Upload
ricardo-araya
View
214
Download
0
Embed Size (px)
Citation preview
ARTICLE IN PRESS
0277-9536/$ - se
doi:10.1016/j.so
�CorrespondE-mail addr
wmsfdjd@fores
(R. Playle), Tho
weisrp.Meds.Uw
glyn.lewis@bris
Social Science & Medicine 62 (2006) 3072–3083
www.elsevier.com/locate/socscimed
Perceptions of social capital and the built environment andmental health
Ricardo Arayaa,�, Frank Dunstanb, Rebecca Playleb, Hollie Thomasc,Stephen Palmerb, Glyn Lewisa
aUniversity of Bristol, Bristol, UKbDepartment of Epidemiology, Statistics and Public Health, University of Wales College of Medicine, Cardiff, UK
cDepartment of Psychological Medicine, University of Wales College of Medicine, Cardiff, UK
Available online 24 January 2006
Abstract
There has been much speculation about a possible association between the social and built environment and health, but
the empirical evidence is still elusive. The social and built environments are best seen as contextual concepts but they are
usually estimated as an aggregation of individual compositional measures, such as perceptions on trust or the desirability
to live in an area. If these aggregated compositional measures were valid measures, one would expect that they would
evince correlations at higher levels of data collection (e.g., neighbourhood). The aims of this paper are: (1) to investigate
the factor structure of a self-administered questionnaire measuring individual perceptions of trust, social participation,
social cohesion, social control, and the built environment; (2) to investigate variation in these factors at higher than the
individual level (households and postcodes) in order to assess if these constructs reflect some contextual effect; and (3) to
study the association between mental health, as measured by the General Health Questionnaire-12 (GHQ-12), and these
derived factors. A cross-sectional household survey was undertaken during May–August 2001 in a district of South Wales
with a population of 140,000. We found that factor analysis grouped our questions in factors similar to the theoretical ones
we had previously envisaged. We also found that approximately one-third of the variance for neighbourhood quality and
10% for social control was explained at postcode (neighbourhood) level after adjusting for individual variables, thus
suggesting that some of our compositional measures capture contextual characteristics of the built and social environment.
After adjusting for individual variables, trust and social cohesion, two key social capital components were the only factors
to show statistically significant associations with GHQ-12 scores. However, these factors also showed little variation at
postcode levels, suggesting a stronger individual determination. We conclude that our results provide some evidence in
support of an association between mental health (GHQ-12 scores) and perceptions of social capital, but less support for the
contextual nature of social capital.
r 2005 Elsevier Ltd. All rights reserved.
Keywords: Social capital; Mental health; Contextual effects; Wales
e front matter r 2005 Elsevier Ltd. All rights reserved
cscimed.2005.11.037
ing author. Tel.: +44 117 9546702.
esses: [email protected] (R. Araya),
t.cf.ac.uk (F. Dunstan), [email protected]
[email protected] (H. Thomas),
[email protected] (S. Palmer),
.ac.uk (G. Lewis).
Introduction
There has been an ongoing debate on theimportance of people and places for health. Thisinterest has been stimulated by the debate on social
.
ARTICLE IN PRESSR. Araya et al. / Social Science & Medicine 62 (2006) 3072–3083 3073
capital and health (Kawachi & Berkman, 2000;Muntaner & Lynch, 2002; O’Brien Caughy, O’Cam-po, & Muntaner, 2003; Pearce & Davey-Smith,2003; Sampson, 2003; Tunstall, Shaw, & Dorling,2004) and studies examining health variationsacross geographical areas (Ellaway, Macintyre, &Kearns, 2001; Macintyre, Ellaway, & Cummins,2002; Macintyre, Maciver, & Sooman, 1993; Ska-pinakis, Lewis, Araya, Jones, & Williams, 2005).
Interest in finding better ways of measuring andcapturing the contextual nature of places, neigh-bourhoods, and communities has also grown. Someof the difficulties in the measurement of contextualvariables can be appreciated when examining howsocial capital has been measured so far.
Although there is no consensually agreed defini-tion of social capital, one commonly used one isthat it refers to how social relations and networksinfluence collective action for mutual benefit(Kawachi, Kennedy, Lochner, & Prothrow-Stith,1997; Putnam, 1993). The concept of social capitalcan be disaggregated into at least two importantcomponents, structural and cognitive (Bain &Hicks, 1998). Whilst the former refers to the extentand intensity of associational links, the latter hasmore to do with qualitative aspects of these links,such as levels of trust or reciprocity.
Much of the research in social capital and healthhas used these so-called cognitive features assessedat an individual level to estimate indirectly theamount of social capital in an area (Kawachi,Kennedy, & Glass, 1999; Kawachi et al., 1997;Lochner, Kawachi, Brennan, & Buka, 2003; Loch-ner, Kawachi, & Kennedy, 1999; McCulloch, 2003;Subramanian, Lochner, & Kawachi, 2003). This isproblematic because these cognitive aspects of socialcapital are meant to represent a contextual con-struct, rather than a compositional one obtainedthrough aggregating individual data. Two different,but not mutually exclusive, approaches to pursuingthis challenge have been used. Firstly, these featurescan be measured through direct observation ofcertain collective behaviours in an area; e.g., peoplenot respecting zebra crossings or arguing in publicspaces might indicate lower levels of social capital.Although this approach might be more contextuallyvalid, it can be resource intensive, some features arenot directly observable, and some observationsrequire subjective inferences, such as deciding iftwo people engaged in a discussion can be regardedas ‘arguing’. Secondly, some of these cognitivecomponents can also be measured indirectly by
investigating individuals’ perceptions of, e.g., howmuch people trust each other in the neighbourhood.These individual perceptions are then aggregated oranalysed at higher levels of aggregation usingmultilevel models to obtain an estimate of the levelof trust in the area. Several self-reported question-naires have been developed to assess these percep-tions on various aspects of social capital withdifferences depending on the specific aims for whichthey were designed (Lochner et al., 1999; Sampson,Raudenbush, & Earls, 1997; The World Bank SocialCapital Thematic Group, 2002).
Although this methodology is simple and feasible,it is still questionable if these perceptions are validcontextual measures or just simply the sum ofindividuals’ perceptions (compositional). Further-more, even if these perceptions reflected a trulycontextual characteristic it is still possible that thisestimate could be somehow confounded by thecharacteristics of the individuals living in that place.One way of indirectly attempting to clarify if theseperceptions represent some contextual construct isby trying to ascertain what proportion of thevariance on any of these constructs (e.g., percep-tions on trust or reciprocity) is explained at higherlevels, such as neighbourhoods, after accounting forindividual factors. In a recent study, Subramanianet al. (2003) used individual data on the perceptionof trust by individuals in Chicago, USA, to examinewhether there were true differences in trust betweenneighbourhoods after accounting for individualvariation. Their results suggested that, even afteraccounting for individual socio-demographic vari-ables, significant neighbourhood variation remainedin the individual perception of trust (Subramanianet al., 2003).
The built environment can be assessed usingdirect observations of the characteristics of geogra-phical areas (Perkins, Meeks, & Taylor, 1992;Weich, Holt, Twigg, Jones, & Lewis, 2003) orthrough perceptions of residents on the quality oftheir built environment (Dalgard & Tambs, 1997).Perceptions on the built environment, such asgraffiti on walls or dirtiness, are hypotheticallysubject to similar respondent bias as perceptions onsocial capital, e.g., trust or social cohesion.
It could be argued that the quality of the builtenvironment is a consequence of different levels ofsocial capital or vice versa. For instance, a highproportion of houses with broken windows isprobably the consequence of low levels of socialcapital in the area. However, it is also plausible that
ARTICLE IN PRESSR. Araya et al. / Social Science & Medicine 62 (2006) 3072–30833074
unfavourable changes in the physical environmentmight lead to deterioration in trust or socialcohesion.
A few studies have been published in peer-reviewed journals reporting on the associationbetween social capital and common mental dis-orders among adults. Most of these studies haveonly measured and analysed social capital data atthe individual level (Ellaway et al., 2001; Harpham,Grant, & Rodriguez, 2004; Ross, 2000; Silver,Mulvey, & Swanson, 2002; Steptoe & Feldman,2001). Others have either aggregated individual datato create compositional variables representing high-er levels (e.g., neighbourhood) or analysed indivi-dual data using hierarchical multilevel models toestimate the level of variation at different levels(Cutrona, Russell, & Hessling, 2000; Skapinakiset al., 2005).
The underlying, but yet unproven, assumptionhas been that social capital is good for mentalhealth. Although some studies analysing data atindividual level have found inverse associationsbetween mental illness and social capital, moststudies using aggregated data or multilevel modelshave failed to find statistically significant associa-tions between social capital and common mentalillness at higher levels, such as neighbourhoods. It isworth emphasising that most of these studies haveused different sampling designs, measures ofsocial capital and mental health, and differenthierarchical data structures, making comparisonsrather problematic.
Studies on the built environment and mentalhealth have focused on residents’ perceptions oftheir environment (Dalgard & Tambs, 1997; Ross,2000) and geographical area variations or contex-tual assessments of the quality of the built environ-ment (Duncan, Jones, & Moon, 1995; Pickett &Pearl, 2001; Reijneveld & Schene, 1998; Wainwright& Surtes, 2003; Weich, Blanchard, & Prince, 2002;Weich et al., 2003). Most of these studies have alsofailed to find statistically significant area effects onmental health after accounting for individualfactors. Amongst studies that have used multilevelmodelling with positive findings, Skapinakis et al.(2005) found a small but significant associationbetween mental health and geographical areas in anationally representative sample in Wales but nospecific factor explained these findings (Skapinakiset al., 2005). Ross (2000) found that neighbourhooddisorder and residential instability were associatedwith depression (Ross, 2000). Other studies using
multivariate analysis of individual data have shownthat characteristics of the built environment canbe associated with psychiatric symptomatology(Dalgard & Tambs, 1997; Sampson, 2003; Weichet al., 2002).
The complex way in which the social and builtenvironment might interact to affect mental healthis unknown but there is no shortage of speculationon the potential mechanisms. It is possible that thesocial and built environment can effect changes ineach other and eventually impact on mental health.For instance, a poorly maintained built environ-ment with derelict buildings and covered in rubbishmight affect the sense of social cohesion in theneighbourhood, the combination of both leading topoorer mental health among its residents. But it isalso possible that poor social cohesion might lead toa poorer built environment as residents might havelittle interest to look after their common areas.Poorer mental health among residents might alsolead to less interest to keep the neighbourhood tidyand to engage in social interactions. All thesecombinations are possible and thus it is importantto conduct studies in which both the perceptions ofthe social and built environment are simultaneouslyassessed and analysed at different levels of dataaggregation. Different aspects of the built environ-ment may affect residents’ perceptions of theirneighbourhood and lead to behaviours congruentwith these beliefs. For instance, empty and boardedhouses may facilitate criminal activity and lead toperceptions of lack of safety and unwillingness tointeract with other people. Social withdrawal,isolation, and fear are likely to lead to theemergence of psychiatric symptoms among vulner-able people. However, there is as yet no empiricalevidence showing unequivocally these or manyother potential associations between the social andbuilt environment and mental health.
This study was part of a comprehensive researchprogramme [housing and neighbourhood andhealth (HANAH)] investigating the relationshipsbetween the built and social environment andhealth. In this paper, we present the results of theself-administered questionnaire developed to mea-sure the residents’ perceptions of aspects of thesocial and built environment. The aims of this paperwere: (1) to look for common factors grouping itemsof a self-administered questionnaire to measureperceptions of the social and built environment; (2)to investigate variation in factors derived from thequestionnaire at area level, after accounting for
ARTICLE IN PRESSR. Araya et al. / Social Science & Medicine 62 (2006) 3072–3083 3075
individual variables, in order to estimate if thesecompositional constructs might reflect some con-textual effect; and (3) to study the associationbetween mental health and these factors.
Methods
Sampling strategy
A cross-sectional household survey was under-taken during May–August 2001 in a district of SouthWales with a population of 140,000. The area’seconomy used to be dominated by heavy industry,including coal, steel, and petrochemicals, but thesehave seen a significant decline since the 1970s. Thishas left a legacy of environmental pollution fromheavy industry, poor standards of housing andamenities, high levels of poverty and economicinactivity. Some large-scale post-war housing devel-opments are now places of social isolation and lackof social and economic investment. Standards ofhealth in the former steel and coal communities,including the area in this study, are typical of similardeprived areas throughout the UK.
As it has been argued that ecological associationsare best explored using data from small areas (Curtis& Rees Jones, 1998; Perkins et al., 1992), our primarysampling unit was the postcode, often a single streetof houses, with a mean of 16.5 domestic householdsper postcode. This small scale relates to Barton,Grant, and Guise’s (2003) ‘home patch’ that isincreasingly seen as a useful unit for urban design(Barton et al., 2003). A stratified random sample of51 postcodes was chosen with probability of selectionproportional to their size. The strata were based ontwo factors, the degree of urbanicity and the level ofdeprivation as measured by the Townsend Depriva-tion Index (Townsend, Phillimore, & Beattie, 1998).This is a commonly used index of area deprivation inthe UK based on census data on unemployment,ownership of cars and housing, and overcrowding fordiverse areas. Seventy-five per cent of householdswere chosen randomly within each postcode, with amaximum of 20 per postcode. Questionnaires weresent to all adults between the ages of 16 and 75 withineach selected household (more details on samplingcan be requested from the authors
The questionnaire
A self-completion questionnaire was designed inwhich we included instruments and questions that
had been used to measure social capital and thequality of the built environment in other studies(Lochner et al., 1999; Ross, 2000; Sampson, 2003;The World Bank Social Capital Thematic Group,2002). Questions were included to reflect percep-tions on social cohesion (SC), trust (T), socialparticipation (SP), informal social control (ISC),neighbourhood quality (NQ), and neighbourhoodaccessibility (NA). These theorised constructs andthe questions belonging to each one of them arelisted in Table 1. A copy of this questionnaire can beobtained from the authors. Responses to questionswere rated on ordinal scales and recoded wherenecessary so that a higher score represented a higherlevel of social capital or a more desirable builtenvironment. The outcome measure for mentalhealth was the General Health Questionnaire-12(GHQ-12), incorporated into the self-completionquestionnaire (Goldberg & Williams, 1988). This isa commonly used questionnaire with 12 questionsasking about psychological well-being over the last7 days. Each question is answered on a 0–3 Likertscale. We used the sum of all items as a continuousscore, but we also used it as a binary measure, witha case being defined as scoring 3 or more afterquestions had been recoded as 0 or 1 (Goldberg &Williams, 1988). Since the GHQ-12 uses non-interval scales, treating its score as a binary variable(case/non-case) rather than a continuous measuremay be appropriate. However, previous researchhas also shown that it can be used as a continuousmeasures since total scores represent differentseverity levels (Goldberg & Williams, 1988). Thus,we have used both approaches in this paper.
Statistical analysis
An exploratory preliminary analysis was carriedout to estimate basic summary statistics and overallcorrelation structure of the responses. Principalcomponent analysis was used to simplify theresponses and to determine common underlyingfactors for the chosen items. A scree plot andeigenvalues were used to determine the optimumnumber of factors to select from the analysis. Theextraction method was principal components ana-lysis and an orthogonal rotation was performedwith varimax and Kaiser normalisation using SPSS11.0 (SPSS Inc., 2004).
Since all data were collected at individual level,these factor scores are defined for individuals,nested within households and postcodes. Rather
ARTICLE IN PRESS
Table
1
Principalcomponentanalysisofthequestionnaireonperceptionsofsocialcapitalandbuiltenvironmentin
South
Wales(n¼
931)
Proposed
theoretical
factor
Derived
factora
1Social
cohesion
2Neighbourhood
quality
3Inform
al
socialcontrol
4Trust
5Neighbourhood
accessibility
6Social
participation
Per
cen
tag
eo
fva
ria
tio
nex
pla
ined:
27.5
8.5
8.0
6.3
4.8
4.2
Ifeel
likeIbelongaroundhere.
SC
0.82
Ienjoylivingaroundhere.
SC
0.75
Ithinkoftheareaaroundhereasarealhomenot
just
aplace.
SC
0.73
Given
theopportunityIwould
liketo
moveaway
from
here.
SC
0.66
Iregularlystopandtalk
withpeople
around
here.
SC
0.60
Ifeel
differentfrom
people
aroundhere.
SC
0.55
Itrust
people
aroundhere.
T0.55
0.47
Litterisaproblem
aroundhere.
NQ
0.80
Graffiti/vandalism
isaproblem
aroundhere.
NQ
0.75
Thereare
notenoughgreen
areasortreesaround
here.
NQ
0.62
Theproperties
aroundhereare
tooclose
together.
NQ
0.61
Theareaaroundhereisnicelykeptbyits
residents.
NQ
0.60
Ithinkofthisareaasadesirable
place
tolive.
NQ
0.44
0.53
Thereare
other
placeswhichare
more
desirable
placesin
whichto
live.
NQ
0.51
Ithinkthisisagoodplace
tobringupchildren.
NQ/SC
0.42
0.44
0.42
How
likelyisitthatpeople
aroundherewould
interveneifchildrenwereshowingdisrespectto
anadult.
ISC
0.82
How
likelyisitthatpeople
aroundherewould
interveneifchildrenwerespray-paintinggraffiti
onalocalbuilding.
ISC
0.81
R. Araya et al. / Social Science & Medicine 62 (2006) 3072–30833076
ARTICLE IN PRESSHow
likelyisitthatpeople
aroundherewould
interveneifchildrenweretruantingfrom
school
andhangingaroundonstreet
corners.
ISC
0.75
How
likelyisitthatpeople
aroundherewould
interveneifafightwasbreakingoutin
frontof
theirhouse.
ISC
0.74
How
likelyisitthatpeople
aroundherewould
interveneifthefire
stationclosest
totheirhome
wasbeingthreatened
bybudget
cuts.
ISC
0.51
Ifeel
safe
aroundhereatnighttim
e.T
0.72
Ifeel
safe
aroundherein
thedaytime.
T0.65
Generallyspeaking,would
yousaythatmost
people
can’tbetrusted
orthatyoucan’tbetoo
carefulin
dealingwithpeople?
T0.53
Shoppingfacilities
are
within
easy
reach
ofmy
home.
NA
0.81
Healthfacilities
are
within
easy
reach
ofmy
home.
NA
0.81
Social/leisure
facilities
are
within
easy
reach
of
myhome.
NA
0.80
How
often
doyouparticipate
in:communityor
religiousactivities,voluntary
orlocalcommunity
group,adulteducationornightschoolclass,
leisure
centre,
socialoutingsandlibrary
use.
SP
0.82
Are
youactivelyinvolved
in:sports/sport
supportersclub,hobby/interest
group,political
party,neighbourhoodwatchschem
e,parent
teacher
association,tenant’sgroup,residents’
group,neighbourhoodcouncilorother
local
group.
SP
0.82
SC,socialcohesion;NQ,neighbourhoodquality;ISC,inform
alsocialcontrol;NA,neighbourhoodaccessibility;SP,socialparticipation;T,trust.
aFactorloadingsofless
than0.4
notincluded.
R. Araya et al. / Social Science & Medicine 62 (2006) 3072–3083 3077
ARTICLE IN PRESSR. Araya et al. / Social Science & Medicine 62 (2006) 3072–30833078
than aggregating these individual measures to higherlevels, multilevel models (three-level) were fitted tofind the percentages of variation that could beattributed to higher levels, such as households andpostcodes. Variation between factor scores arisesfrom a number of sources—between postcodes,between households in postcodes, and betweenindividuals within households. Multilevel modellingallows simultaneous estimation of the magnitude ofthese sources of variation which, for simplicity, wewill refer to as variation attributable to postcodes,households and individuals, respectively. Multilevelmodels are described fully elsewhere (Goldstein,1995). The association between GHQ total score andeach one of these factors was also estimated usinglinear and logistic multilevel regressions, dependingon whether the GHQ-12 was used as a continuous orbinary measure. Models were adjusted for age,gender, financial status (living comfortably, doingalright, just about getting by, finding it difficult, andfinding it very difficult), unaffordable items (mea-sured by the number of items, from a list, which weredesired but unaffordable), and employment status(employed, unemployed seeking work, housekeeper,student, retired, or permanently unable to work). Inthe continuous case, results were expressed in termsof estimated effect sizes, calculating the expectedchange in GHQ score for a change of 1 SD in thefactor score, while for the binary case an odds ratiowas calculated to assess the effect of such a change.The modelling was carried out using MLwiN version1.10 software, using iterated generalised leastsquares for the linear case and penalised quasi-likelihood with second-order linearisation in thelogistic case (Rasbash et al., 2000).
Results
Sample characteristics
In total, 1058 adults from 647 householdscompleted the questionnaires. This represents anindividual response rate of 66% and a householdresponse rate of 73%. The number of individualsper postcode varied from 1 to 47 with a mean of20.7. There was a range between 1 and 22 house-holds per postcode. The mean age of residents was46 years (SD 16.1 years), 55% were females, and32% of the households had children under the age16. Two-thirds were married or co-habiting, 19%were single, and the remainder were divorced,separated or widowed. As for educational qualifica-
tions, 17% had either received higher education orachieved qualifications enabling entry (‘A’ level),33% had no educational qualifications at all, andthe remaining 50% were in between these twoextremes. Just under half of the participants wereemployed and 22% were retired. Approximately30% reported experiencing financial difficulties.
The sample residing in the area was reasonablystable, 69% had lived in the area for at least 10 yearsand another 9% between 6 and 9 years. The mostcommon type of housing was semi-detached (42%)or mid-terrace (26%) with a wide range of propertyages. Approximately 13% of participants reportedovercrowding in their house, defined as a subjectivereport of overcrowding by the respondents. Almosttwo-thirds (67%) reported that their health ingeneral was good or excellent and only 12%reported poor or very poor health.
Questionnaire analysis
Thirty-one items were included in the analysis.Three items had rotated component loadings of lessthan 0.4 and hence did not relate strongly to any ofthe factors and were removed from the finalanalysis. These items were: ‘There are areas of (thistown) which are less desirable places in which tolive’, ‘How likely is it that people around here wouldpick up other peoples’ rubbish?’ and ‘I borrowthings and exchange favours with my neighbours’.
The contributions of the remaining 28 items tothe derived factors are shown in Table 1, contribu-tions less than 0.4 are not given. There were sixfactors with eigenvalues greater than 1 and the screeplot suggested this was a reasonable number offactors to select. All were readily interpretable andso this was selected as the optimal number offactors. As seen in Table 1, the rotated solution withsix factors was able to separate items with reason-able clarity into individual factors that matchedwith the theoretical factors defined prior to theanalysis. The six factors explained a total of 59.5%of the total variance; percentages accounted for bythe individual factors are shown in Table 1. Thosequestions that had high loadings on the factors wereexamined and it was found that the six factors werebroadly interpretable as ‘social cohesion’, ‘neigh-bourhood quality’, ‘informal social control’, ‘neigh-bourhood accessibility’, ‘trust’, and ‘socialparticipation’. These were very similar to thetheoretical constructs on which the questionnairewas designed and Table 1 shows the particular
ARTICLE IN PRESSR. Araya et al. / Social Science & Medicine 62 (2006) 3072–3083 3079
construct to which each question was expected tobelong. Another aim of the analysis was to producea set of summary scores from the completedquestionnaires. Each question was allocated to thefactor on which it had the highest loading and foreach factor the scores of allocated questions weresummed. The resulting factors are no longeruncorrelated since the weights are no longer thesame as the optimal loadings but they have themerit of simplicity. The scales were constructed sothat high scores correspond to positive views aboutthe area.
Contextual or compositional measure?
Table 2 shows the percentages of variationaccounted for by different levels in the multilevel
Table 2
Percentages of variance explained at different levels for each of the qu
environment in South Wales (2001)
Model Mean scores (SD) Va
Le
Unadjusted Ind
Factor Neighbourhood quality (NQ) 25.6 (5.73) 10
(6.
Social cohesion (SC) 26.1 (4.95) 2.4
(1.
Trust (T) 12.2 (2.58) 0.3
(0.
Neighbourhood accessibility (NA) 11.1 (2.32) 0.6
(0.
Informal social control (ISC) 5.05 (2.62) 0.7
(0.
Social participation (SP) 2.42 (2.00) 0.0
(0.
Adjustedb Ind
Factor Neighbourhood quality (NQ) 9.8
(5.
Social cohesion (SC) 1.3
(0.
Trust (T) 0.2
(0.
Neighbourhood accessibility (NA) 0.5
(0.
Informal social control (ISC) 0.7
(0.
Social participation (SP) 0.0
(0.
Multilevel models including individual, household, and postcode levelsa95% confidence intervals were estimated using MCMC models.bAdjusted model for age, gender, unaffordable items, financial, and
models, both unadjusted and also adjusted for age,gender, employment status, deprivation, and finan-cial status. ‘Neighbourhood quality’ showed thehighest percentage of variation at postcode level,accounting for nearly a third of the total variation,an effect that remained almost unchanged afteradjustments. ‘Social cohesion’, ‘neighbourhoodaccessibility’, and ‘informal social control’ factorseach showed about 10% each of residual variationat postcode level, substantially less than at otherlevels. ‘Trust’ and ‘social participation’ at thepostcode level contributed much less to the totalvariation, especially the latter whose contribution tothe variance after adjustments was below 1%.Adjustments for individual characteristics generallyhad little effect on the percentages accounted for bythe postcode level, but did lead to a noticeable
estionnaire factors on perceptions of social capital and the built
riance attributable to level (SE, 95% CIa)
vels
ividual (N ¼ 926) Household (N ¼ 588) Postcode (N ¼ 51)
.15 (2.48) 9.54 (1.31) 12.53 (0.98)
26, 15.87) (7.06, 12.22) (10.76, 14.60)
4 (0.87) 7.96 (1.27) 14.48 (1.09)
10, 4.46) (5.54, 10.54) (12.50, 16.76)
9 (0.20) 2.66 ((0.38) 3.73 (0.29)
09, 0.85) (1.94, 3.44) (3.21, 4.34)
1 (0.22) 1.37 (0.29) 3.54 (0.27)
27, 1.12) (0.82, 1.96) (3.05, 4.11)
7 (0.25) 2.07 (0.35) 4.02 (0.31)
39, 1.35) (1.42, 2.79) (3.47, 4.65)
7 (0.07) 1.46 (0.22) 2.47 (0.19)
00, 0.23) (1.05, 1.90) (2.16, 2.88)
ividual (N ¼ 861) Household (N ¼ 559) Postcode (N ¼ 50)
1 (2.51) 10.42 (1.38) 11.57 (0.96)
91, 15.63) (7.80, 13.22) (9.82, 13.61)
9 (0.66) 8.49 (1.25) 12.19 (0.98)
36, 2.93) (6.13, 11.02) (10.41, 14.27)
0 (0.16) 2.64 (0.39) 3.33 (0.28)
00, 0.58) (1.90, 3.40) (2.82, 3.92)
8 (0.21) 1.42 (0.31) 3.34 (0.28)
25, 1.09) (0.83, 2.04) (2.83, 3.92)
2 (0.24) 2.30 (0.36) 3.65 (0.29)
35, 1.28) (1.63, 3.03) (3.12, 4.27)
3 (0.04) 1.50 (0.22) 2.33 (0.18)
00, 0.14) (1.07, 1.94) (2.00, 2.71)
.
employment status.
ARTICLE IN PRESSR. Araya et al. / Social Science & Medicine 62 (2006) 3072–30833080
reduction for factors representing social and cogni-tive aspects of the neighbourhood, such as ‘socialparticipation’, ‘trust’, and ‘social cohesion’.
Mental health and perceptions of the built and social
environment
Table 3 presents the results of the multilevelregression analysis of GHQ total score on the fivefactors derived from questionnaire. This was doneusing the GHQ total score as a continuous andbinary variable. Results are presented for unad-justed models and for models adjusted for age,gender, financial status, deprivation, and employ-ment status. In both unadjusted models, all factorsare significantly associated with high GHQ scores.When the models are adjusted for individualcharacteristics, only ‘social cohesion’ and ‘trust’remain significantly associated with both high GHQscores and GHQ caseness. In the continuous model,‘neighbourhood quality’ also remains significantlyassociated with high GHQ scores. Effect sizes in thebinary model are summarised by an odds ratio,corresponding to a change of 1 SD in the factorscore, while in the continuous model estimatedchanges in the mean GHQ score for a change of
Table 3
The association between GHQ-12 and perceptions of social capital and
Model Cha
in faUnadjusteda
Con
Factor Social cohesion (SC) �1.
Trust (T) �1.
Neighbourhood quality (NQ) �0.
Participation (P) �0.
Accessibility (NA) �0.
Informal social control (ISC) �0.
Adjustedb,c
Factor Social cohesion (SC) �0.
Trust (T) �0.
Neighbourhood quality (NQ) �0.
Participation (P) �0.
Accessibility (NA) �0.
Informal social control (ISC) �0.
Multilevel models including individual, household, and postcode levels
*Significant [coefficient72 (SE) approximately 95% CI excludes zero].aUnadjusted models N for individuals, household, and postcode levebAdjusted model for age, gender, unaffordable items, financial, andcAdjusted models N for individuals, household, and postcode levels
1 SD in the factor score are presented to demon-strate the magnitude of the effect of the factors.
Discussion
The design of our questionnaire was a compre-hensive attempt to assemble a wide array ofquestions relating to important factors included insocial capital definitions and studies on the qualityof the built environment. Principal componentanalysis grouped questions in factors similar to thetheoretical constructs we had envisaged. Ourneighbourhood quality factor captured almost athird of the variance at higher levels. Although alower proportion of the variance was explained athigher levels for variables representing key compo-nents of social capital, 10% of the variance forinformal social control and 6% for trust wasexplained at postcode level, a result in keeping withprevious research (Subramanian et al., 2003). Thus,our compositional measures seem to capture somecontextual characteristics of the social and builtenvironment. We also found that key componentsof social capital, such as trust and social cohesion,were the only factors significantly associated withGHQ scores after adjusting for individual variables.
the built environment in South Wales (2001)
nge in GHQ score for 1 SD
ctor score (SE)
Odds ratios (95% CI)
tinuous GHQ scores
Binary GHQ cases
059 (0.188)* 0.67 (0.56, 0.79)*
298 (0.186)* 0.58 (0.49, 0.70)*
917 (0.195)* 0.68 (0.57, 0.81)*
600 (0.184)* 0.86 (0.72, 1.03)
575 (0.186)* 0.81 (0.68, 0.97)*
393 (0.191)* 0.82 (0.69, 0.97)*
653 (0.178)* 0.74 (0.57, 0.95)*
730 (0.175)* 0.71 (0.55, 0.91)*
487 (0.172)* 0.78 (0.60, 1.01)
220 (0.168) 0.92 (0.73, 1.15)
267 (0.167) 0.81 (0.65, 1.00)
270 (0.170) 0.82 (0.64, 1.04)
.
ls (909,586, 50).
employment status.
(842, 553, 50).
ARTICLE IN PRESSR. Araya et al. / Social Science & Medicine 62 (2006) 3072–3083 3081
Sampling small geographical areas (postcode) fordata collection and interviewing all members of thehousehold in the selected addresses were strengthsof this study. Most of the questions we chose toinclude in our questionnaire had been used pre-viously in other studies (Kawachi et al., 1997;Lochner et al., 1999, 2003; Ross, 2000; Sampsonet al., 1997; Subramanian et al., 2003).
One of the aims of this study was to try todisentangle how much individuals, households, andpostcodes contributed to variations in the ques-tionnaire scores for the factors derived. In keepingwith other studies (Lindstrom, Merlo, & Ostergren,2002; Subramanian et al., 2003), we foundthat, even after accounting for individual socio-demographic variables, significant neighbourhoodvariation remained for most of our factors, inparticular for the factor depicting neighbourhoodquality.
It is likely that there might be better agreementamong neighbours on perceptions about the qualityof objective aspects of their built environment thanon perceptions of social features, such as trust orcohesion. It is also possible that social issues mightbe more sensitive to the influence of individualfactors, such as personality. People may find it hardto circumscribe their answers to issues, such as trustto specific geographical boundaries. Instead theiranswers might reflect to a larger extent theirpersonal social environment, which would involvefriends, colleagues, and distant relatives. However,in order to reduce this confusion, we explicitlyincluded a reference to the ‘area around here’ inmost questions to force the respondent to thinkabout where they live rather than on other moredistant geographical areas. Questions on socialparticipation did not refer to the area around therebut as to whether or not people participated insocial activities or organisations. Thus, it is possiblesome people could have answered to this questionthinking on participation in groups or activitiesoutside the area where they live and this mightexplain the little variation explained at postcodelevel.
In view of all these limitations and uncertainties,establishing with confidence how much any of thesefactors represented a compositional or contextualconstruct is rather difficult. As experts in this fieldhave argued, it may even be a rather artificialoversimplification that distracts our attention fromthe more important interactions between contextand composition (Kawachi & Berkman, 2003).
Places make people as well as people shape ourplaces (Macintyre et al., 2002).
One of our aims was to try to ascertain if therewas an association between social capital andmental health. Intuitively one might suppose thata pleasant, safe, and supportive environment wouldhelp to increase our sense of happiness andsatisfaction resulting in better mental health. Inkeeping with this, we found that there weresignificant associations between the GHQ-12 scoresand most factors in the unadjusted models. How-ever, the only statistically significant associationsthat remained after adjusting for individual vari-ables were for two key components of socialcapital—trust and social cohesion. Nevertheless,only a small proportion of the variance for thesetwo factors was to be found at higher levels,suggesting that these are not strongly contextualfactors. Equally, it is difficult to rule out anypotential response bias introduced by the mentalstate of the responders. However, one would expectthat if a response bias were the most likelyexplanation for this association this could haveaffected all questions in a similar way. Thus, it isintriguing to note that the informal social controlfactor did not show associations with mental health.The neighbourhood quality factor included severalitems depicting neighbourhood disorder and dis-advantage, such as graffiti and litter, features thathad been associated with depression previously(Ross, 2000). We expected that strong perceptionsof disorder in the neighbourhood would correlatestrongly with perceptions of lack of informal socialcontrol and increased symptoms of anxiety anddepression. However, we did not find evidence of anassociation between informal social control andmental health. It is possible that respondents couldhave decided to downplay their expectations whenappraising this aspect of their social environment asa way to protect themselves from despair (Sen,1992). But if this were correct we would need to beable to explain why individuals downplay someaspects of their environment and not others.
Although associations between the social envir-onment and mental health are possible, and we havefound some in this study, the complex mechanismsand pathways underlying these associations stillremain elusive and at present the source ofmuch conjecture (McKenzie, Whitley, & Weich,2002). Depressed people are unlikely to interactmuch with their neighbours regardless of where theylive and people living in places where there is little
ARTICLE IN PRESSR. Araya et al. / Social Science & Medicine 62 (2006) 3072–30833082
contact between neighbours are perhaps more likelyto feel downhearted and depressed. This study tookplace in just one small area in the South of Wales sosome caution should be exercised when attemptingto generalise our results. The sample was ofmoderate size and it is possible that the study failedto detect small effect sizes. This was a cross-sectional study and as such it cannot reveal thecausal direction of these associations but at least itprovides some insight suggesting potentially inter-esting mechanisms that can be tested in futureresearch. Equally, this design prevents us fromknowing how stable these features are over time(Blakely & Woodward, 2000) or any further detailsabout their spatial distribution (Macintyre & Ell-away, 2003). Further, refinement of measures ofsocial capital, especially developing some trulycontextual ones, is also needed. However, thesedifficulties and challenges must not paralyse ourefforts to make gradual progress in understandingbetter how complex social phenomena influencehealth. In the meantime, the little evidence thathas accumulated suggests that there might be sometrue association between social capital andmental health but we have to be cautious not to‘embrace’ too closely this rather intuitive connectionbetween the social and built environment andmental health.
Acknowledgements
We would like to thank N. Weaver, J. Patterson,T. Bell, and P. Jones for their contribution to thisstudy. We thank Gareth Williams and Ben Rolfe fortheir help in developing the questionnaire. Theresearch was supported by a joint grant from theUK Medical Research Council and the Engineeringand Physical Sciences Research Council (grantreference number G9900679).
References
Bain, K., & Hicks, N. (1998). Building social capital and reaching
out to excluded groups: The challenge of partnerships. Paper
presented at the CELAM meeting on the struggle against
poverty towards the turn of the millennium, Washington, DC.
Barton, H., Grant, M., & Guise, R. (2003). Shaping neighbour-
hoods: A guide for health, sustainability and vitality. London:
Spon Press.
Blakely, T. A., & Woodward, A. J. (2000). Ecological effects in
multi-level studies. Journal of Epidemiology & Community
Health, 54, 367–374.
Curtis, S., & Rees Jones, I. (1998). Is there a place for geography
in the analysis of health inequality? Sociology of Health &
Illness, 20(5), 645–672.
Cutrona, C. E., Russell, D. W., & Hessling, R. M. (2000). Direct
and moderating effects of community context on the
psychological well-being of African American women. Jour-
nal of Personality and Social Psychology, 79(6), 1088–1101.
Dalgard, O. S., & Tambs, K. (1997). Urban environment and
mental health: A longitudinal study. British Journal of
Psychiatry, 171, 530–536.
Duncan, C., Jones, K., & Moon, G. (1995). Psychiatric
morbidity: A multilevel approach to regional variations in
the UK. Journal of Epidemiology & Community Health, 49,
290–295.
Ellaway, A., Macintyre, S., & Kearns, A. (2001). Perceptions of
place and health in socially contrasting neighbourhoods.
Urban Studies, 38(12), 2299–2316.
Goldberg, D., & Williams, P. (1988). The user’s guide to the
general health questionnaire. Windsor: NFER-NELSON.
Goldstein, H. (1995). Multilevel statistical models. London.
Harpham, T., Grant, E., & Rodriguez, C. (2004). Mental health
and social capital in Cali, Colombia. Social Science &
Medicine, 58(11), 2267–2278.
Kawachi, I., & Berkman, F. (2000). Social cohesion, social
capital, and health. In F. Berkman, & I. Kawachi (Eds.),
Social epidemiology (pp. 174–190). New York: Oxford
University Press.
Kawachi, I., & Berkman, L. (2003). Neighborhoods and health.
Oxford: Oxford University Press.
Kawachi, I., Kennedy, B. P., & Glass, D. R. (1999). Social capital
and self-rated health: A contextual analysis. American Journal
of Public Health, 89(8), 1187–1193.
Kawachi, I., Kennedy, B. P., Lochner, K. A., & Prothrow-Stith,
D. (1997). Social capital, income inequality and mortality.
American Journal of Public Health, 87, 1491–1498.
Lindstrom, M., Merlo, J., & Ostergren, P.-O. (2002). Individual
and neighbourhood determinants of social participation and
social capital: A multilevel analysis of the city of Malmo,
Sweden. Social Science & Medicine, 54, 1779–1791.
Lochner, K. A., Kawachi, I., Brennan, R. T., & Buka, S. L.
(2003). Social capital and neighborhood mortality rates in
Chicago. Social Science & Medicine, 56, 1797–1805.
Lochner, K. A., Kawachi, I., & Kennedy, B. P. (1999). Social
capital: A guide to its measurement. Health & Place, 5,
259–270.
Macintyre, S., & Ellaway, A. (2003). Neighborhoods and health:
An overview. In I. Kawachi, & L. Berkman (Eds.),
Neighborhoods and health (pp. 20–41). Oxford: Oxford
University Press.
Macintyre, S., Ellaway, A., & Cummins, S. (2002). Place effects
on health: How can we conceptualise, operationalise and
measure them? Social Science & Medicine, 55, 125–139.
Macintyre, S., Maciver, S., & Sooman, A. (1993). Area, class and
health: Should we be focusing on places or people? Journal of
Social Policy, 22(2), 213–234.
McCulloch, A. (2003). An examination of social capital and
social disorganisation in neighbourhoods in the British
Household panel study. Social Science & Medicine, 56,
1425–1438.
McKenzie, K., Whitley, R., & Weich, S. (2002). Social capital
and mental health. British Journal of Psychiatry, 181(4),
280–283.
ARTICLE IN PRESSR. Araya et al. / Social Science & Medicine 62 (2006) 3072–3083 3083
Muntaner, C., & Lynch, J. (2002). Social capital, class gender and
race conflict, and population health: An essay review of
Bowling alone’s implications for social epidemiology. Inter-
national Journal of Epidemiology, 31, 261–267.
O’Brien Caughy, M., O’Campo, P., & Muntaner, C. (2003).
When being alone might be better: Neighbourhood poverty,
social capital, and child mental health. Social Science &
Medicine, 57, 227–237.
Pearce, N., & Davey-Smith, G. (2003). Is social capital the key to
inequalities in health? American Journal of Public Health, 93,
122–129.
Perkins, D., Meeks, J., & Taylor, R. B. (1992). The physical
environment of street blocks and resident perceptions of crime
and disorder: Implications for theory and measurement.
Journal of Environmental Psychology, 12, 21–34.
Pickett, K. E., & Pearl, M. (2001). Multilevel analyses of
neighbourhood socioeconomic context and health outcomes:
A critical review. Journal of Epidemiology & Community
Health, 55, 111–122.
Putnam, R. D. (1993). Making democracy work. Princeton, NJ:
Princeton University Press.
Rasbash, J., Browne, W., Goldstein, H., Yang, M., Plewis, I.,
Healy, M., et al. (2000). A user’s guide to MLwinN, Version
2.1. London: Institute of Education, University of London.
Reijneveld, S. A., & Schene, A. H. (1998). Higher prevalence of
mental disorders in socioeconomically deprived urban areas
in the Netherlands: Community or personal disadvantage?
Journal of Epidemiology & Community Health, 52, 2–7.
Ross, C. E. (2000). Neighborhood disadvantage and adult
depression. Journal of Health and Social Behaviour, 41(2),
177–187.
Sampson, R., Raudenbush, S., & Earls, F. (1997). Neighbor-
hoods and violent crime: A multilevel study of collective
efficacy. Science, 277, 918–924.
Sampson, R. J. (2003). Neighborhood-level context and health:
Lessons from sociology. In I. Kawachi, & L. Berkman (Eds.),
Neighborhoods and health (pp. 132–146). Oxford: Oxford
University Press.
Sen, A. (1992). Inequality re-examined. Cambridge, MA: Harvard
University Press.
Silver, E., Mulvey, E. P., & Swanson, J. W. (2002). Neighbor-
hood structural characteristics and mental disorder: Faris and
Dunham revisited. Social Science & Medicine, 55, 1457–1470.
Skapinakis, P., Lewis, G., Araya, R., Jones, K., & Williams, G.
(2005). Mental health inequalities in Wales, United Kingdom:
A multilevel effect of area deprivation. British Journal of
Psychiatry, 186, 417–422.
SPSS Inc. Statistical package for the social sciences (SPSS).
(11.0). 2004.
Steptoe, A., & Feldman, P. J. (2001). Neighbourhood problems
as sources of chronic stress: Development of a measure of
neighbourhood problems and associations with socio-eco-
nomic status and health. Annals of Behavioral Medicine, 23(3),
177–185.
Subramanian, S. V., Lochner, K. A., & Kawachi, I. (2003).
Neighborhood difference in social capital: A compositional
artifact or a contextual construct? Health & Place, 9, 33–44.
The World Bank Social Capital Thematic Group. (2002). In C.
Grootaert, D. Narayan, V. Nyhan-Jones, & M. Woolcock
(Eds.), Integrated questionnaire for the measurement of social
capital. Washington, DC: World Bank.
Townsend, P., Phillimore, P., & Beattie, A. (1998). Health and
deprivation: Inequality and the North. London: Routledge.
Tunstall, H. V. Z., Shaw, M., & Dorling, D. (2004). Places and
health. Journal of Epidemiology & Community Health, 58, 6–10.
Wainwright, N. W. J., & Surtes, P. G. (2003). Places, people, and
their physical and mental functional health. Journal of
Epidemiology & Community Health, 58, 333–339.
Weich, S., Blanchard, C., Prince, M., et al. (2002). Mental health
and the built environment: Cross-sectional survey of indivi-
dual and contextual risk factors for depression. British
Journal of Psychiatry, 180, 428–433.
Weich, S., Holt, G., Twigg, L., Jones, K., & Lewis, G. (2003).
Geographic variation in the prevalence of common mental
disorders in Britain: A multilevel investigation. American
Journal of Epidemiology, 157(8), 730–737.