ARTICLE IN PRESS

0277-9536/$ - se

doi:10.1016/j.so

�Correspond

E-mail addr

(R. Thomas).

Social Science & Medicine 60 (2005) 2773–2783

www.elsevier.com/locate/socscimed

Housing improvement and self-reported mental distress amongcouncil estate residents

Richard Thomasa,�, Sherrill Evansb, Peter Huxleyb, Claire Gatelyb, Anne Rogersc

aSchool of Geography, University of Manchester, Oxford Road, Manchester, M13 9PL, UKbHealth Services Research Department, Institute of Psychiatry, King’s College, SE5 8AF London, UK

cNational Primary Care Research and Development Centre, University of Manchester, Manchester, M13 9PL, UK

Available online 26 January 2005

Abstract

This paper is concerned with how housing improvements instigated either publicly or privately influence the degree of

psychological stress reported by council estate residents in South Manchester. Stress is measured on the GHQ12 scale

containing standard symptomatic items. Potential sources of variation in this indicator are analysed within a

geographical setting where repeated samples of residents were drawn from two adjacent suburban council housing

estates before and after the implementation of a single regeneration budget (SRB) housing initiative in late 1999. The

residents of one of these estates (Wythenshawe) were targeted by this funding while those in the other (Mersey Bank)

were not. The latter, therefore, serve as a control for the effects of the enhanced incidence of housing improvement

activity promoted by this SRB. Regression analyses revealed that stress was raised significantly among the SRB

residents perhaps on account of the additional environmental nuisance they encountered. The experience of stress

among all residents, however, was dominated by measures of personal psychosocial risk and it is argued that future

regeneration initiatives should address the manifestation of these risks in the effort to achieve better mental health.

r 2004 Elsevier Ltd. All rights reserved.

Keywords: Manchester, UK; Council estate residents; Mental health; Urban regeneration

Introduction

The supposition that changing socio-economic cir-

cumstances might affect the mental health of a commu-

nity has been informed through the refinement of a

number of psychological constructions (Weich & Lewis,

1998; Marmot & Bobak, 2000; Rogers et al., 2001;

WHO, 2001). The initial ideas concerning this connec-

tion emphasised the importance of social structure

evidenced by the positive association between psychia-

tric morbidity and social disadvantage and adversity

e front matter r 2004 Elsevier Ltd. All rights reserve

cscimed.2004.11.015

ing author.

ess: richard.w.thomas@man.ac.uk

(Holingshead & Redlich (1958)) and, later, with

disparities in resources like income, occupation and

years in education (Bartley, Blane & Davey-Smith,

1998). A second strand of this debate, often referred to

as the psychosocial perspective, has noted the more

immediate contribution to the onset of mental distress of

precipitating personal factors like the experience of

stressful life events or changed social circumstances

(Dohrenwend & Dohrenwend, 1982; Brown & Harris,

1978). The evolution of this research has also witnessed

a switch from a preoccupation with the psychiatric

epidemiology elicited from individuals being treated for

mental health problems to a more recent focus on the

origins of symptomatic distress among the community

as a whole (Aneshensel & Sucoff, 1996; Elliott, 2000). A

d.

ARTICLE IN PRESSR. Thomas et al. / Social Science & Medicine 60 (2005) 2773–27832774

corollary of all this effort is that initiatives aimed at

improving the local environment might impact indirectly

upon the mental health of the recipients of these actions.

More particularly, the research reported here is

concerned with how housing improvements instigated

either publicly or privately influence the degree of

psychological stress reported by those most immediately

affected. The measurement of stress is made quantita-

tively by recourse to the General Health Questionnaire

12 point scale (GHQ12) containing standard sympto-

matic items (Goldberg & Williams, 1988). Potential

sources of variation in this indicator are analysed within

a geographical setting where repeated samples of

residents were drawn from two adjacent suburban

council housing estates in South Manchester before

and after the implementation of a single regeneration

budget (SRB) housing initiative in late 1999. The

residents of one of these estates (Wythenshawe) were

targeted by this funding while those in the other (Mersey

Bank) were not. The latter, therefore, serve as a control

for the effects of the enhanced incidence of improvement

activity promoted by this SRB. This design facilitates

the exploration of two general hypotheses. First, are the

GHQ12 scores of residents altered either negatively or

positively by the experience of housing improvement?

Second, are such possible outcomes further affected by

the more intense occurrence of regeneration activity in

the targeted estate? The assessment of these essentially

environmental stimuli also includes an examination of

their leverage relative to known psychological risks

factors for the susceptibility to mental distress.

The paper is organised as follows. Section 2 considers

recent research concerning the relationship between

environmental influences and the psychosocial risks for

mental distress. Section 3 describes the survey methods

together with the experimental design. The latter

consists of a sequence of multiple regression models

specified first to establish the leverage of the incidence of

housing improvements on GHQ12 scores and then to

include variation associated with the presence of known

psychosocial risks. Section 4 describes the results of

these statistical procedures. The discussion relates these

findings to the debate about the influence of spatially

tailored social policy interventions on community

mental health.

Environmental and individual risks for mental distress

The structural and psychosocial constructions of

mental distress are not mutually exclusive because both

attach importance to the generalised risks associated

with socio-economic deprivation. The former, however,

emphasises the influence of environmental deprivation

while the latter highlights the more immediate contribu-

tion of adverse personal experiences. A hierarchical

framework that integrates these perspectives has been

proposed by Stansfeld, Head and Marmot (1998) and

Stansfeld, Fuhrer, Cattell, Wardle, and Head (1999). In

their scheme, risks originate in the physical environment

through the gradation of material resources. In turn, this

variation affects the frequency with which psychological

risk factors are present in the local social environment.

Poverty, for example, might be expected both to

enhance the likelihood of unemployment and to exacer-

bate the personal consequences of such an event.

Alternatively, access to financial resources might act

positively to buffer these potential impacts. Whether or

not such circumstances are expressed as symptoms of

distress depends upon the perceptions of each indivi-

dual, which is presumed to serve as the last filter on the

pathway to poor mental health.

A method for operationalising the main elements of

this scheme has been outlined in Thomas et al (2002)

where the distinction was made between risk factors

defined by state variables and those that represent

discrete events. States refer to persistent effects that

gradually enhance the likelihood of psychosocial risk.

They may be chosen to reflect structural characteristics

of the physical environment, like living in a community

with persistently high levels of unemployment, or to

measures of personal vulnerability more specifically

linked to the prevalence of common mental health

problems. Instances of such vulnerability include people

with a limiting disability or those with parents who died

during childhood. By contrast, events refer to occur-

rences thought to precipitate the onset of symptomatic

distress like the encounter of psychosocial risks. In

addition, individual perceptions may be reflected as

events representing reactions to current life circum-

stances. Feelings of entrapment, for example, which

have been found to be crucial to the formation of

depression in both patient and non-patient series

(Brown, Harris, & Hepworth, 1995), may be evident

by the reporting of a restricted opportunity like being

unable to move home. Such feelings may also become

manifest in the failure to achieve personal goals like the

work aspirations of those currently unemployed (Nor-

denmark & Strandh, 1999). Powerlessness, loss and

humiliation characterise the final pathway to depression

and both naturalistic studies and controlled trials

suggest that psychosocial situations reflecting new hope

(or fresh starts) characterise a similar pathway to

remission from depression (Harris, 2001).

Relationships between variables representing these

constructs have been analysed using data surveyed from

the South Manchester council estate residents prior to

the Wythenshawe SRB (Thomas et al., 2002). The

analysis revealed that the variation in GHQ12 scores

correlated more strongly with psychosocial event vari-

ables than those representing environmental states. In

addition both GHQ12 and the event variables were

ARTICLE IN PRESS

1However, there are a number of reasons to suppose this

sample may well be representative. A relatively low response

rate does not necessarily indicate low representativeness (Cook,

Heath, & Thompson, 2000; Krosnik, 1999). We compared the

respondents gender, age, ethnicity, marital status and employ-

ment status with the 1991 census data for the area, and the only

major difference was in the age distribution; 19–24 year olds

were under-represented in the sample (9.5% compared to

15.4% in the census). National sample surveys also report fewer

respondents in this age group (Krosnik, 1999), but there has

also been a reduction in this age group in the population

between 1991 and 1999; 15–24 year olds were 12% of the UK

population in 1999 (ONS, 2001). According to figures for the

study wards, there were 10.8% in this age group by 1998

R. Thomas et al. / Social Science & Medicine 60 (2005) 2773–2783 2775

found to be inversely related to the age of the resident.

These results are consistent with previous studies that, in

particular, report a relatively low incidence of depres-

sion in old age (Paykel, 1991). Similarly, the less

frequent occurrence of the precipitators of stress (life

events, goal-setting and restricted opportunities) among

the elderly also helps to account for their comparatively

better mental health.

Housing is seen to be implicit in the structure of this

general framework as a state of the physical environ-

ment. Type of tenure, for example, has been viewed as a

relational resource linked to psychological characteris-

tics such as a sense of identity and aspirations which

provide the basis for security, mastery, self-esteem and

overall life satisfaction (Mcintyre, Hiscock, Kearns, &

Ellaway, 2001; Nettleton & Burrows, 2000). This

research has demonstrated that psychological character-

istics are distributed unequally among residents in ways

that are likely to have different health impacts on those

who rent property compared to owner-occupiers. The

distribution of housing stressors, like overcrowding and

dampness, together with perceptions of the local

environment (area reputation) have been shown to

partly explain the relationship between housing tenure

and both physical and mental health (Ellaway &

Macintyre 1998). In addition, there is more limited

evidence to suggest that such potential health risks can

be reduced by interventions targeted at those with

specific vulnerabilities. Favourable outcomes have been

found for improved housing conditions (Halpern, 1980)

and, in particular, for those re-housed on the grounds of

poor mental health (Elton & Packer 1986). Similar

findings have been reported for local interventions

aimed at the unemployed (Price, van Ryan, & Vinokur,

1992) and pregnant teenagers living in poverty (Olds,

Henderson, Tatelbaum, & Chamberlin, 1988).

The effects of area-based housing improvement

initiatives, however, are more ambiguous. While better

living conditions are anticipated to be beneficial in the

long term, the immediate effects of repair and construc-

tion are more likely to be sources of stress. Such

outcomes have led a recent review of the health effects of

housing improvement interventions (Thomson, Petti-

crew, & Morrison, 2001) to conclude that large scale

studies investigating their wider social context are

required. This claim, therefore, provides a rationale for

our investigation of the mental health consequences of

the Wythenshawe SRB.

(Manchester City Council, 2000). Furthermore, a comparison

between our sample and the MORI survey (MORI, 2001;

n ¼ 3480) in eight comparable SRB areas, for 15 variables

common to both studies, showed that the present study mean

for all variables was always within the MORI range and close to

the mean. The means for six variables (employment, length of

residence, long standing illness, having no car, crime rating, and

safety at night) were all within one percentage point of the

MORI mean.

Design

Survey details

The Wythenshawe SRB area was matched with

neighbouring wards forming the Mersey Bank council

housing estate using the index of deprivation supple-

mented by locally available statistics. An initial postal

survey to addresses randomly selected from the electoral

register was conducted in March 1999 prior to the SRB

initiative. The 2596 respondents to this survey repre-

sented a relatively low response rate (17%) which is not

uncommon for postal surveys in deprived areas, and in

line with the pilot study response from a neighbouring

non-study area (18%).1

This survey was repeated 22 months later and sought

information from those people who had responded to

the postal survey and who had not moved away from the

area. By follow-up, 522 baseline respondents had moved

and 1344 of the remainder (65%) replied. Socio-

demographic variables predicting failure to return the

questionnaire at follow-up for whatever reason were age

(younger), gender (men) and marital status (single). The

respondents in index and control areas were similar in

age and gender: mean age 51 years (sd 18.4) vs. 53 (17.0),

p ¼ 0:20; percentage male 52% in both areas (p ¼ 0:95).

However there were slightly higher proportions of non-

whites and single people in the control area (6% vs. 2%,

po0.001 and 31% vs. 21% for control versus index

area, po0.001 in both cases).

The intervention

This research was founded on the expectation that the

SRB would lead to more changes in the index than in the

control area. Very little SRB expenditure had been

committed up to the time of the baseline survey, but

over £2m was invested over the study period. Some of

the major developments are ‘complementary’, that is,

not part of this £2m, but part of the overall package of

ARTICLE IN PRESS

Table 1

Types and frequencies of housing improvement reported by residents at follow-up

Type Wythenshawe

residents in

receipt(WR)

Wythenshawe

residents not in

receipt (WNR)

Mersey Bank

residents in

receipt(MBR)

Mersey Bank

residents not in

receipt (MBNR)

Heating 157 547 73 567

Damp proofing 41 663 53 587

Lighting/electrics 48 656 48 592

Roofing 28 676 55 585

Bathroom 110 594 43 597

Plumbing 25 678 32 608

Kitchen 73 631 33 607

Windows 133 571 79 561

Other 38 666 33 607

At least one of the above (area %) 360(51.1) 344(48.9) 225(35.2) 415(64.8)

R. Thomas et al. / Social Science & Medicine 60 (2005) 2773–27832776

interventions in the area, including private sector

finance. The changes of tenancy from local authority

to housing trust status, in the Willow Park area of

Wythenshawe, were part of the associated complemen-

tary developments. The total investment from all sources

to the end of the study period was about £45 m.

The follow-up survey asked respondents to report any

housing improvements completed since 1999. The

frequencies of these improvements by type are listed in

Table 1 for both Wythenshawe and Mersey Bank. The

survey did not identify how these improvements were

funded and, therefore, it is not possible to identify the

exact numbers attributable to the SRB. The differences

between the frequencies in the two areas, however, are

indicative of how improvements in Wythenshawe were

affected by this area intervention. In this respect, 51.1%

of the Wythenshawe respondents were in receipt of at

least one type of housing improvement compared to

35.2% for those in Mersey Bank. Moreover, the

frequencies for the different types of alteration indicate

the main effect on housing conditions of the SRB was to

promote improvements to heating, bathrooms, kitchens

and windows.

Methods and measures

The analysis entails the estimation of a series of

regression models specified with variables collected from

both the baseline and follow-up surveys. In these

models, the dependent variable is the GHQ12 score

which is taken to be indicative of the degree of mental

distress. This measure is a 12-item screening instrument

covering a range of psychiatric symptoms such as

anxiety, depression, somatic and social dysfunction.

These items are ‘been unable to concentrate on whatever

you’re doing’, ‘lost much sleep over worry’, ‘felt you

were not playing a useful part in things’, ‘felt incapable

of making decisions about things’, ‘felt constantly under

strain’, ‘felt you couldn’t overcome your difficulties’,

‘been unable to enjoy your normal day to day activities’,

‘been unable to face up to your problems’, ‘been feeling

unhappy or depressed’, ‘been losing confidence in

yourself’, ‘been thinking of yourself as a worthless

person’ and ‘been feeling reasonably unhappy all things

considered’. Each of these items is presented to the

respondent on a four-point scale containing two

gradations of ‘problem’–‘no problem’. A value of one

is added to the GHQ12 score if the respondent rated

either of the ‘problem’ gradations and where a final

score greater than two indicates a mental health problem

with high sensitivity and specificity (Goldberg &

Williams, 1988).

Independent variables have been chosen to represent

individual susceptibility to mental distress and the

respondent’s status with regard to the SRB and the

receipt of housing improvement during the episode

between the surveys. The susceptibility of the individual

is reflected by variables measuring the age of the

respondent (AGE) and number of restricted opportu-

nities (RO) reported to be incident on the date of survey.

More specifically, RO was constructed as an indicator of

psychosocial risk and was measured on an eight-point

scale counting positive responses to the following items:

‘lacked money to enjoy life’, ‘like more leisure but

cannot’, ‘more active social life but unable to’, ‘wanted

to move but could not’, ‘wanted to improve living

conditions but could not’, ‘wanted to improve person

safety but could not’, ‘wanted to participate in family

activity but could not’, and ‘wanted help with health

problems but could not get it’.

The justification for this parsimonious choice of two

susceptibility indicators is based upon a previous

analysis of the baseline survey (Thomas et al., 2002)

that estimated regression relationships between GHQ12

ARTICLE IN PRESSR. Thomas et al. / Social Science & Medicine 60 (2005) 2773–2783 2777

and an array of variables representing facets of both

structural and psychosocial risk.2 Here, AGE and RO

were found to be the dominant explanatory factors for

the variation in GHQ12. In addition, RO was the one

psychosocial construct that correlated strongly with the

indicators of structural risk and, therefore, serves as a

useful proxy for this variation in the present reduced

analysis. Moreover, in this analysis, RO is always

measured as the count reported at the time of the

baseline survey. This specification is made to avoid the

possibility that the scoring of the constituent items of

RO might be affected by mental health problems

encountered after this date. For this reason, such

problems are anticipated to be manifest solely in the

value of GHQ12 reported at follow-up.

The potential effects of living in the SRB intervention

zone are represented by the variable AREA, which

comprises of the two categories Mersey Bank (MB) and

Wythenshawe (W—the targeted estate). These categories

are distinguished as a dummy variable where zero

indicates residence in Mersey Bank and unity residence

in Wythenshawe. Similarly, the types of housing

improvement listed in Table 1 are represented by a

separate categorical variable that includes the generic

label IMP in its title. For each of these variables a value

of unity indicates receipt (R) of the specified improve-

ment during the intervention episode, while zero

indicates non-receipt (NR). Among this set, the central

focus of the analysis is the ‘at least one’ category, which

distinguishes between those who received some kind of

alteration to their housing and those who did not.

The independent variables described above are the

main effects selected to explain the variation in GHQ12.

The inclusion of the two dummy variables in the design,

however, allows for the possibility of their interaction

with the ratio variables AGE and RO. The functional

form of the relationship between GHQ12 and AGE in

Wythenshawe, for example, might be significantly

different from that estimated for Mersey Bank. To test

for all such outcomes, the main effects AREA and IMP

are disaggregated into the four dummy categories

labelled WR, WNR, MBR and MBNR. Thus, WR

refers to the subset of respondents in Wythenshawe in

receipt of the specified improvement while WNR

denotes the complementary Wythenshawe residents

who were non-recipients. Membership of each dummy

category is denoted by unity (MBR ¼ 1 if the respon-

dent is both resident in Mersey Bank and in receipt of

2In addition to restricted opportunities, psychosocial risk in

this study was also represented by variables measuring the

incidence of negative life events, the frequency of goal setting

behaviours and the quality of life, while structural risk included

measures of individual socio-economic deprivation and perso-

nal vulnerability. The analysis was derived from the 2596

respondents to the baseline survey.

the improvement) and non-membership by zero (all the

other respondents). Then, the ratio variables are

themselves disaggregated according to these dummy

specifications. AGE, for example, can be partitioned

into the interaction variables AGE(WR), AGE(WNR),

AGE(MBR) and AGE(MBNR). Accordingly,

AGE(MBR) contains the ages of those respondents

who are members of MBR and zero otherwise. To

execute the regression analysis one of the interaction

categories is specified as a fixed effect against which the

significance of the relationships with the other categories

is assessed (Johnston, 1980). Throughout, this fixed

category is MBNR. The regression coefficients estimated

for the dummy categories (WR etc.) are intercepts (a)expressed in units of GHQ12, while those for the

interaction variables [AGE(WR) etc.] are b-values

describing the direction and strength of their relation-

ship with GHQ12.

The following analysis first examines regression

relationships involving the main effects and then tests

the significance of the interactions associated with the

AGE and RO variables. The presentation is cross-

sectional in style and entails a comparison of results

obtained from models where the dependent variable is

GHQ12 at baseline and those where this specification is

replaced by the score reported at follow-up.

Results

Main effects models

Bivariate regression statistics estimated by OLS for

the relationship of each main effect variable with

GHQ12 at both baseline (t1) and follow-up (t2) are

listed in Table 2. At both dates, AGE and RO display

highly significant relationships with GHQ12 with larger

adjusted R2 values estimated for the baseline variables.

RO is positively associated with GHQ12 and, typically,

explains about 15% of the variation in this score. By

contrast the reporting of symptoms of distress declines

with AGE and this relationship explains about 2% of

the variation in GHQ12.

At t1, the bivariate regression for the AREA effect

(b ¼ 0.278) is not significant (p ¼ 0.110). This b-value

refers to the difference between the mean GHQ12 score

in Wythenshawe [a(W, t1) ¼ 2.528] and the mean score

in Mersey Bank [a(MB, t1) ¼ 2.250] and this insignif-

icant prior outcome is some justification for our

selection of the Mersey Bank estate as a control for

the SRB intervention in Wythenshawe. At t2 the AREA

effect (b ¼ 0.314) is of modest significance (p ¼ 0:079)

and the mean GHQ12 score for Wythenshawe residents

(2.621) is greater than that estimated for Mersey Bank

(2.307). The bivariate regressions for the At least 1 IMP

effect are significant at t1 (b ¼ 0.411, p ¼ 0.019) and t2

ARTICLE IN PRESS

Table 3

Paired sample t–tests for the difference between mean GHQ12 scores at baseline (t1) and at follow-up (t2) for each categorical effect

Effect GHQ12(t1) mean GHQ12(t2) mean Mean difference (t2– t1) t–Statistic Exact significance

AREA:

MB 2.250 2.307 0.057 0.322 0.747

W 2.528 2.621 0.093 0.457 0.647

At least 1 IMP:

NR 2.217 2.309 0.092 0.620 0.535

R 2.628 2.681 0.053 0.121 0.904

Total 2.409 2.459 0.050 0.556 0.578

Table 2

Bivariate regressions for the main effects as predictors of GHQ12 at baseline (t1) and at follow-up (t2)

Effect a b Exact significance R2 Exact significance

(a) GHQ12(t1)

AGE 3.384 �0.028 0.000 0.023 0.000

RO(t1) 0.218 0.619 0.000 0.172 0.000

AREA:

MB 2.250

W 2.528 0.278 0.110 0.001 0.110

At least 1 IMP:

NR 2.217

R 2.628 0.411 0.019 0.003 0.019

(b) GHQ12(t2)

AGE 3.730 �0.024 0.000 0.017 0.000

RO(t1) 0.589 0.535 0.000 0.123 0.000

AREA: 2.307

MB 2.621 0.314 0.079 0.002 0.079

W

At least 1 IMP:

NR 2.309

R 2.681 0.372 0.039 0.002 0.039

Exact significance: for the AREA and At least 1 IMP effects this probability refers to the difference between the intercept terms, that is

b(W) ¼ a(W)–a(MB). These intercepts are the mean GHQ12 score for the given category.

R. Thomas et al. / Social Science & Medicine 60 (2005) 2773–27832778

(b ¼ 0.372, p ¼ 0.039). For both regressions, the mean

GHQ12 score for the receipt category (R) are greater

than the mean scores for those in the NR category. Since

membership of R refers to receipt of an improvement in

the episode after the baseline survey, the significant

difference estimated at t1 indicates those in the R

category maintained higher scores than those in NR

before any receipt of an actual housing improvement.

The regression statistics reveal differences between the

category effects at a specified survey date but do not test

for the change in the mean scores observed for a single

category on the these dates. The latter differences are

tested by paired sample T-statistics (Table 3). The results

are all insignificant and indicate a high degree of

stability between the mean GHQ12 scores observed for

each category over the study period. The same outcome

occurs for the mean scores for the entire sample (Total)

where the insignificant change in GHQ12 (0.050) is

positive over the interval from t1 to t2. The correlation

coefficient for the GHQ12 scores at t1 and t2 is 0.487

(p ¼ 0.000).

Models with interaction effects

The significant differences between the main cate-

gories at specific survey dates suggest the possibility of

their interaction with both RO and AGE. In this respect,

the regression model for the disaggregation of RO

ARTICLE IN PRESS

Table 4

Multiple regressions for AGE interactions with AREA�At least 1 IMP as predictors of GHQ12 at baseline (t1) and at follow-up (t2)

Effect a b Exact significance R2 Exact significance

(a) GHQ12(t1)

AREA�At least 1 IMP:

MBNR 2.813

MBR 4.613 1.800 0.018

WNR 4.168 1.35 0.066

WR 4.274 5.461 0.041

AGE(AREA�At least 1 IMP):

MBNR �0.015 0.092

MBR �0.039 0.084

WNR �0.034 0.160

WR �0.031 0.211

Model 0.028 0.000

(b) GHQ12(t2)

AREA�At least 1 IMP:

MBNR 2.868 0.396 0.612

MBR 3.264 1.586 0.037

WNR 4.454 1.736 0.019

WR 4.604

AGE(AREA�At least 1 IMP):

MBNR �0.013 0.137

MBR �0.013 0.984

WNR �0.038 0.077

WR �0.034 0.114

Model 0.021 0.000

Exact significance: for the AREA�At least 1 IMP effects this probability refers to the difference between the intercept term and the

corresponding fixed effect, that is b(.) ¼ a(.)–a(MBNR). For the AGE(AREA�At least 1 IMP) interactions the probability refers to

the difference term Db(.) ¼ b(.)–b(MBNR). In (b), for example, Db(WR) ¼ b(�0.034)–b(�0.013) ¼ �0.21.

R. Thomas et al. / Social Science & Medicine 60 (2005) 2773–2783 2779

according to the categorical effects revealed no sig-

nificant interactions with GHQ12 either at t1 or t2 (not

illustrated). The regression models for the AGE(AR-

EA�At least 1 IMP) interactions, however, yielded

more positive outcomes (Table 4). Here, the intercepts

(a) are the estimated value of GHQ12 when AGE ¼ 0.

The b-coefficients for the AGE interactions are all

negative and denote the decrement to the value of their

respective intercept that is consequent upon each

additional year of life. Accordingly, these intercepts

and b-coefficients describe the linear relationship be-

tween GHQ12 and AGE for each of the interaction

effects. These relationships are visualised in Fig. 1 for

the regression models (a) estimated at baseline (b) and

follow-up.

The functional form of these interactions is seen to

differ at the survey dates. At t1, the significance statistics

(Table 4a) distinguish between relationship obtained for

the MBNR category and the functional forms obtained

for the three remaining categories. The predicted values

of GHQ12 estimated for MBNR are lower than the rest

and are subject to a reduced b-coefficient for the

negative effect of AGE (Fig. 1a). At t2, however, the

inter-relations between the categories are more symme-

trical. Compared to those for both MB categories, the

equivalent functions for Wythenshawe are subject to

both higher intercept values (increased predicted

GHQ12) and steeper b-coefficients for the AGE effect

(Fig. 1b). Moreover, the intercept values estimated for

the non-receipt categories (MBNR and WNR) are less

than the values obtained for the corresponding receipt

categories (MBR and WR). Despite this more regular

outcome, the estimate of R2¼ 0.021 obtained for the t2

regression model explains less of the variation in

GHQ12 than the t1 model (R2¼ 0.028).

The preceding analysis has identified the separate

significance of the hypothesised sources of variation in

GHQ12 at both baseline and follow-up. The regression

model listed in Table 5 examines the effects of these

sources in combination by specifying RO and the

AGE(AREA�At least 1 IMP) interactions as the

predictors of GHQ12 at t2 (Table 5). The adjusted

value of R2¼ 0.125 compares with the estimate of

R2¼ 0.123 (Table 2b) obtained when RO was specified

as the single predictor of GHQ12(t2). Thus, RO

accounts for the majority of the variation in these scores

ARTICLE IN PRESS

(a)

(b)

Fig. 1. AGE(AREA�At least 1 IMP) interation plots against

(a) predicted GHQ12 at baseline (b) and at follow-up.

Table 5

Multiple regression for RO and the AGE(AREA�At least 1 IMP) i

Effect a b

AREA�At least 1 IMP:

MBNR 0.508 �0.009

MBR 0.499 1.176

WNR 1.684 1.392

WR 1.900

AGE(AREA�At least 1 IMP):

MBNR 0.001

MBR 0.004

WNR �0.020

WR �0.022

RO(t1) 0.514

Model

Exact significance: for the AREA�At least 1 IMP effects this probab

corresponding fixed effect, that is b(.) ¼ a(.)–a(MBNR). For the AGE

the difference term Db(.) ¼ b(.)–b(MBNR).

R. Thomas et al. / Social Science & Medicine 60 (2005) 2773–27832780

while the AGE interaction retains a degree of signifi-

cance. The functional form of the categorical relation-

ships, however, is altered in the presence of RO (Fig. 2).

In this last respect, the intercept terms in the

combined model refer to the predicted value of

GHQ12 for each category when both AGE and RO

are zero. For the Mersey Bank categories (MBR,

MBNR) neither their intercepts nor their b-coefficients

are significantly different from zero (Table 5). The

intercepts for the Wythenshawe categories (WR, WNR),

however, are both positive and retain similar negative

AGE gradients. The inclusion of RO in this specifica-

tion, therefore, subsumes the previously identified age

effect among MB residents and the lower predicted

GHQ12 values estimated for those in the NR categories.

These differences are evident from the Fig. 2, where the

Wythenshawe categories are those that display a similar

negative relationship with AGE.

For completeness, Table 6 provides statistics for

combined regression models where the IMP category

has been re-specified for each type of housing improve-

ment that was promoted by the SRB initiative. In

general, the statistics estimated in these regression

models are quite consistent with those obtained for the

At least 1 IMP specification. The degrees of model

explanation (R2) and contributions of the RO variable

are virtually identical and the intercepts their

b-coefficients for the Mersey Bank AGE interactions

are not significantly different from zero. The Wythen-

shawe categories maintain the negative AGE gradient

although the strength of this relationship varies across

the WR and WNR categories. In the Bathroom IMP

regression the strongest relationship is estimated

for the WR category whereas, for the other types of

nteractions as predictors of GHQ12(t2)

Exact significance R2 Exact significance

0.990

0.102

0.046

0.876

0.813

0.107

0.069

0.000

0.125 0.000

ility refers to the difference between the intercept term and the

(AREA�At least 1 IMP) interactions the probability refers to

ARTICLE IN PRESSR. Thomas et al. / Social Science & Medicine 60 (2005) 2773–2783 2781

improvement, this outcome is associated with the WNR

category.

Discussion

Better mental health was not a specific target of the

SRB intervention in Wythenshawe. Yet, our first ideas

about this matter were predicated upon the expectation

that such an extensive investment in the socio-economic

infrastructure would indirectly improve mental health

within this community. The analysis presented in this

paper was designed to investigate this broad hypothesis

with particular regard to the potential benefits of

Fig. 2. RO and the AGE(AREA�At least 1 IMP) interactions

plotted against predicted GHQ12 at follow-up.

Table 6

Regression models for each SRB housing improvements as predictor

IMP WR WNR

Heating a 0.706 1.956

ba�0.013 �0.024

Bathroom a 2.173 1.683

ba�0.025 �0.020

Kitchen a 0.982 1.858

ba�0.009 �0.022

Windows a 1.792 1.779

ba�0.018 �0.022

At least 1 a 1.900 1.684

ba�0.022 �0.020

Independent variables are RO and the AGE(AREA� IMP) interacti

Bold indicates a coefficient is significant at pr0.05. Italics indicates aaThe first four values in each row are age b�-coefficients, the fifth i

At least 1 IMP regression are also listed in Table 5 which provides a

housing improvement. The following interpretation of

the results, however, provides evidence that is often

counter to this expectation.

The regression results demonstrate in sequence how

housing improvement activity and indicators of pyscho-

social risk combine to affect the reporting of mental

distress by the surveyed residents. In this respect, the

GHQ12 scores reported by those in Wythenshawe SRB

zone are of special interest. In the bivariate regression at

baseline, their mean score is not significantly different

from that estimated for Mersey Bank residents while, at

follow-up, this difference is significant and positive. The

latter outcome persists both in the separate analysis of

the AGE interactions and in the combined model. One

implication of this difference is that the reporting of

mental distress among Wythenshawe residents was

disturbed by the implementation of the SRB. It is quite

possible these residents experienced a degree of stress

from local environmental nuisance that was attributable

to enhanced improvement activity on the estate. This

additional stress is more evident among younger

Wythenshawe residents and tends to dissipate in those

over 60 years of age.

The results obtained for the housing improvement

effect are more ambiguous. At baseline, for example,

those who subsequently received at least one type of

improvement reported additional symptoms of stress

irrespective of their residential estate. Such an outcome

could be interpreted as a structural effect related to

living in a house in need of repair. The persistence of this

enhanced stress at follow-up, however, might be

attributable to the disruption to home life engendered

by house improvement. The clearer delineation between

the R and NR categories obtained for the separate

analysis of the AGE interactions at follow-up (Fig. 1b)

would appear to add weight to the case for such

s of GHQ12 at follow-up

MBR MBNR RO(t1) R2

0.366 0.556 0.127

0.011 0.000 0.515

1.898 0.336 0.127

�0.028 0.005 0.518

�0.163 0.516 0.126

0.008 0.002 0.519

0.454 0.521 0.126

0.008 0.001 0.517

0.499 0.508 0.125

0.004 0.011 0.514

ons.

coefficient is significant in the range pr0.10, p40.05.

s the restricted opportunities b- coefficient. The statistics for the

key for the other regression results given above.

ARTICLE IN PRESSR. Thomas et al. / Social Science & Medicine 60 (2005) 2773–27832782

disruption. This distinction, however, is largely absent in

the combined model estimates when the restricted

opportunities variable is included in the specification.

A clearer interpretation of these improvement effects

might have been possible had the mean GHQ12 scores

for the categorical variables altered between the two

survey dates. Their stability over this interval (Table 3),

however, indicates the nuisance effect served only to re-

distribute these category means at each specific date.

Moreover, the local leverage of this environmental effect

on GHQ12 scores is small in comparison to that

estimated for the reporting of restricted opportunities.

Predicted GHQ12 at t2, for example, increases from

about 0 to 6 within the reporting range of RO (0 to 8

items) compared to the unit response exhibited by the

negative AGE interactions for Wythenshawe residents

(see Fig. 2).

Thus, the results of our analysis suggest better mental

health outcomes are associated with low personal

assessments of psychosocial risk (RO). In contrast, the

effects of housing improvement tend to be detrimental.

From the policy perspective, this evidence indicates

structural investment is unlikely to reduce the reporting

of symptoms of mental distress and, instead, a surer

strategy to achieve this aim is to devise interventions

that target the incidence of psychosocial risk. Such a

conclusion, however, must be tempered by limitations of

the present research design. The follow-up survey was

undertaken 22 months into the duration of the SRB and,

therefore, captured only the immediate impacts of this

intervention. The long-term impacts on residents of

these structural changes remain to be established.

Similarly, the relatively low response rates to our surveys

might influence the generality of our findings with

regard to the populations from which they were drawn.

Finally, the research presented here was complemen-

ted by in-depth interviews with a sub-sample of the

residents during the intervention episode. These findings

will be reported elsewhere (Rogers, Gately, Evans,

Huxley, & Thomas, 2005), however, some of the issues

they raised are pertinent to the current discussion. Many

of these respondents referred to a lack of opportunities

on the estates and, in particular, expressed concern

about the poor provision of play and leisure facilities for

children and youth, the reputation of the area, and

restrictions on travel, especially at night. They also

suggested the perceived benefits of housing improve-

ments were viewed as constituting cosmetic changes

when compared to the quality of private housing being

built within the same area. Others pointed out that

substantial alterations had been made by tenants

themselves and there was a fear of increased rental

charges which was a seemingly negative outcome to be

set against making improvements. Therefore, they

regarded the structural changes we measured as both

modest and insufficient to match their expectations for

the intervention. Accordingly, targeting such concerns

prior to the implementation of urban regeneration

initiatives might precipitate more decisive mental health

outcomes than those reported here.

Acknowledgement

The research in this paper was funded by ESRC

research award L128 25 1041 whose support we grate-

fully acknowledge.

References

Aneshensel, C., & Sucoff, C. (1996). The neighbourhood

context of adolescent mental health. Journal of Health and

Social Behaviour, 37, 293–310.

Bartley, M., Blane, D., & Davey-Smith, G. (1998). Introduc-

tion: beyond the Black Report. Sociology of Health and

Illness, 20(5), 563–578.

Brown, G. W., & Harris, T. O. (1978). Social oigins of

depression: a study of psychiatric disorder in women. London:

Tavistock Publications.

Brown, G. W., Harris, T. O., & Hepworth, C. (1995). Loss,

humiliation and entrapment among women developing

depression: a patient and non-patient comparison. Psycho-

logical Medicine, 25, 7–21.

Cook, C., Heath, F., & Thompson, R. L. (2000). A meta-

analysis of response rates in Web or internet-based surveys.

Educational and Psychological Measurement, 90(6), 821–836.

Dohrenwend, B. P., & Dohrenwend, B. S. (1982). Perspectives

on the past and future of psychiatric epidemiology.

American Journal of Public Health, 72, 1271–1279.

Ellaway, A., & Macintyre, S. (1998). Does housing tenure

predict health in the UK because it exposed people to

different levels of housing related hazards in the home or its

surroundings? Health and Place, 4, 141–150.

Elliott, M. (2000). The stress process in neighbourhood context.

Health and Place, 6, 287–299.

Elton, P. J., & Packer, J. M. (1986). A prospective randomised

trial of the value of re-housing on the grounds of mental ill-

health. Journal of Chronic Disease, 39, 221–227.

Goldberg, D. P., & Williams, P. (1988). A users’ guide to the

general health questionnaire. Windsor: NFER-Nelson.

Halpern, D. (1980). Mental health and the built environment.

London: Taylor and Francis.

Harris, T. (2001). Recent developments in understanding the

psychosocial aspects of depression. British Medical Bulletin.

Depression, 57, 17–32.

Holingshead, A. B., & Redlich, F. C. (1958). Social class and

mental illness. New York: Wiley.

Johnston, R. J. (1980). Multivariate statistical analysis in

geography. London: Longman.

Krosnik, J. A. (1999). Survey research. Annual Review of

Psychology, 50, 537–567.

Manchester City Council (2000). The 1998 local census: ward

comparison profile Manchester City Council Local Census

Planning Studies Manchester

ARTICLE IN PRESSR. Thomas et al. / Social Science & Medicine 60 (2005) 2773–2783 2783

Marmot, M., & Bobak, M. (2000). International comparators

and poverty and health in Europe. British Medical Journal,

321, 1124–1128.

Mcintyre, S., Hiscock, R., Kearns, A., & Ellaway, A. (2001).

Housing tenure and car access: further exploration of the

nature of their relations with health in a UK setting. Journal

of Epidemiology and Community Health, 55, 330–331.

Nettleton, S., & Burrows, S. (2000). When a capital investment

becomes an emotional loss; the health consequences of the

experience of mortgage possession in England. Housing

Studies, 15, 463–479.

MORI, (2001). Single regeneration budget challenge fund

evaluation [1996–2001] London, Market Opinion Research

International.

Nordenmark, M., & Strandh, M. (1999). Towards a socio-

logical understanding of mental well-being among the

unemployed: the role of economic and psychosocial factors.

Sociology of Health and Illness, 21, 577–598.

Olds, D. L., Henderson, C. R., Tatelbaum, R., & Chamberlin,

R. (1988). Improving the life-course development of socially

disadvantaged mothers: a randomised trial of nurse home

visitation. American Journal of Public Health, 78,

1436–1445.

ONS, (2001). Population Trends (winter). London: Office of

National Statistics.

Paykel, E. S. (1991). Depression in women. British Journal of

Psychiatry, 158(s10), 22–29.

Price, R. H., van Ryan, M., & Vinokur, A. (1992). Impact of a

preventative job search intervention on the likelihood of

depression among the unemployed. Journal of Health and

Social Behaviour, 33, 158–167.

Rogers, A., Gately, C., Evans, S., Huxley, P.J., Thomas, R.W.

(2005). More than jobs and houses: mental health, quality of

life and the perceptions of locality in an area undergoing

urban regeneration (forthcoming).

Rogers, A., Huxley, P. J., Thomas, R. W., Robson, B. T.,

Evans, S., Stordy, J., & Gately, C. (2001). Evaluating the

impact of a locality based social policy intervention on

mental health: conceptual and methodological issues.

International Journal of Social Psychiatry, 47, 41–55.

Stansfeld, S. A., Head, J., & Marmot, M. G. (1998). Explaining

social class differences in depression and well being. Social

Psychiatric Epidemiology, 33, 1–9.

Stansfeld, S. A., Fuhrer, R., Cattell, V., Wardle, J., & Head, J.

(1999). Psychosocial factors and the explanation of socio-

economic gradients in common mental disorder. Health

Variations, 4, 4–5.

Thomas, R. W., Evans, S., Gately, C., Stordy, J., Huxley, P.,

Rogers, A., & Robson, B. T. (2002). State-event relations

among indicators of susceptibility to mental distress in

Wythenshawe in the UK. Social Science and Medicine, 55,

921–935.

Thomson, H., Petticrew, M., & Morrison, D. (2001). Health

effects of housing improvement: systematic review of

intervention studies. British Medical Journal, 323, 187–190.

Weich, S., & Lewis, G. (1998). Material standard of living,

social class and the prevalence of the common mental

disorders in Great Britain. Journal of Epidemiology and

Community Health, 52, 8–14.

WHO, (2001). The world health report 20: mental health:

new understanding, new hope Geneva, World Health

Organisation.