BRITISH JOURNAL OF PSYCHIATRY (2007), 191, 500^505. doi: 10.1192/bjp.bp.107.039438

MentalMental health and quality of residential contextual measures of residential environ- ment quality and geographical accessibility environment of local services.

HOLLIE THOMAS, NIKKI WEAVER, JOANNE PATTERSON, PHIL JONES, METHOD TRUDA BELL, REBECCA PLAYLE, FRANK DUNSTAN, STEPHEN PALMER, GLYN LEWISand RICARDO ARAYA Sampling strategy The study was part of a research pro- gramme – Housing And Neighbourhood And Health (HANAH) – developed to in- vestigate the relationships between the built Background There is increasing The common mental disorders of depres- environment, the social and economic con- interest in the proposition that residential sion and anxiety can lead to substantial dis- text, and health. A cross-sectional survey ability. The possibility that characteristics was conducted in Neath Port Talbot environment can affect . of neighbourhoods, in addition to charac- County Borough in South Wales, UK. We Aims Tostudy the degree to which teristics of residents, can affect mental restricted the sampling to postcodes with health is of increasing interest (Macintyre at least three households. Of the 3972 post- common clusters et aletal,,1993). There are a number of sug- code units identified within the area by according to postcode units and gested neighbourhood or contextual char- means of the Postcode Address File, a stra- households.Toinvestigate whether acteristics that might affect mental health. tified random sample of 51 postcode units contextual measures of residential These include social capital, defined (after was selected using a probability of selection environmentquality and geographical Putnam, 1993) as the features of social proportional to their size. The average size organisation (such as networks, norms of selected postcode units was 20.5 ad- accessibility are associated with symptoms and trust) that facilitate coordination and dresses, although 20% contained 30 or of common mental disorder. cooperation for mutual benefit. However, more and the range was 3 to 86. Postcode neighbourhood effects might also act units were sampled from six strata to repre- MethodMethod Atotalof1058individualsagedAtotal of1058 individuals aged through the quality of the residential sent low (bottom 15%), medium (middle 16^75 years (response rate 66%) environment or its contribution to social 70%) and high (top 15%) areas of socio- participatedin a cross-sectional survey. cohesion, or the self-esteem felt by an indi- economic deprivation using Townsend The12-item General Health vidual. A few studies have measured con- scores (Townsend et aletal, 1988), in addition textual effects on mental health (Birtchnell to both urban and semi-urban areas. Questionnaire measured symptoms of et aletal, 1988; Aneshensel & Sucoff, 1996; Neath Port Talbot County Borough is common mental disorder. Dalgard & Tambs, 1997; Ross, 2000; the fourth most deprived of the 22 county EllawayEllaway et aletal, 2001; Steptoe & Feldman, boroughs in Wales, according to the Welsh ResultsResults Only 22% % (95% CI 0^6) ofofthe the 2001; Silver et aletal, 2002; Weich et aletal,, Index of Multiple Deprivation based on the unexplained variationvariationin in symptoms existed 2003; Wainwright & Surtees, 2004; 2001 census. However, overall, areas in our at postcode unit level, whereas 37% (95% MathesonMatheson et aletal, 2006; Fone et aletal, 2007),2007) sample were only a little more deprived CI 27^49) existed athousehold-level, but but almost all have relied on aggregated than the average for Wales and included a residents’ perceptions of their environment reasonable spread of Townsend score thepostcodeunit variationwasreducedto or census data (compositional data) instead values. Wales is somewhat more deprived zero after adjustments.There was little of independently measured contextual than England on average, but direct com- evidence to suggestthat residential quality characteristics such as residential quality parison is difficult as the Welsh and English or accessibility were associated with or geographical accessibility to local ser- Indices of Multiple Deprivation are not symptoms.symptoms. vices (McKenzie et aletal, 2002). Furthermore, comparable. the choice of the area level to be investi- A total of 1523 addresses were identified Conclusions There was substantial gated is also crucial. It has been argued that in the 51 postcode units. As the number of unexplained variation atthe household ecological associations are best explored addresses varied greatly between postcode using data for small areas (Curtis & Rees units, a sampling fraction of 0.7 was level but we could find no evidence of Jones, 1998), and the ‘home patch’ (Barton applied to each postcode unit with up to postcode unit variation and no association et aletal, 2003) is increasingly seen as a useful 36 addresses, giving a maximum of 25 with residential environmental quality or unit for urban design, yet many studies sampled addresses in these postcode units. geographicalaccessibility.Itislikely thatthe have investigated much larger areas. In the For postcode units with more than 36 UK, postcode units comprise on average addresses, 25 were chosen at random. Of psychosocial environment is more 15–20 addresses, and often define a single these 1523 addresses, 140 were not eligible importantthan the physical environment street. Our aim was to investigate the (e.g. commercial or empty properties) and a in relation to common mental disorder. amount of variation in symptoms of com- further 148 contained only occupants who mon mental disorder between postcode were outside the age limit of 16–75 years. Declaration of interest None. units and between households, and whether In Neath Port Talbot, there were about any such variation could be explained by 2.02 individuals per household on average.

500500

Downloaded from https://www.cambridge.org/core. 01 Oct 2021 at 17:39:33, subject to the Cambridge Core terms of use. MENTAL HEALTH AND ENVIRONMENT

Measures (green space or buildings) and density of Ta b l e 1 Characteristics of the sample Self-administered questionnaire housing. Given the different nature of these constructs, it was decided that an overall All residents aged 16–75 years in each CharacteristicCharacteristic nn (%)(%) score would require different items to be sampled household were asked to complete given different weights. For example, the Gender the questionnaire survey. Questionnaires presence of burnt-out cars would probably Male 474474(45.0) were left at 887 households (72% of the be given more importance than the exis- FemaleFemale 579(55.0) eligible sampling frame). Questionnaire dis- tence of recreational space. In order to Working status tribution began on 15 May 2001 and was obtain these weights we conducted a sep- Employed 509 (48.7)(48.7) completed on 5 August 2001. Common Seeking work 34 (3.3)(3.3) arate survey of a random sample of 150 mental disorder was measured by the 12- CarerorlookingafterCarer or looking after residents from the Neath Port Talbot item General Health Questionnaire (GHQ;(GHQ; children/house 102(9.8) County Borough’s citizens’ panel, in which Goldberg & Williams, 1988), with a score Student or on training scheme42 (4.0) they were asked the degree to which the of 3 or more used for case definition. The Retired 232 (22.2)(22.2) presence or absence of each characteristic survey included additional self-administered Permanently unable to work126 (12.0) was felt to be desirable or undesirable. A questions regarding social capital, social Financial situation questionnaire was posted and 97 (65%) cohesion, perceptions of the local area, Living comfortably 182(17.5) completed and returned it. The survey was Doing all right 387(37.3) and individual-level socio-demographic and also posted on the local authority staff Just about getting by 336(32.4) socio-economic variables. website, and a further 37 responses were re- Finding it difficult 82(7.9) The variables listed in Table 1 were ceived from members of staff. The results Finding it very difficult51 (4.9) used in further analysis. Financial situation from the survey were used to generate an Unaffordable lifestyle items (0^9) was assessed using the question, ‘How well integer weight between 1 and 3, based on 050 55454(57.0) do you feel you are managing financially the median value of the responses to the 1^2 199 (20.5)(20.5) these days?’ The ‘unaffordable lifestyle’ citizens’ panel survey. The scores were 3^53^5 155(16.0) items were ‘keep household warm’, ‘keep multiplied by the weight and summed to 6^9 63 (6.5)(6.5) house damp-free’, ‘keep house in decent give a total score. The decision to use inte- Proportion of household income state of decoration’, ‘replace worn-out from benefits ger weights was to simplify the use of the furniture’, ‘have friends and family to your NoneNone 477(48.6) scale (Dunstan et aletal, 2005). A high overall home for a drink or meal at least once a Very little 154(15.7) score indicated an area of general low month’, ‘have a week’s annual holiday Quarter 83 (8.5)(8.5) quality, with a greater number of negative away from home’, ‘have new rather than Half to three-quarters109 (11.0) or undesirable features. In order to perform second-hand clothes’, ‘eat meat, chicken, AllAll 159(16.2) a sensitivity analysis we also calculated a fish or vegetarian equivalent at least every Duration of residence in area residential quality score before application 551year1year 41 (3.9)(3.9) second day’ and ‘eat fresh fruit and vegeta- of weights.ofweights. 1year1year 43(4.1) bles every day’. The response ‘would like to The total REAT score had a range from 2^5 years2^5years 143143(13.5) but can’t afford it’ was coded as 1 for each 0 to 68. For the 51 sampled postcode units 6^9 years 100(9.5) of these items and the total score was there- in the borough of Neath Port Talbot, the 10+ years 726 (68.6)(68.6) fore between 0 and 9. These items are from REAT score ranged from 8 to 46 (mean Property type the Breadline Britain surveys (Gordon & 23.9, s.d.23.9,s.d.¼8.3). We analysed the data from Detached 191(18.1) Pantazis, 1997). Overcrowding was a self- REAT as a continuous variable, but for the Semi-detached 445(42.1) reported item. Type and age of property analyses presented in this paper total REAT End-terraceEnd-terrace 94 (9.0)(9.0) were obtained from the Welsh School of Mid-terraceMid-terrace 270 (25.6)(25.6) score was split into thirds of the distribu- Architecture database for Neath Port Talbot. FlatFlat 55 (5.2)(5.2) tion (scoretion(score 5521.0, 21.0–27.5, 28+). The Age of property reliability of REAT scores was assessed by Residential Environment Assessment Tool Pre-1919 328(31.1) comparing them with ratings by a second 1919^1944 132 (12.5)(12.5) The Residential Environment Assessment observer. Twenty-four of the 28 items had 1945^19641945^1964 249(23.6) Tool (REAT; Dunstan et aletal, 2005) was a kappa statistic of more than 0.8. The 1965^19801965^1980 200200(19.0) designed to measure directly the observable lowest value (kk¼0.58) occurred for the con- Post-1980 146(13.8) characteristics of urban residential environ- dition of the paths, whereas for the mainte- Household crowding ment. Full details of the scale and its devel- nance of shared space kk¼0.67. Kappa Too crowded 132(12.6) opment are provided by Dunstan et aletal values for the density of housing and main- Just right 837(79.6) (2005). Residential environmental assess- tenance of houses were 0.7–0.8. The intra- Too big 83(7.8) ments of each postcode unit were under- class correlation coefficient for the total Area of residence taken over 3 days at the end of June 2001 score was over 0.9. Semi-urbanSemi-urban 663 (62.7)(62.7) by four raters. UrbanUrban 39537.3 Area socio-economic deprivation The 28 environmental characteristics Geographical accessibility scores (Townsend Score) rated included property vandalism, stray The locations of a range of facilities were Least deprived 20420419.3 dogs, presence of hedges and fences, garden provided by Neath Port Talbot County Mid deprived 71367.4 and property maintenance, presence of re- Borough Council and were mapped on a Most deprived 14113.3 creational space, the predominant outlook geographical information system. Facility

501501

Downloaded from https://www.cambridge.org/core. 01 Oct 2021 at 17:39:33, subject to the Cambridge Core terms of use. THOMAS ET AL

categories were leisure (e.g. cinema, public the data were analysed at a single level 25 and the total number of respondents house, children’s play park); sports (e.g. (again allowing for clustering) in order to per postcode unit ranged from 6 to 47. swimming pool, sports centre, playing help inform inclusion of confounding vari- The socio-demographic and socio- field); transport (e.g. bus stop, train station, ables in the multilevel models. Together economic characteristics of the 1058 study cycle paths); shopping (e.g. post office, with the variables in Table 1, the following participants are presented in Table 1. The local shop, pharmacy); and public services variables were investigated for an indepen- average age of the sample was 46.0 years, (e.g. general practice, community centre, dent association with GHQ case status: and 55% were women. Just under half of school). The citizens’ panel survey also marital status, having children at home, the participants were employed and 22% asked residents’ views on suitable levels of car ownership,housing tenure, household were retired. Over half reported their finan- access. These results, together with existing monthly income, total floor area of property, cial situation as either ‘comfortable’ or ‘all indices (Barton et aletal, 2003), allowed each council tax band of property and urbanicity. right’, similarly, 57% reported no lifestyle facility to be allocated to one of four cate- Multilevel modelling (Goldstein, 1995) item that they desired but were unable to gories reflecting priority of importance: used MLwiN version 1.10 software (Insti- afford. However, 13% were finding their financial situation either difficult or very (a)(a)categorycategory 1: nearest bus stop, local shop, tute of Education, University of London, pharmacy;pharmacy; UK). All analyses using multilevel model- difficult, and 16% reported that all of their ling excluded the seven postcode units with household income was derived from bene- (b)(b)categorycategory 2: general practice, post office, five or fewer replies. Sensitivity analyses fits. The majority had lived in their area cycle path, primary school, children’s were also completed, excluding 318 house- for 10 years or more. The most common play park;playpark; holds with only one response per household type of housing was semi-detached or (c)(c)categorycategory 3: playing field, public house, (while also excluding postcode units with mid-terrace, with a wide range of ages of supermarket, community centre, chil- five or fewer replies) to check the robust- property. Approximately 13% of parti- dren’s nursery, bus station, secondary ness of the estimate of residual variation cipants reported overcrowding in their school, train station, swimming pool, in GHQ symptoms at the household level. house. As a consequence of the sampling sports centre, restaurant; A simple variance components null strategy, the majority of respondents lived (d)(d)categorycategory 4: cinema, non-food stores, model to estimate the residual variation at in semi-urban areas, and 13% lived in the bowling green, tennis courts. postcode unit, household and individual le- most deprived areas. vels was fitted first using GHQ scores both Each of the four categories was then as- as a continuous total score and a binary Variance components null model signed a distance band indicating good, fair outcome (GHQ case v.v. non-case). Analyses for symptoms of common mental and poor levels of geographical accessi- involving the continuous outcome were disorderdisorder bility. Category 1 facilities were assigned based on a normally distributed multilevel We estimated that approximately 2% (95% distances of less than 300 m (good), 300– model whereas those involving the binary CI 0–6) of the unexplained residual varia- 500 m (fair) and over 500 m (poor); for outcome were based on a binomial multi- tion in symptoms of common mental disor- category 2 the distances were less than level model using a logit link function. der was at the postcode unit level, 37% 600 m, 600–800 m and over 800 m; for Markov chain Monte Carlo (MCMC) (95% CI 26–49) at the household level category 3 they were less than 800 m, procedures using Gibbs sampling generally and 61% at the individual level (Table 2). 800–1900 m and over 1900 m for category provide more accurate parameter estimates More than a quarter of the sample 4 they were less than 1300 m, 1300– (Rodriguez & Goldman, 2001) and these (26.5%, 95% CI 23.5–29.4) were scored 1900 m and over 1900 m. were used throughout. Individual-, house- as cases on the 12-item GHQ. The null An automated process using a geogra- hold- and postcode-level predictors of model for the binary GHQ outcome led to phical information system calculated the GHQ were then added to the model as similar estimates of percentage residual distance from the nearest facility to each fixed effects in a cumulative manner and variation of postcode unit, household and postcode unit, allocating a score of 2 for changes in variance were noted. individual level (Table 2). Although these good, 1 for fair and 0 for poor accessibility. results were consistent with those using The scores were summed to create a geo- GHQ score as a continuous outcome, the graphical accessibility score for each post- RESULTSRESULTS MCMC modelling of this binary outcome code unit with a range of 0–46, with Characteristics of sample proved difficult as convergence was slow, higher scores indicating better levels of ac- A total of 1058 questionnaires were re- with high levels of autocorrelation in the cessibility. The geographical accessibility turned, giving a response rate of 66%. series that gave the posterior distributions scores ranged from 16 to 42 for the 51 These 1058 individuals were clustered in of the variances at the three levels. This sampled postcode units. We took a similar 647 households (household response rate suggested that estimates were unreliable. approach towards analysis as described 73%) within 51 postcode units. Seven post- We therefore decided to restrict further above for the REAT score. code units included replies from five or few- analyses to models using GHQ total er participants; after excluding these symptom score as the outcome variable. Statistical analyses postcode units the multilevel analyses were Sample characteristics and prevalence of based on 1042 individuals nested within Including individual, household common mental disorder were derived 634 households within 44 postcode units. and postcode unit characteristics using commands in Stata version 6.0 for The number of respondents per household in the model Windows to allow for clustering by post- ranged from 1 to 5, the number of house- Individual, household and postcode unit- code unit. Prior to multilevel modelling holds per postcode unit ranged from 4 to level fixed effects were added to the null

502502

Downloaded from https://www.cambridge.org/core. 01 Oct 2021 at 17:39:33, subject to the Cambridge Core terms of use. MENTAL HEALTH AND ENVIRONMENT

Ta b l e 2 Comparison of unexplained variance in GHQ total symptom score and prevalence of GHQ case status at postcode unit, household and individual level

GHQ total score11 GHQ casecaseGHQ 22

Variance(s.e.) Percentage variance(95% CI)Variance (s.e.) Percentage variance(95% CI)

Model 1 (null) Postcode 0.54 (0.58)(0.58) 1.71.7 (0.01^6.2)(0.01^6.2)11 0.06 (0.09)(0.09) 1.21.2 (0.02^5.6)(0.02^5.6) Household 12.09(1.87) 37.3 (26.4^48.9)(26.4^48.9) 2.082.08 (0.76)(0.76) 38.338.3(15.9^70.0) IndividualIndividual 19.78(1.48) 61.0(52.7^70.5) 3.29 60.5 Total 32.41 5.43

GHQ,GHQ,General General Health Questionnaire. 1. Results for GHQ total score based on Markov chain Monte Carlo procedures using Gibbs sampling with gamma priors for variance parameters, 25 000 monitoring iterations after 1000 burn-in iterations. 2. Results for GHQ case (binary outcome) based on Markov chain Monte Carlo procedures using Metropolis Hastings sampling with gamma priors for varianceceparameters,100 parameters,100 000 monitoring iterations after1000 burn-in iterations.

model using GHQ symptom score as the the estimates for the null model were post- level remained almost identical after adding continuous outcome (Table 3). Inclusion code unit variance 0.45% (95% CI 0.0– potential explanatory household factors. of the individual-level exposures reduced 2.78), household variance 32.4% (95% CI ChandolaChandola et aletal (2003) reported a similar the total residual variation and also the 22.0–44.2) and individual variance 67.1% finding when attempting to account for percentage of residual variation at postcode (95% CI 58.1–77.7) when excluding house- the 20% of total variation in self-rated gen- level, whereas the percentage residual var- holds with single responses. eral health attributable to households. iation at household and individual level in- We also performed analyses using a Household variation in mental health creased slightly. Further inclusion of either residential environment score in which the suggests that who you live with is more household or postcode unit exposures had weights had not been used. The correlation important than the internal or external en- a less dramatic effect on the total amount between this unweighted REAT score and vironment in its effect on mental health. of residual variation and did not greatly weighted REATwas 0.98. We had very si- Other possible explanations include a alter the estimate of percentage residual milar and non-significantnon-significant results using this household effect on perceptions and ex- variance attributable to each level. alternative residential score. pectations of the external residential environment. The physical environment of Residential environment quality the home might also affect self-esteem and and geographical accessibility DISCUSSION psychological well-being either directly, or indirectly through a lack of social contact The quality of the residential environment The most striking result suggests that if the home environment inhibits visits from was not statistically significantly associated approximately 37% (95% CI 25–50) of friends and family. with symptoms of common mental disorder the variation in symptoms of common Our results suggest little variation in (Table 4) although symptoms were less mental disorder is explained at the house- symptoms of common mental disorder at common in areas with lower REAT scores hold level, whereas there appeared to be the postcode unit level. The confidence in- (more attractive areas). After adjusting virtually no variation across postcode units terval suggests that the maximum variation for the individual-level variables these once individual characteristics had been ta- compatible with our data could be approxi- differences were reduced. The geographical ken into account. Few previous household mately 2.4% after taking account of indi- accessibility score was not statistically sig- surveys of mental health have been able to vidual and household factors. We are not nificantly associated with GHQ symptom quantify the variation between households aware of any previous research in the UK score either before or after adjusting for because they have sampled one individual that has used postcode units in a multilevel individual-level variables (Table 4). The per household. Weich et aletal (2003) esti- model to assess the area-level variation in geographical accessibility of leisure and mated that 14% of the variation in preva- common mental disorder. However, pre- entertainment facilities was most strongly lence of common mental disorder vious findings from larger areas suggest associated with GHQ score, but the asso- occurred at the household level in the that between 1% and 3% of the total var- ciation was not statistically significant British Household Panel Survey. Our re- iance in common mental disorder can be either before or after adjustment for sults suggest a stronger association in attributed to differences among UK regions individual-level variables. GHQ score between individuals living in (Duncan(Duncan et aletal, 1995), Welsh unitary autho- the same household in this much smaller rities (Skapinakis et aletal, 2005), UK electoral Sensitivity analysis area. We do not have any satisfactory wards (Weich et aletal, 2003; Wainwright & The sensitivity analysis (excluding 318 explanation for this difference, but one Surtees, 2004; Fone et aletal, 2007), census households with only one response) was might expect that these characteristics tracts in urban areas of Canada (Matheson based on 687 individuals within 329 house- would vary between different geographical et aletal, 2006) and boroughs of Amsterdam holds within 45 postcode units, and the locations in the UK. (Reijneveld & Schene, 1998). results were comparable with those re- The proportion of unexplained variance It is difficult, though, to exclude the ported in Tables 2 and 3. For example, in common mental disorder at the household possibility that larger contextual effects

503

Downloaded from https://www.cambridge.org/core. 01 Oct 2021 at 17:39:33, subject to the Cambridge Core terms of use. THOMAS ET AL

Ta b l e 3 Effect of inclusion of individual, household and postcode unit fixed effects on the unexplained variance boundaries) than between postcode sectors in General Health Questionnaire (GHQ) symptom score (designed to include similar numbers of addresses but not delineated by natural boundaries) in Amsterdam, although the GHQ symptom score results were not always statistically signifi- Variance(s.e.) Percentage variance(95% CI) cant. Most research on contextual effects, including our own, has been based on ad- Model 1 (null) ministrative rather than geographic classifi- Postcode unit 0.54 (0.58)(0.58) 1.71.7 (0.01^6.2)(0.01^6.2)11 cations and might therefore fail to detect Household 12.09(1.87) 37.337.3 (26.4^48.9)(26.4^48.9) effects because of this. IndividualIndividual 19.78(1.48) 61.061.0 (52.7^70.5) Total 32.41 Quality of residential environment Model 22Model 22 and geographical access Postcode unit 0.06 (0.11)(0.11) 0.3 (0.004^1.6) We did not find that the contextual mea- Household 8.77(1.43) 37.1 (25.5^49.3)(25.5^49.3) sure of residential environment quality IndividualIndividual 14.81 (1.20)(1.20) 62.6 (53.5^73.2) was associated with symptoms of common Total 23.64 mental disorder. Weich et aletal (2002) did(2002)did Model 33Model 33 report that some independently rated household and neighbourhood characteris- Postcode unit 0.07 (0.12)(0.12) 0.3 (0.004^1.8) tics were associated with the prevalence of Household 8.37 (1.42) 36.536.5 (24.8^49.1)(24.8^49.1) common mental disorder, although multi- IndividualIndividual 14.46(1.21) 63.263.2 (53.4^74.1) level models were not used to analyse their Total 22.8922.89 data. After adjustment, deck access proper- Model 44Model 44 ties, age of property and private gardens Postcode unit 0.08 (0.15)(0.15) 0.250.25 (0.004^2.4)(0.004^2.4) were associated with common mental dis- Household 8.56(1.46) 37.1 (25.1^49.7)(25.1^49.7) order. In Chile, Araya et aletal (2007) used a IndividualIndividual 14.41 (1.22)(1.22) 62.5 (52.9^73.9) method analogous to ours and did find an Total 23.05 association between the physical environ- ment in large areas of Santiago and mental 1. Confidence limit does not cross zero as % variance cannot hold negative values. 2. Model1plus individual exposures (gender, age, working status, financial status, unaffordable items). health. However, the circumstances in 3. Model 2 plus household exposures (proportion income from benefits, crowding in house, level of social support). Chile differ in many respects from those 4. Model 3 plus postcode unit exposures (socio-economic deprivation category,Residential Environment Assessment in the UK, and socio-economic differences Tool score).score).Tool are more marked. Geographical accessibility of local facil- influence mental health. The choice of the fact that any inaccuracies in exposure ities was not associated with symptoms of geographical category will affect the results measurement will be more noticeable in common mental disorder. We are not (Blakely & Woodward, 2000) and a small areas (Jarman, 1997). Reijnveld et aletal aware of any previous studies of such a balance is needed between defining an area (2000), for example, found more variation measure of access. Thomson et aletal (2003)(2003) small enough to indicate a homogeneous between neighbourhoods (areas with simi- assessed, using qualitative methods, the community while taking into consideration lar types of building delineated by natural health impact of a new swimming pool in Ta b l e 4 Difference in GHQ symptom score categorised by quality of residential environment and geographi- Glasgow, UK. The residents, especially cal accessibility scores mothers of young children, reported mental health benefits from the social contact encouraged by the pool. These data illus- Mean difference (95% CI)11 trate that the relationship between the pro- Unadjusted Adjusted22 vision of local facilities and mental health is complex and inadequately summarised by REAT simple measures of distance. Likewise, a Total scorescoreTotal 5521.0 ReferenceReference Reference variety of other social and monetary factors can affect access. Total score 21.0^27.5 0.20 ((771.12 to 1.16) 770.18( 771.07 to 0.72) Total score 28+ 0.54( 770.58 to 1.66) 0.060.06 ((770.88 to 0.96) Geographical accessibility Strengths and limitations Total scorescoreTotal 5525.025.0 Reference Reference Our study used independent measures of Total score 25.0^31.0 0.260.26 ((770.82 to 1.34)0.23 ( 770.82 to 1.34) residential quality and geographical accessi- Total score 32+ 770.89 ((772.11to2.11 to 0.29) 770.21( 771.16 to 0.75)0.75)1.16 bility, and carried out an analysis that took the hierarchical structure of the data into GHQ,General Health Questionnaire; REAT,Residential Environment Assessment Tool. account. The multilevel models allowed us 1. Values are the mean difference between the reference category and the other categories of REATor geographical accessibility score. to estimate unexplained variation at higher 2. Adjusted for gender, age, working status, financial situation and number of unaffordable lifestyle items. levels and allowed us to model higher-level

504504

Downloaded from https://www.cambridge.org/core. 01 Oct 2021 at 17:39:33, subject to the Cambridge Core terms of use. MENTAL HEALTH AND ENVIRONMENT

variables such as REAT scores in a robust manner. However, the study investigated a HOLLIE THOMAS,THOMAS, DPhil, Department of Psychological Medicine, School of Medicine,; NIKKI WEAVER,WEAVER, MSc,JOANNE PATTERSON,PATTERSON,BSc,PHILJONES BSc, PHIL JONES,,PhD,CentreForResearchintheBuilt PhD,Centre For Research in the Built relatively small area, and perhaps there Environment,Welsh School of Architecture,Architecture,Cardiff Cardiff University;TRUDA BELL,BELL, MSc,REBECCA PLAYLE,PLAYLE,PhD, PhD, was not enough variation between our FRANK DUNSTAN,DUNSTAN,DPhil,STEPHENPALMER DPhil, STEPHEN PALMER,, FFPHM, Department of Epidemiology,Statistics and Public postcode units to detect using our methods. Health,Health,SchoolofMedicine,CardiffUniversity,Cardiff; School of Medicine,Cardiff University,Cardiff;GLYN LEWIS,LEWIS, FRCPsych, PhD,RICARDO ARAYA,ARAYA, The confidence intervals for our estimates MRCPsych, PhD, Division of Psychiatry,, Bristol, UK are also relatively wide and it is possible that we had insufficient statistical power Correspondence:Correspondence: Professor Glyn Lewis,Division of Psychiatry,University of Bristol,Cotham House, to detect any differences. Furthermore, Cotham Hill,Bristol BS6 6JL,UK.Tel: +44 (0)117 954 6796; fax: +44 (0)117 954 6672; email: Glyn.LewisGlyn.Lewis@@bristol.ac.uk our independent measures at the highest level concentrated on the physical aspects (First received 25 April 2007, final revision 20 July 2007, accepted 1August 2007) of the environment, mostly because we

thought that these could be measured Araya,Araya,R.,Montgomery,A.,Rojas,G., R., Montgomery, A., Rojas, G., et aletal (2007)(2007) McKenzie, K.,Whitley,K., Whitley, R. & Weich, S. (2002) SocialSocial reliably. The quality of the environment, Common mental disorders and the built environment in capital and mental health. British Journal of Psychiatry,, the presence or absence of graffiti and the Santiago,Chile.Santiago, Chile. British Journal of Psychiatry,, 190190, 394^401 181181, 280^283.,280^283. maintenance of properties also reflect Barton, H., Grant, M. & Guise, R. (2003) ShapingShaping Putnam, R. (1993) The prosperous community: social something about the psychosocial environ- Neighbourhoods: A Guide for Health, Sustainability and capital and public life. American Prospect,, 13,35^42. VitalityVitality.Spon. ment, but we did not measure these aspects Reijneveld, S. A. & Schene, A. H. (1998) HigherHigher directly. We therefore did not study some Birtchnell, J., Masters, N. & Deahl, M. (1988) prevalence of mental disorders in socioeconomically Depression and the physical environment: a study of deprived areas in the Netherlands: community or of the less easily measured constructs, such young married women on a London housing estate. personal disadvantage? Journal of Epidemiology and as social cohesion or social capital, at the British Journal of Psychiatry,, 153, 56^64.,56^64. Community Health,, 52,,2-7. 2-7.

postcode unit level. Blakely,T. A. & Woodward, A. J. (2000) Ecological Reijneveld, S. A.,Verheij, R. A. & de Bakker, D.H. effects in multi-level studies. Journal of Epidemiology and (2000) The impact of area deprivation on differences in Community Health,, 5454, 367^374. health: does the choice of geographical classification Contextual effects on mental matter?matter? Journal of Epidemiology and Community Health,, Chandola, T., Bartley, M.,Wiggins, R., et aletal (2003)(2003) healthhealth Social inequalities in health by individual and household 5454, 306^313. The most striking finding is the consider- measures of social position in a cohort of healthy people. Rodriguez, G. & Goldman, N. (2001) ImprovedImproved Journal of Epidemiology and Community Health,, 5757,, able clustering of common mental disorder estimation procedures for multilevel models with binary 56^62. response: a case study. Journal of the Royal Statistical within households, in contrast to the rela- Curtis, C. & Rees Jones, I. (1998) Is there a place for Society Series A,, 164164, 339^355.,339^355. tively tiny or non-existent clustering at geography in the analysis of health inequality? Sociology Ross, C. E. (2000) Neighborhood disadvantage and postcode level. It seems likely that the most of Health and Illness,, 2020, 645^672.,645^672. adult depression. Journal of Health and Social Behaviour,, important contextual influence on common Dalgard,O. S. & Tambs, K. (1997) Urban environment 4141,177^187.,177^187. mental disorder is that provided by the peo- and mental health: a longitudinal study. British Journal of Silver, E., Mulvey, E. P., Swanson, J.W. (2002) ple with whom an individual lives in a Psychiatry,, 171171,530^536. Neighborhood structural characteristics and mental household. Our research concentrated on Duncan, C., Jones, K. & Moon, G. (1995) PsychiatricPsychiatric disorder: Faris and Dunham revisited. Social Science and MedicineMedicine,, 5555,1457^1470. classifying the physical environment in morbidity: a multilevel approach to regional variations in the UK. Journal of Epidemiology and Community Health,, et aletal neighbourhoods, but our results suggest Skapinakis, P., Lewis, G., Araya, R., (2005) 49, 290^295.,290^295. Mental health inequalities in Wales,UK: multi-level that it is the psychosocial environment that Dunstan, F.,Weaver, N., Araya, R., et aletal (2005) AnAn investigation of the effect of area deprivation. BritishBritish we need to understand. Perhaps future re- observation tool to assist with the assessment of urban Journal of Psychiatry,, 186186, 417^422.,417^422. search should concentrate more upon the residential environments Journal of Environmental Steptoe, A. & Feldman, P.J.P. J. (2001) Neighbourhood PsychologyPsychology,, 2525, 293^305.,293^305. psychosocial characteristics of households problems as sources of chronic stress: development of a measure of neighbourhood problems and associations and neighbourhoods, rather than attempt- Ellaway, A., Macintyre, S. & Kearns, A. (2001) Perceptions of place and health in socially contrasting with socio-economic status and health. Annals ofofAnnals ing the easier task of assessing the physical neighbourhoods. Urban Studies,, 3838, 2299^2316. Behavioral Medicine,, 2323,177^185.,177^185. environment. Fone, D., Dunstan, F.,Williams,F.,Williams, G.,etal(2007) Thomson, H., Kearns, A. & Petticrew, M. (2003) Places, people and mental health: a multilevel analysis of Assessing the health impact of local amenities: a ACKNOWLEDGEMENTS economic inactivity. Social Science and Medicine,, 6464,, qualitative study of contrasting experiences of local 633^645.633^645. swimming pool and leisure provision in two areas of Glasgow.Glasgow. Journal of Epidemiology and Community Health,, We thank Neath Port Talbot County Borough Goldberg,Goldberg,D.&Williams,P.(1988) D. & Williams, P. (1988) A User’s Guide to the 5757, 663^667.,663^667. Council and its Executive for their crucial support General Health Questionnaire.nfer-Nelson. Townsend, P., Phillimore,Phillimore, P. & Beattie, A. (1988)(198 8) and cooperation in carrying out this study. Thanks Goldstein, H. (1995) Multilevel Statistical Models.Wiley..Wiley. Health and Deprivation: Inequality and the North. CroomCroom to Professor Gareth Williams and to Mr Ben Rolfe Gordon, D. & Pantazis, C. (1997) Breadline Britain in Helm. for their help in developing the questionnaire. The the 1990s1990sthe .Ashgate. Wainwright,Wainwright,N.W.J.&SurteesP.G.(2004) N.W.J. & Surtees P.G. (2004) Area and research was supported by a joint grant from the Jarman, B. (1997) Deprivation payments to general individual circumstances and mood disorder prevalence. UK Medical Research Council and the Engineering practitioners. Scores should be based on enumeration British Journal of Psychiatry,, 185185,,227^232. 227^232. and Physical Sciences Research Council, grant refer- districts, and payments should be phased in gradually. Weich, S., Blanchard, M., Prince, M., et aletal (2002) ence number G9900679. BMJBMJ,, 314314, 228^229.,228^229. Mental health and the built environment: cross-sectional Macintyre, S., Maciver, S. & Sooman, A (1993) Area,Area, survey of individual and contextual risk factors for REFERENCES class and health; should we be focusing on places or depression. British Journal of Psychiatry,, 180180, 428^433.,428^433. people?people? Journal of Social Policy,, 22, 213^234. Weich, S., Holt, G.,Twigg, L., et al (2003)(2003) Geographic Aneshensel, C. S. & Sucoff, C. A. (1996) TheThe Matheson, F., Moineddin, R., Dunn, J., et aletal (2006) variation in the prevalence of common mental disorders neighborhood context of adolescent mental health. Urban neighbourhoods, chronic stress, gender and in Britain: a multilevel investigation. American Journal of Journal of Health and Social Behaviour,, 3737, 293^310. depression.depression. Social Science and Medicine,, 6363, 2604^2616. Epidemiology,, 157, 730^737.,730^737.

505505

Downloaded from https://www.cambridge.org/core. 01 Oct 2021 at 17:39:33, subject to the Cambridge Core terms of use.