The notion that environmental features may be related to psychological well-being and mental health has a long history. As far back as 1939, Faris and Dunham1 found that schizophrenia and substance abuse rates were highest amongst individuals living in socially disorganised Chicago neighbourhoods.2 In My Name is Legion, published in 1959, Alexander Leighton3 explored how the expression of mental illness was shaped by local context and concluded that processes underlying the sociocultural disintegration of neighbourhoods may be shaping patterns of mental health and psychiatric disorder.4

In recent years, there has been an explosion of interest in the peer-reviewed medical and public health literature about the ways in which neighbourhoods and residential environments may affect a variety of health outcomes, including mental health and depression.5 This interest has been spurred by theoretical discussions of the ecologic determinants of health6 7 as well as by the growing popularity and availability of multilevel analysis,8 a statistical technique that has been used to assess the relation between neighbourhood context and health after controlling for potential individual-level confounders.9 10

There are many theoretical reasons why neighbourhood environments may be particularly relevant to mental health, and specifically to depression and depressive symptoms. Features of neighbourhoods such as lack of resources, disorder, violence, inadequate housing, and lack of green spaces may function as stressors.11 12 Neighbourhood features may also act as buffers of individual-based sources of stress related to mental illness. For example, physical and social features of neighbourhoods may affect social connections and the levels of social support experienced by residents. Social support may in turn affect residents’ vulnerability to stress and depressive symptoms.13

sDespite some theoretical rationale for neighbourhood effects on depressive symptoms, the results of the literature in this area are still somewhat mixed.14 In this review we take stock of the published observational studies of neighbourhoods and depression and depressive symptoms in order to identify future directions for the field. We summarise the main research questions, study populations, neighbourhood definitions, neighbourhood measures, depressive symptom measures, study designs, analytic techniques and results from these studies. The review concludes by discussing the remaining gaps in our knowledge about the relationship between neighbourhood context and depression, and suggests future research directions. This review complements a prior review of neighbourhoods and mental health by focusing specifically on the more narrow outcomes of depression and depressive symptoms, extending the review to also encompass articles published from 2004 to 2007 (a time of increasing publications in this area), and focusing on observational studies and their limitations.15

METHODS

Studies were primarily identified using a biomedical database (PubMed) and two databases of psychological literature (PsycINFO and PsycARTICLES). The search terms “depression,” “depressive symptoms,” or “psychological distress” were entered together with “neighbourhood” or “neighbourhood characteristics”. These terms were selected since we were interested in any type of neighbourhood effect on either depression or depressive symptoms. These searches retrieved 79 articles in PubMed and 168 articles in PscyINFO. PsycARTICLES did not turn up any studies that were not found using the PsycINFO database. Additional studies were identified from the reference lists of the papers identified in the PubMed and PsycINFO searches. Studies included in this review are English-language cross-sectional and longitudinal studies that used at least one neighbourhood-level variable in the analysis, and had either depression or depressive symptoms as the outcome. All studies were published between January 1990 and August 2007. Excluded articles included reviews and opinion pieces, studies without any geographical component, studies that looked at depression or depressive symptoms only as mediators, and articles that did not differentiate between depression and other psychiatric conditions such as schizophrenia. In total, 45 reports of observational studies of the relation between depression and neighbourhood characteristics were identified using these search methods.

RESULTS

The main research questions, study populations, neighbourhood definitions, neighbourhood features, depression measures, study design, analytical technique and key results of the 45 studies are described in table 1 (online).

Research questions

Of the 45 studies reviewed, the majority (n = 26) focused solely on the main effects of neighbourhood-level variables on depression,2 11 13 16^–38 three were primarily interested in how neighbourhood characteristics moderate the association between individual-level risk factors and depressive symptoms,39^–41 and 15 examined both the main effects of neighbourhood conditions and the interactions of these characteristics with individual-level variables.42^–56 One study was primarily interested in the interaction of two neighbourhood characteristics.57

Study population

Studies have varied widely both in sample size and in the characteristics of the populations studied. The size of study populations varied from 117 to 56 428 subjects. Some of the studies restricted their populations to specific racial/ethnic groups or age categories, whereas others included a wide range of demographic characteristics. Twenty-nine studies examined the association between neighbourhood characteristics and depressive symptoms in adult populations across broad age ranges,2 11 21 23^–2527^–3840^–4347494650^–5257 10 studies focused on groups of children or teenagers,18^–2022263944535456 and six restricted their populations to people aged 65 and over.13 16 17 45 48 55 The gender distribution across most studies was relatively evenly balanced. Five of the studies excluded men from analyses,37 41 42 49 51 whereas 40 sampled both men and women. Eight studies restricted their study population to African-Americans19 22 39 41 42 49 53 56 and one study only examined Mexican-Americans.17 The remaining studies enrolled a mixture of racial/ethnic groups, most commonly using random sampling of their study populations. The majority of studies were conducted in metropolitan or urban areas: only 14 studies included non-urban dwelling subjects in their populations.16 17 19 22 23 25 29 32^–34 44 46 50 51

Neighbourhood definitions

The definition and size of a neighbourhood varied widely across studies. Neighbourhood definitions ranged from participant-defined areas to census-defined areas (census blocks, tracts or clusters of tracts). Among the 34 studies conducted in the USA the vast majority (n = 21) used census or administratively defined areas: five used census block groups (average population approximately 1000 people),39 40 42 49 52 nine used census tracts (average population approximately 4000),2 11 13 16 17 38 48 55^–57 and seven used clusters of block groups or tracts.18^–21 27 28 44 Twelve studies asked each study participant to define their own neighbourhood22^–26 41 45 47 51 53 54 56 and one study used circular buffers of varying sizes around residences to define neighbourhoods.36

The nine studies conducted in the UK used government-defined areas as proxies for neighbourhood, ranging from British electoral wards (mean population about 5500) to larger regional units, such as the 22 regional unitary authorities of Wales (mean population 122 850).29^–35 46 50 Studies conducted in Canada and the Netherlands also used administratively defined areas (census tracts in Canada43 and boroughs in Amsterdam30).

Neighbourhood features

The neighbourhood characteristics investigated fall into two categories: structural characteristics—such as neighbourhood socioeconomic and racial/ethnic composition, residential stability, and the built and service environments—and measures of social processes—such as neighbourhood disorder, social cohesion and ties with neighbours, and perceived exposure to crime, violence, drug use and graffiti. Structural characteristics were the most common features examined (33 out of the 45 studies). Twenty-five studies examined the contextual8 effect of neighbourhood socioeconomic position (after accounting for compositional differences)2 11 13 16 17 19^–2128^–3437394042434648^–5057 and nine of these studies included no other type of neighbourhood characteristic.21 28^–3032^–344650 Racial/ethnic composition (examined in 10 studies)13 16 17 19 20 38 40 43 48 55 and residential mobility (examined in eight studies)2 11 13 16 18 20 43 48 were the other two structural characteristics most commonly examined. Four studies investigated the role of the built environment26 27 35 36 and one study examined the available service environment.13

Twenty-five of the 45 studies examined the association between neighbourhood social processes and depressive symptoms.11 18 19 22^–2531373941424445474951^–57 Of these, 10 examined neighbourhood disorder and related domains,11 24 44 45 47 49 51 52 56 57 16 examined social interactions between neighbours 18 19 22^–243137394142475254^–57 and 12 investigated exposure to violence and other hazardous conditions18 19 22 25 26 41 42 44 52^–5456

Twelve studies examined both neighbourhood structural characteristics and social processes.11 18^–203137394249515557 Nine of the 12 studies looked at both neighbourhood socioeconomic characteristics and social processes.11 19 20 31 37 39 42 49 57

Measurement of neighbourhood features

Neighbourhood characteristics were measured using a variety of techniques. Census-derived neighbourhood variables were the most common measures used (16 studies),2 16 17 21 28^–3032^–34384043464850 followed by self-reports of neighbourhood characteristics by study participants (14 studies).22^–26363741454752^–5456 Ten studies included both census-derived measures and participant self-reports.11 18^–20394249515557 A small number of studies created measures by using objective raters who did not live in the neighbourhoods 27 35 44 or by using resources such as phone books to construct neighbourhood measures,13 and investigated the measures so constructed either on their own or in combination with census measures.

Depression measures

The most common outcome measure examined was the CES-D (Center for Epidemiologic Studies-Depression) scale (either full or modified) (19 out of 45 studies).11 13 16 17 21 22 24 25 35 36 37 40 41 45 48 51 52 55 57 Nine studies relied on measures approximating DSM-IV (Diagnostic and Statistical Manual of Mental Disorders) criteria, a measure of clinical depression.2 27 28 33 38 42 43 47 49 Studies of children or adolescents also tended to used instruments that approximated DSM criteria.18–202639445354 Six studies, mainly carried out in the UK, used the GHQ (General Health Questionnaire),30 31 34 46 47 50 a scale created to assess four elements of non-psychotic distress, including depression.58 The SF-36 (Mental health index of the Short Form Health Survey 36) was used in two studies29 32 and a question from a general health survey (Behavioural Risk Factors Surveillance System) was used in one study.23

Study designs

The majority of the studies (35) were cross-sectional in nature. Only 10 of the 45 studies used any type of follow-up or prospective analysis.20 21 24 28 34 44 49 51 53 54 Thirty-two studies had multilevel designs in that they included data on individuals nested within neighbourhoods and collected data at both levels.2 11 13 16 17 19^–212327^–3537384042^–444648^–525557 Thirteen were purely individual-level studies in which individual-level reports of neighbourhood characteristics were linked to individual-level outcomes in an individual-level analysis.18 22 24^–263639414547535456

Analytical techniques

Twenty-one of the studies used linear or logistic multilevel models to investigate the relationship between depression or depressive symptoms and area-level characteristics.11 13 16 19 20 27^–303234374042^–4448^–505257 The remaining studies used single-level linear or logistic regression,2 18 21 22 24^–2635363945^–475153–56 structural equation modelling,41 or tests of significance of the difference in prevalence rates between groups.38 The four studies that contrasted results from multilevel analysis with an analysis ignoring the multilevel structure reported similar results with both approaches.17 23 31 33 Individual-level confounders, most commonly age, gender, marital status, ethnicity, education, employment status, financial strain, and number of current physical health problems, were included in models in all 45 studies.

Study results

Thirty-seven of the 45 studies found support for an association between neighbourhood characteristics and depression or depressive symptoms after controlling for a variety of individual-level characteristics, usually a combination of race/ethnicity, age, gender, marital status, education and income. When categorised by study design six of the seven purely longitudinal studies20 21 24 28 44 51 and 29 of the 35 purely cross-sectional studies reported associations.2 11 13 17^–19222325–272931–3335^–3941^–4345474855–57 Three studies had both cross-sectional and prospective elements: one of these found a significant association with both types of analyses,53 54 while another only found a significant association in their cross-sectional data.49 The six studies that reported ICCs (intraclass correlations) for depression measures generally reported ICCs in the 0.4–2.9% range for cross-sectional studies of adult populations,30^–3343 11% for children20 and 1% for longitudinal analyses.44

Differences based on neighbourhood characteristics and definitions

Study results differed depending on which neighbourhood characteristics were being studied. Overall, 24 of the 46 different structural characteristics (52%) examined were significantly associated with depressive symptoms/depression. Thirteen of the 25 studies that examined the effect of neighbourhood socioeconomic position on depressive symptoms found evidence to support the presence of an association after adjustment for individual-level characteristics. 11 13 17 20 21 28 29 32 33 37 41 43 49 Four of the eight studies that examined the association between depression and residential mobility found evidence of an association.2 43 48 57 Only four of the 10 studies that examined racial/ethnic composition of neighbourhoods found support for the association between neighbourhood context and depression.17 19 38 55 All four studies that looked at the association between depressive symptoms and the built environment (specifically the internal and external built environment, the quality of housing areas, the walking environment and a negative neighbourhood environment, identified by factors such as violence, abandoned buildings and homeless people on the streets) found an association with depressive symptoms.26 27 35 36

Twenty-five of the 37 social processes (68%) examined in the studies were significantly associated with depression/depressive symptoms. All but one52 of the nine studies that assessed whether individual perceptions of the conditions and disorder in one’s neighbourhood affected risk of depression concluded that these factors were associated with depressive symptoms.11 24 44 45 47 51 56 57 Eleven out of 16 studies found positive social interactions between neighbours to be a protective factor against depression.18 22 24 37 39 41 47 54 55 56 57 Exposure to violence and hazardous conditions was found to be associated with depressive symptoms in six out of 12 studies.18 22 25 41 42 44

Both of the studies that systematically compared results for different scales found no consistent evidence that results differed systematically by neighbourhood size, although one study suggested that small scales (smaller than electoral ward in the UK) may be most relevant to depression.34 36 Study results differed somewhat in the USA and in the UK. Regardless of how they defined neighbourhood, UK studies found evidence for associations between neighbourhood environments and depression in only two-thirds of the studies (6 out of 9), whereas studies in the USA, independent of the size or definition of neighbourhood, found associations between at least one neighbourhood characteristic and depression in 30 out of 34 studies.2 11 13 16^–283638^–42444547^–4951^–57

Heterogeneity in the effects of neighbourhood-level variables

It is often hypothesised that the effect of neighbourhood context on depression may vary by gender, age, racial/ethnic group or socioeconomic position. Of the nine studies that reported results either stratified by gender or with interaction terms between gender and neighbourhood characteristics, two found that neighbourhood characteristics were more strongly associated with depressive symptoms in women22 54 and one found a stronger association in men,36 whereas others had mixed results40 45 53 56 or found no difference between genders.32 43 Although the number of studies of children or of older people was generally small, there was limited evidence of more consistent associations in children or older people: four of the five studies that restricted their populations to older people and 9 of the 10 studies of children aged 18 and under found evidence of an association between neighbourhood characteristics and depressive symptoms, compared with 24 of 30 studies of adult populations. Very few studies have investigated heterogeneity by race/ethnicity.17 18 40 47 In a Baltimore study, community cohesion was associated with less depression amongst White people, but was not associated with depression amongst African-Americans.47 One study found Mexican-Americans had better mental health in areas with high concentrations of Mexican-Americans, whereas another study found that African-Americans had worse mental health in areas with higher concentrations of African-Americans, although this association disappeared after adjustment for individual-level variables.17 40

Five studies examined interactions of neighbourhood characteristics with individual-level socioeconomic position. Three of these studies found no interaction,40 48 49 whereas two found a significant interaction between individual-level economic status and neighbourhood conditions.46 50 Wealthy individuals living in areas with high income inequality had higher levels of mental disorders than those living in more equal areas, but the opposite was true for poor individuals.46 Living in a poverty area was only associated with worse mental health outcomes among the unemployed in another study.50 Other sets of interactions have also been examined in a small number of studies: knowing one’s neighbours was more strongly associated with higher levels of childhood anxiety and depression in poverty area neighbourhoods than in wealthy neighbourhoods;39 parents’ use of inductive reasoning was a protective factor for African-American teenagers’ levels of depressive symptoms only for those living in disordered neighbourhoods;44 and residential stability was associated with lower levels of depressive symptoms in wealthy neighbourhoods and higher levels in poor neighbourhoods.57

Longitudinal studies

Ten of the 45 studies used some type of follow-up or prospective analysis.20 21 24 28 34 44 49 51 53 54 Two studies had 1 year or less of follow-up time,24 34 six studies had 1–2 years of follow-up,20 28 44 49 51 53 one study had 7–8 years of follow-up54 and one study followed subjects for 10 years.21 Nine of the 10 studies had two waves of data,20 21 24 34 44 49 51 53 54 whereas one study used three waves of data collection.28 Four studies defined incident depression/depressive symptoms as all subjects who did not have depression or fell below a certain cut-off level of depressive symptoms at baseline, but who did have depression or were above the cut-off level at follow-up time(s),21 28 34 49 one study used a change score,51 and five studies simply used the level/presence of depressive symptoms at follow-up, with three of these controlling for baseline levels in their models.20 24 44 53 54 Four studies restricted their populations to children20 44 53 54 and two to women,49 51 whereas the remainder enrolled representative adult populations.21 24 28 34 Each of these studies used a different definition of neighbourhood: New York City community districts, census block groups, clusters of census block group areas, British electoral wards, clusters of multiple census tracts, poverty areas/non-poverty areas and participant-defined neighbourhoods. Five of these studies focused on measures of neighbourhood socioeconomic position and disadvantage,20 21 28 34 49 and two of these additionally examined social cohesion and neighbourhood disorder as predictors.20 49 Four of the five studies that examined the association between neighbourhood socioeconomic status and development of depressive symptoms found evidence of an association,20 21 28 49 after controlling for combinations of age, education, sex, race/ethnicity, income, stressors, marital status, number of children, receiving government assistance, perceived health status, body mass index, smoking and alcohol consumption, whereas one found no association.34 Neighbourhood disorder was prospectively associated with depressive symptoms in four out of five studies.24 45 51 53 Neighbourhood cohesiveness was associated with depressive symptoms in two20 54 of the three studies that examined this process.20 24 54

DISCUSSION

Of the 45 studies reviewed, 37 reported associations of at least one neighbourhood characteristic with depression or depressive symptoms after controlling for individual-level characteristics. The percentage of positive results was similar in cross-sectional (82%) and longitudinal (70%) studies. The associations of depressive symptoms/depression with structural features were less consistent (52% significantly associated) than with social processes (68%). Among the structural features, measures of the built environment appeared to be more consistently associated with depression than socioeconomic deprivation, residential stability or race composition, although only a few studies to date have investigated the built environment.

Although a wide variety of area definitions were investigated, very few studies systematically compared area definitions and no clear pattern emerged from the comparison of studies using different-sized areas. Controlling for individual-level confounders often reduced the magnitude of the association between neighbourhood characteristics and depression/depressive symptoms, although the association rarely disappeared all together. Interactions were investigated in only a small number of studies making it difficult to draw any conclusions about vulnerable groups, although there was some evidence of stronger effects in children and older people than in adult populations. The studies varied widely in neighbourhood definitions, in the neighbourhood-level variables investigated and in the individual-level covariates examined, making it impossible to conduct a meta-analysis of study results. Increasing comparability across studies in the geographic areas, the variables and the outcomes examined to conduct systematic reviews is an important need in the field.

Current limitations in this body of literature include limited theory about how neighbourhoods may influence depression and depressive symptoms; the lack of consistency in the definitions of neighbourhoods and the measures of neighbourhood-level properties examined; the possibility of reporting bias, reverse causation and residual confounding; the dearth of studies exploring different spatial scales; and the relative lack of longitudinal studies. Five important research directions emerge from the reviewed works. These research directions are (1) developing theory on the processes through which specific area features may affect mental health, including theories on the most vulnerable groups; (2) improving the measures of neighbourhood or area-level factors necessary to test these theories empirically; (3) investigation of a broad range of areas (or spatial scales) and neighbourhood–person interactions; (4) addressing issues of reporting bias, reverse causation and residual confounding; and (5) increasing the use longitudinal designs and quasi-experimental or experimental designs.

Developing theory on the processes through which area features may be associated with depression and depressive symptoms and empirically testing specific predictions derived from these theories is fundamental to strengthening causal inference. Empirical investigations of the processes linking neighbourhood characteristics to depression will require the measurement of the specific neighbourhood attributes involved. To date, the majority of studies have used measures of the socioeconomic composition of areas as a proxy for the more specific area attributes that may be relevant. A growing number of studies have attempted to measure specific attributes of neighbourhoods such as the built environment, social cohesion disorder or crime.11 22 24 27 41 45 47 It is interesting to note that findings have generally been more consistent for studies focusing on specific neighbourhood attributes than those focusing on aggregate measures of socioeconomic position or deprivation. However, the measures used across studies have varied widely, making comparisons difficult. Developing standardised measurement instruments that can be applied across studies so that findings can be systematically compared will be an important advancement. One methodology that could potentially be explored further involves the use of geographic information systems (GIS) to construct measures of the built environment and the physical layout of neighbourhoods hypothesised to be related to mental health or to create synthetic geographical areas with optimised homogeneity of social characteristics.59

There is little consensus on what spatial scales (ranging from the immediate built environment of the home to broader regional characteristics) may be relevant to depression or depressive symptoms in different population groups. The development of hypotheses on relevant spatial definitions will require more sophisticated theory on how persons interact with and are affected by spatial contexts. In the absence of clear theory on the spatial scale relevant to a particular process, researchers can conduct sensitivity analyses to determine the effects of different definitions of “neighbourhoods” on the results of their research.59 The definition and size of a neighbourhood varied widely across the studies in this review and few studies have examined sensitivity of results to the use of measures corresponding to different-sized areas.36 50 The use of spatial analytic methods is another promising arena that has not yet been extensively used in this body of literature.60^–62 These methods can be used to investigate the spatial patterning of health outcomes without relying on arbitrary defined boundaries. This spatial patterning can provide information on the spatial scale at which the relevant processes may be operating.

An important methodological challenge in investigating neighbourhood effects on depression is reporting bias (sometimes also referred to as same-source bias). Reporting bias may arise for example if people who are already depressed report lower levels of social cohesion and a worse external environment because of their depression. Many of the studies in this review measured neighbourhood conditions from the same sample of people from whom they took measurements on depressive symptoms. The association between social cohesion and depressive symptoms, for example, might exist because depressed people feel more alone, even though their neighbourhood, objectively, does not have low social cohesion. A growing body of work on ecological measurement has begun to develop alternative ways to use survey data or objectively collected data on the built environment (though publicly available data or systematic social observation) to characterise neighbourhood environments in ways that avoid same-source bias.63 64 Greater use of these methods in the area of neighbourhood characteristics and depression is needed.

Reverse causation and residual confounding by individual-level variables are two additional methodological problems. Reverse causation would arise if people who are depressed tend to stay in or move into deprived neighbourhoods. In this case the exposure to the neighbourhood condition is a consequence of (and not a cause of) depression. All cross-sectional studies are vulnerable to the problems of reverse causation. Longitudinal designs are necessary to rule out reverse causation as an explanation for cross-sectional associations. As in other neighbourhood effects research, the possibility of residual confounding by individual-level variables is an important limitation of observational studies of neighbourhoods and depression Most studies in this review attempted to address this issue by controlling for a variety of individual-level variables, but there is no consensus on what the key confounders are likely to be, or on the sensitivity of results to plausible amounts of residual confounding. Other approaches sometimes used to control for multiple confounders such as propensity score analysis 65 66 have not been used in research on neighbourhoods and depression.

The majority of existing studies of neighbourhoods and depression are cross-sectional. As noted above, longitudinal studies are necessary to rule out reverse causation. They are also needed to investigate time lags and cumulative effects of neighbourhoods on depression. Short of the ideal randomised experiment, natural experiments or quasi-experimental designs may also provide opportunities to examine causal effects of neighbourhood or area attributes on depression avoiding some of the pitfalls of observational studies. For example, a study could examine changes in depressive symptoms over time in a neighbourhood in response to some source of exogenous variation such as the inauguration of a new public space, or the implementation of a new community policing approach. These interventions, which are “naturally occurring” in neighbourhoods all the time, provide valuable but as yet untapped opportunities to investigate area or neighbourhood effects on depression. In summary, existing observational evidence supports a role of neighbourhood conditions in the development of depression and depressive symptoms. However, more refined observational work (including the study of natural experiments) is needed to determine whether the associations observed are causal and what the relevant neighbourhood-level attributes and mediating variables might be.