determi-nants. Moreover, the degree towhich neighborhoods are impor-tant to health may differ, de-pending on the broader socialcontext. The importance ofneighborhood or area differ-ences, for example, may be smallin comparison with the over-whelming health effects of soci-etywide factors (such as trends inthe economy, restructuring ofwork, and mass production offoods).

The effects of neighborhoodsmay also be small in comparisonwith the individual-level effect ofbeing a member of a discrimi-nated group, having a low in-come, being uneducated, or hav-ing an unskilled job in anunequal society. Neighborhoodshave become the “contexts” ofchoice in epidemiology in partbecause of the availability of ad-ministrative data that can beused to characterize them. Butmany other “contexts” (e.g., fami-lies, peer groups, other socialgroups or communities that arenot necessarily geographicallybounded) may be more relevantthan neighborhoods for some in-dividuals or some outcomes.

Second, as noted earlier,“neighborhoods” are only one ofa number of nested and overlap-ping geographic areas (or levels)that may have implications forhealth. Properties defined at

these different levels (includingphysical, resource, social, andpolicy features) may be related tohealth outcomes. Moreover, justas individuals are interacting andinterdependent parts of socialgroups,90 “neighborhoods” (aswell as other geographically de-fined areas) are interdependentand interacting parts withinlarger wholes. For example,neighborhoods may play differ-ent roles within the social andeconomic structure of a city, andhealth-related differences acrossneighborhoods may be partlyshaped by how neighborhoodsrelate to each other within thelarger city structure. The pres-ence of multiple levels, as well asthe roles of dynamic interactionswithin and between levels, is achallenge in the investigation of“neighborhood” effects as it is forepidemiology generally.91

Third, use of the term neigh-borhoods may often unintention-ally conjure up a somewhatidealized notion of independentand socially cohesive neighbor-hood communities. The concep-tualization of neighborhoods asdistinct and independent entitieswith properties that are some-how under the “neighborhood’s”(or residents’) control may leadto “blame the neighborhood” ex-planations analogous to “blamethe victim” explanations for thecauses of diseases in individuals.In addition, in emphasizing thesocial attributes of neighbor-hoods, we should not forget thatneighborhoods are fundamen-tally places that exist in physicalspace and often differ markedlyin environmental, service, and re-source characteristics.

Finally, epidemiologists mustguard against simplistic explana-tions that reduce “neighbor-hoods” or areas to yet another“variable” to be included in the


COMMENTARY

web of causation. Examination ofarea or neighborhood effects pro-vides a unique opportunity inepidemiology, because it high-lights the need to develop mod-els and analytic strategies that in-corporate systems defined atmultiple levels: regions, cities,neighborhoods, people. Investi-gating area or neighborhood ef-fects challenges us to theorizeand investigate how large-scalemacro processes—processes de-termining residential segregation,for example—ultimately trickledown to people and their bodieswithout ignoring the interdepen-dencies and mutual influencesbetween people, between places,and between people and theplaces in which they live.

About the AuthorThe author is with the Division of GeneralMedicine, Columbia College of Physiciansand Surgeons, and the Division of Epide-miology, Joseph T. Mailman School ofPublic Health, Columbia University, NewYork City.

Requests for reprints should be sent toAna V. Diez Roux, MD, PhD, Division ofGeneral Medicine, Columbia PresbyterianMedical Center, Ph 9 East Room 105,622 W 168th St, New York, NY 10032(e-mail: [email protected]).

This commentary was accepted Janu-ary 24, 2001.

AcknowledgmentsThis work was supported by grant R29HL59386 from the National Heart,Lung and Blood Institute of the NationalInstitutes of Health.

Many thanks to Bruce Link for help-ful comments on an earlier version andto Sharon Stein Merkin for assistancewith references.

References1. Barrett FA. Finke’s 1792 map ofhuman diseases: the first world diseasemap? Soc Sci Med. 2000;50:915–921.

2. Harbug E, Erfurt J, Chape C,Hauenstein LS, Schull WJ, Schork MA.Socioecological stressor areas and black-white blood pressure. J Chronic Dis.1973;26:595–611.

3. James SA, Kleinbaum DG. Socio-ecologic stress and hypertension relatedmortality rates in North Carolina. Am JPublic Health. 1976;66:354–358.

4. Kasl SV, Harburg E. Mental healthand the urban environment: somedoubts and second thoughts. J HealthSoc Behav. 1975;16:268–282.

5. Carstairs V, Morris R. Deprivationand mortality. An alternative to socialclass? Community Med. 1989;11:210–219.

6. Diehr P, Koepsell T, Cheadle A,Psaty BM, Wagner E, Curry S. Do com-munities differ in health behaviors?J Clin Epidemiol. 1993;46:1141–1149.

7. Haan M, Kaplan G, Camacho T.Poverty and health: prospective evi-dence from the Alameda County Study.Am J Epidemiol. 1987;125:989–998.

8. Humphreys K, Carr-Hill R. Areavariations in health outcomes: artefactor ecology. Int J Epidemiol. 1991;20:251–258.

9. Krieger N. Overcoming the ab-sence of socioeconomic data in medicalrecords: validation and application of acensus-based methodology. Am J PublicHealth. 1992;82:703–710.

10. Anderson R. Mortality effects ofcommunity socioeconomic status. Epide-miology. 1997;8:42–47.

11. Davey Smith G, Hart C, Watt G,Hole D, Hawthorne V. Individual socialclass, area-based deprivation, cardiovas-cular disease risk factors, and mortality:the Renfrew and Paisley study. J Epi-demiol Community Health. 1998;52:399–405.

12. Diez-Roux A, Nieto F, Muntaner C,et al. Neighborhood environments andcoronary heart disease: a multilevelanalysis. Am J Epidemiol. 1997;146:48–63.

13. Kleinschmidt I, Hills M, Elliott P.Smoking behavior can be predicted byneighborhood deprivation measures.J Epidemiol Community Health. 1995;49(suppl):S72–S77.

14. Leclere F, Rogers R, Peters K.Neighborhood social context and racialdifferences in women’s heart diseasemortality. J Health Soc Behav. 1998;39:91–107.

15. Matteson D, Burr J, Marshall J. In-fant mortality: a multi-level analysis ofindividual and community risk factors.Soc Sci Med. 1998;47:1841–1854.

16. O’Campo P, Xue X, Wang M,Caughy M. Neighborhood risk factorsfor low birthweight in Baltimore: amultilevel analysis. Am J Public Health.1997;87:1113–1118.

17. Shouls S, Congdon P, Curtis S.Modelling inequality in reported longterm illness in the UK: combining indi-vidual and area characteristics. J Epi-demiol Community Health. 1996;50:366–376.

18. Robert S. Community-level socio-economic status effects on adult health.J Health Soc Behav. 1998;39:18–37.

19. Roberts EM. Neighborhood socialenvironments and the distribution oflow birth weight in Chicago. Am J PublicHealth. 1997;87:597–603.

20. Waitzman N, Smith K. Phantom ofthe area: poverty-area residence andmortality in the United States. Am JPublic Health. 1998;88:973–976.

21. Sundquist J, Malmstrom M, Johans-son SE. Cardiovascular risk factors andthe neighbourhood environment: a mul-tilevel analysis. Int J Epidemiol. 1999;28:841–845.

22. Yen I, Kaplan G. Neighborhood so-cial environment and risk of death: mul-tilevel evidence from the AlamedaCounty Study. Am J Epidemiol. 1999;149:898–907.

23. Cubbin C, LeClere FB, Smith GS.Socioeconomic status and injury mortal-ity: individual and neighbourhood de-terminants. J Epidemiol CommunityHealth. 2000;54:517–524.

24. Duncan C, Jones K, Moon G.Smoking and deprivation: are thereneighbourhood effects? Soc Sci Med.1999;48:497–505.

25. Elreedy S, Krieger N, Ryan PB,Sparrow D, Weiss ST, Hu H. Relationsbetween individual and neighborhood-based measures of socioeconomic posi-tion and bone lead concentrationsamong community exposed men. Am JEpidemiol. 1999;150:129–141.

26. Sloggett A, Joshi H. Higher mortal-ity in deprived areas: community orpersonal disadvantage. BMJ. 1994;309:1470–1474.

27. O’Campo P, Gielen A, Faden R,Xue X, Kass N, Wang M. Violence bymale partners against women duringthe childbearing year: a contextual anal-ysis. Am J Public Health. 1995;85:1092–1097.

28. Sloggett A, Joshi H. Deprivation in-dicators as predictors of life events1981–1992 based on the UK ONSLongitudinal Study. J Epidemiol Commu-nity Health. 1998;52:228–233.

29. Rauh VA, Andrews HF, GarfinkelR. The contribution of maternal age toracial disparities in birthweight: a multi-level perspective. Am J Public Health.2001;91:1815–1824.

30. Pearl M, Braveman P, Abrams B.The relationship of neighborhood so-cioeconomic characteristics to birth-weight among five ethnic groups inCalifornia. Am J Public Health. 2001;91:1808–1824.

31. Krieger N. Epidemiology and theweb of causation. Has anyone seen the

spider? Soc Sci Med. 1994;39:887–903.

32. Kaplan GA, Lynch JW. Whitherstudies on the socioeconomic founda-tions of population health? Am J PublicHealth. 1997;87:1409–1411.

33. Macintyre S, Maciver S, Sooman A.Area, class, and health: should we befocusing on places or people? J Soc Pol-icy. 1993;22:213–234.

34. Kaplan G. People and places: con-trasting perspectives on the associationbetween social class and health. Int JHealth Serv. 1996;26:507–519.

35. Jones K, Duncan C. Individualsand their ecologies: analysing the geog-raphy of chronic illness within a multi-level modelling framework. Health Place.1995;1:27–40.

36. Schwartz S. The fallacy of the eco-logical fallacy: the potential misuse of aconcept and its consequences. Am JPublic Health. 1994;84:819–824.

37. Susser M. The logic in ecological, I:the logic of analysis. Am J Public Health.1994;84:825–829.

38. Diez-Roux AV. Bringing contextback into epidemiology: variables andfallacies in multilevel analysis. Am JPublic Health. 1998;88:216–222.

39. Duncan C, Jones K, Moon G. Con-text, composition, and heterogeneity:using multilevel models in health re-search. Soc Sci Med. 1998;46:97–117.

40. DiPrete TA, Forristal JD. Multilevelmodels: methods and substance. AnnuRev Sociol. 1994;20:331–357.

41. Diez-Roux AV. Multilevel analysisin public health research. Annu RevPublic Health. 2000;21:193–221.

42. Von Korff M, Koepsell T, Curry S,Diehr P. Multi-level research in epidemi-ologic research on health behaviors andoutcomes. Am J Epidemiol. 1992;135:1077–1082.

43. Massey DS, Gross AH, Eggers ML.Segregation, the concentration of pov-erty, and the life chances of individuals.Soc Sci Res. 1991;20:397–420.

44. Harvey D. Class structure in a cap-italist society and the theory of residen-tial differentiation. In: Peel R, ChisholmM, Haggett P, eds. Processes in Physicaland Human Geography. London, En-gland: Heinemann Educational BooksLtd; 1975:354–369.

45. Wilson WJ. The Truly Disadvan-taged: The Inner City, the Underclass andUrban Policy. Chicago, Ill: University ofChicago Press; 1987.

46. Massey DS, Eggers ML. The ecol-ogy of inequality: minorities and theconcentration of poverty, 1970–1980.Am J Sociol. 1990;95:1153–1188.


COMMENTARY

47. Jargowsky PA. Poverty and Place:Ghettos, Barrios, and the American City.New York, NY: Russell Sage Foundation;1997.

48. Sampson R, Raudenbush S, EarlsF. Neighborhoods and violent crime: amultilevel study of collective efficacy.Science. 1997;277:918–924.

49. Leventhal T, Brooks-Gunn J. Theneighborhoods they live in: the effectsof neighborhood residence on childand adolescent outcomes. Psychol Bull.2000;126:309–337.

50. Brooks-Gunn J, Duncan GJ, Kle-banov PK, Sealand N. Do neighbor-hoods influence child and adolescentdevelopment? Am J Sociol. 1993;99:353–395.

51. Hogan DP, Kitagawa EM. The im-pact of social status, family structureand neighborhood on the fertility ofblack adolescents. Am J Sociol. 1985;90:825–855.

52. Robert S. Socioeconomic positionand health: the independent contribu-tion of community socioeconomic con-text. Annu Rev Sociol. 1999;25:489–516.

53. Yen IH, Syme SL. The social envi-ronment and health: a discussion of theepidemiologic literature. Annu Rev Pub-lic Health. 1999;20:287–308.

54. Gephart MA. Neighborhoods andcommunities as contexts for develop-ment. In: Brooks-Gunn J, Duncan GJ,Aber JL, eds. Neighborhood Poverty, Vol-ume I: Context and Consequences for Chil-dren. New York, NY: Russell Sage Foun-dation; 1997:1–43.

55. Wing S, Barnett E, Casper M, Ty-roler HA. Geographic and socioeco-nomic variation in the onset of declineof coronary heart disease mortality inwhite women. Am J Public Health. 1992;82:204–209.

56. Tyroler HA, Wing S, Knowles MG.Increasing inequality in coronary heartdisease mortality in relation to educa-tional achievement: profile of places ofresidence, United States, 1962–87. AnnEpidemiol. 1993;3(suppl):S51–S54.

57. Raleigh VS, Kiri VA. Life expect-ancy in England: variations and trendsby gender, health authority and level ofdeprivation. J Epidemiol CommunityHealth. 1997;51:649–658.

58. Morris JN, Blane DB, White IR.Levels of mortality, education, and so-cial conditions in the 107 local educa-tion authority areas of England. J Epi-demiol Community Health. 1996;50:15–17.

59. Briggs R, Leonard W. Mortality andecological structure: a canonical ap-proach. Soc Sci Med. 1977;11:757–762.

60. Paul-Shaheen P, Deane J, WilliamsD. Small area analysis: a review andanalysis of the North American litera-ture. J Health Polit Policy Law. 1987;12:741–809.

61. Eames M, Ben-Shlomom Y, Mar-mot MG. Social deprivation and prema-ture mortality: regional comparisonacross England. BMJ. 1993;307:1097–1102.

62. Townsend P, Phillimore P, BeattieA. Health and Deprivation. Inequality andthe North. London, England: Routledge;1988.

63. Morris R, Carstairs V. Which depri-vation? A comparison of selected depri-vation indices. J Public Health Med.1991;13:318–326.

64. Bullen N, Jones K, Duncan C. Mod-eling complexity: analyzing between-individual and between-place variation—a multilevel tutorial. Environ Plann.1997;29:585–609.

65. Snijders TA, Bosker RJ. Modeledvariance in two-level models. SociolMethods Res. 1994;22:342–363.

66. Boyle MH, Willms JD. Place effectsfor areas defined by administrativeboundaries. Am J Epidemiol. 1999;149:577–585.

67. Ecob R. A multilevel modelling ap-proach to examining the effects of areaof residence on health. J R Stat Soc A.1996;159:61–75.

68. Hart C, Ecob R, Davey Smith G.People, places, and coronary heart dis-ease risk factors: a multilevel analysis ofthe Scottish Heart Health Study archive.Soc Sci Med. 1997;45:893–902.

69. Phillimore PR, Morris D. Discrep-ant legacies: premature mortality in twoindustrial towns. Soc Sci Med. 1991;33:139–152.

70. Forsyth A, Macintyre S, AndersonA. Diets for disease? Intraurban varia-tion in reported food consumption inGlasgow. Appetite. 1994;22:259–274.

71. Ellaway A, Anderson A, MacintyreS. Does area of residence affect bodysize and shape? Int J Obes. 1997;21:304–308.

72. Ellaway A, Macintyre S. Doeswhere you live predict health relatedbehaviours? Health Bull. 1996;54:443–446.

73. Tienda M. Poor people and poorplaces: deciphering neighborhood ef-fects on poverty outcomes. In: Huber J,ed. Macro-Micro Linkages in Sociology.Newbury Park, Calif: Sage Publications;1991:244–262.

74. Furstenberg FF, Hughes ME. Theinfluence of neighborhoods on chil-dren’s development: a theoretical per-spective and a research agenda. In:

Brooks-Gunn J, Duncan GJ, Aber JL,eds. Neighborhood Poverty: Volume II:Policy Implications in Studying Neighbor-hoods. New York, NY: Russell SageFoundation; 1997:23–47.

75. Cook TD, Shagle SC, Degirmen-cioglu SM. Capturing social process fortesting mediational models of neighbor-hood effects. In: Brooks-Gunn J, DuncanGJ, Aber JL, eds. Neighborhood Poverty:Volume II: Policy Implications in StudyingNeighborhoods. New York, NY: RussellSage Foundation; 1997:94–119.

76. Brooks-Gunn J, Duncan GJ, Leven-thal T, Aber JL. Lessons learned and fu-ture directions for research on theneighborhoods in which people live. In:Brooks-Gunn J, Duncan GJ, Aber JL,eds. Neighborhood Poverty, Volume I:Context and Consequences for Children.New York, NY: Russell Sage Foundation;1997:279–297.

77. Raudenbush S, Sampson R. Eco-metrics: toward a science of assessingecological settings, with application tothe systematic social observation ofneighborhoods. Sociol Methodology.1999;29:1–41.

78. Sampson RJ, Raudenbush SW. Sys-tematic social observation of publicspaces: a new look at disorder in urbanneighborhoods. Am J Sociol. 1999;105:603–651.

79. Cheadle A, Wagner E, Koepsell T,Kristal A, Patrick D. Environmental indi-cators: a tool for evaluating community-based health promotion programs. Am JPrev Med. 1992;8:345–350.

80. Blalock HM. Contextual-effectsmodels: theoretical and methodologicalissues. Annu Rev Sociol. 1984;10:353–372.

81. Hauser RM. Context and consex: acautionary tale. Am J Sociol. 1970;75:645–664.

82. Duncan GJ, Connell JP, KlebanovPK. Conceptual and methodological is-sues in estimating causal effects ofneighborhoods and family conditions onindividual development. In: Brooks-Gunn J, Duncan GJ, Aber JL, eds.Neighborhood Poverty, Volume I: Contextand Consequences for Children. NewYork, NY: Russell Sage Foundation;1997:219–250.

83. Jones K, Gould MI, Duncan C.Death and deprivation: an exploratoryanalysis of deaths in the Health andLifestyle Survey. Soc Sci Med. 2000;50:1059–1079.

84. Waitzman NJ, Smith KR. Separatebut lethal: the effects of economic seg-regation on mortality in metropolitanAmerica. Milbank Q. 1998;76:341–373.

85. Yen I, Kaplan G. Poverty area resi-

dence and changes in physical activitylevel. Am J Public Health. 1998;88:1709–1712.

86. Kreft I, deLeeuw J. IntroducingMultilevel Modeling. London, England:Sage Publications; 1998.

87. Snijders T, Bosker R. MultilevelAnalysis. An Introduction to Basic andAdvanced Multilevel Modelling. London,England: Sage Publications; 1999.

88. Korbin JE, Coulton CJ. Under-standing the neighborhood context forchildren and families: combining epi-demiological and ethnographic ap-proaches. In: Brooks-Gunn J, DuncanGJ, Aber JL, eds. Neighborhood Poverty,Volume II: Policy Implications in StudyingNeighborhoods. New York, NY: RussellSage Foundation; 1997:65–79.

89. Macintyre S, Annamdale E, EcobR, et al. The West of Scotland Twenty-07 Study: health in the community. In:Martin C, MacQueen D, eds. Readingsfor a New Public Health. Edinburgh,Scotland: Edinburgh University Press;1989:56–74.

90. Koopman JS, Lynch J. Individualcausal models and population systemmodels in epidemiology. Am J PublicHealth. 1999;89:1170–1174.

91. Schwartz S, Susser E, Susser M. Afuture for epidemiology? Annu Rev Pub-lic Health. 1999;20:15–33.

Documents

Investigating Neighborhood and Area Effects on Health