Content uploaded by Kelvyn Jones
Author content
All content in this area was uploaded by Kelvyn Jones on Nov 06, 2017
Content may be subject to copyright.
Social Science & Medicine 60 (2005) 1267–1283
Social capital, geography and health: a small-area analysis
for England
John Mohan
a,
, Liz Twigg
a
, Steve Barnard
a
, Kelvyn Jones
b
a
Institute for the Geography of Health, University of Portsmouth, Buckingham Building, Lion Terrace, Portsmouth PO1 3HE, UK
b
School of Geographical Sciences, University of Bristol, University Road, Bristol BS8 1SS, UK
Available online 1 October 2004
Abstract
There has recently been much debate about the influence of social capital on health outcomes. In particular it has
been suggested that levels of social capital vary from place to place and that such variations may account for previously
unexplained between-place variations in health outcomes. As yet few studies exist of the influence of small-area
variations in social capital on health outcomes. One reason for this is the difficulty of obtaining indicators for small
areas such as electoral wards in England, and we describe a method used to derive what we term ‘synthetic estimates’ of
aspects of social capital by linking coefficients produced from multi-level analyses of national survey datasets to census
data. We produce estimates for electoral wards in England and apply these in multi-level models of our response
variable, the probability of survival of individuals surveyed in the Health and Lifestyle Survey of England. We report
various combinations of models incorporating individual attributes, health-related behaviours, area measures of
deprivation, and area measures of social capital. Our overall conclusion is that we find little support, at this spatial
scale, for the proposition that area measures of social capital exert a beneficial effect on health outcomes.
r2004 Elsevier Ltd. All rights reserved.
Keywords: Social capital; Place; Health inequalities; Health and Lifestyle Survey; Multi-level modeling; UK
Introduction
A growing body of research contends that area of
residence makes a difference to health-related behaviour
and health outcomes (Jones and Moon, 1993;MacIn-
tyre, MacIver & Sooman, 1993). Health outcomes thus
depend not only on individual characteristics (age,
gender, occupation, etc.) but also on the ‘ecology’, or
the surrounding environment in which individuals live
and work. The general conclusion which can be drawn
from studies of the effect of local social characteristics
on health outcomes appears to be that area of residence
does have an effect over and above effects of population
composition (Pickett & Pearl, 2001;MacIntyre, Ellaway
& Cummins, 2002). There is also evidence that area
effects vary between places, health outcomes and
population groups (Kawachi & Berkman, 2003;MacIn-
tyre & Ellaway, 2003;Robert, 1999).
In searching for explanations of these variations,
attention has been directed, through the work of
Wilkinson (1996, 2001), to the role of the social
environment in influencing life chances. Summarising
some complex debates, it has been argued that, once a
threshold level of development has been reached, it is the
more egalitarian societies which have the best health
records. Social inequality (especially income inequality)
has adverse effects both on individual self-esteem and on
community-level social cohesion. Many authors have
argued (see the review by Szreter & Woolcock, 2004)
ARTICLE IN PRESS
www.elsevier.com/locate/socscimed
0277-9536/$ - see front matter r2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.socscimed.2004.06.050
Corresponding author.
E-mail address: john.mohan@port.ac.uk (J. Mohan).
that social capital provides a missing causal link between
social inequality and health.
Social capital has a long history and many ante-
cedents. The central proposition is that through
participation in associational life of various kinds,
people become members of groups, which ‘both reflect
and help shape identity, norms, beliefs, and priorities’
(Macinko and Starfield, 2001, p. 388). Robert Putnam,
perhaps the key contemporary advocate of the concept,
defined it as ‘features of social organisation such as
networks, norms and trust’ (1993), and Macinko and
Starfield (2001, p. 388) suggest that social capital thus
refers to ‘available resources (capital) that can accrue to
people by virtue of their mutual acquaintance and
recognition (social) and that can be used for a variety of
productive activities’ (capital). A feature of contempor-
ary analysis is the emphasis on social capital as a
property of spatially-defined communities ranging from
villages, through regions (Putnam, 1993), to nation-
states (Fukuyama, 1995). It is suggested that in
communities possessed of high levels of social capital,
a number of beneficial outcomes may obtain.
If social capital is generated through various forms of
associational activity then there are good reasons to
anticipate geographical variations in it (see the review by
Mohan & Mohan, 2002). Political participation and
volunteerism vary, by age, class, ethnicity and gender
(Pattie, Seyd & Whiteley, 2004) and so one would expect
compositional effects to produce spatial variations.
Interestingly, recent studies (Coulthard, Walker &
Morgan, 2002;Williams, 2002) reveal regional varia-
tions in voluntary activity which are greater than one
would predict on the basis of compositional factors.
Other studies suggest a more local dimension to social
capital, arguing that the extent and character of civic
participation is very much shaped by ‘an appreciation of
local issues and problems’, because ‘most people’s lives
are conducted in the locality in which they reside’
(Parry, Moser & Day, 1992; see also Miller, Timpson &
Lessnoff, 1996;Verba, Schlozman & Brady, 1996).
The uneven development of the voluntary sector,
which is well-illustrated by the many statistical analyses
that have been carried out in the USA (e.g. Wolpert,
1990) and elsewhere (Kendall & Knapp, 1996;Salamon,
1995), may also influence social capital formation. The
presence or absence of supportive institutional struc-
tures can affect levels of participation and, thereby,
influence the formation of social capital (e.g. see Hall,
1999;Maloney, Smith & Stoker, 2000).
Finally, contemporary processes of uneven develop-
ment may have an impact on the quality of social
relationships and, therefore, on levels of social capital.
The flight of capital from certain locations has certainly
been associated with a decline in civility and increasing
levels of crime (Anderson, 1990;Campbell, 1993;
Wilson, 1987). This argument implies that there comes
a point at which normal social codes in neighbourhoods
may break down (Subramanian, Kim & Kawachi, 2002,
p. S32). Insofar as the affected neighbourhoods may be
relatively small areas, this raises the question of the
spatial scale at which social capital operates.
If levels of social capital vary geographically, at what
spatial scale ought it to be measured? Putnam does not
commit himself on this point although he recently called
for more work on subnational variations in social
capital (Putnam, 2002). A key problem in developing
measures of social capital for areas is that the concept
refers to community norms, which cannot easily be
measured. Sampson and Raudenbush (1999); see also
Sampson, Morenoff and Earls (1999) have pioneered
‘systematic social observation’ of behavioural norms
through covert observation of urban neighbourhoods,
but these could not easily be extended beyond a small
number of areas without vast resources. Researchers
have therefore had to resort to measures which
correspond to Krishna and Shrader’s (2000) distinction
between ‘structural’ and ‘cognitive’ components of
social capital. The former measures the quantity or
quality of associational links or activity, while the latter
refers to perceptions of support, reciprocity and trust.
Problems of measurement can be illustrated through a
consideration of previous studies of the relationship
between social capital, place and health.
Kawachi, Kennedy, Lochner and Prothrow-Stith
(1997) found statistically significant ecological associa-
tions between various aspects of social capital (trust,
perceived lack of fairness, perceived helpfulness of
others, and membership of groups) and mortality rates
for American states. Their analysis suggested that
income inequality acted through social capital to
influence mortality. Kawachi, Kennedy and Glass
(1999) argued, using a multilevel model, that people
living in states characterised by low levels of social
capital (indexed by measures of trust, reciprocity, and
civic engagement) tended to have higher probabilities of
lower self-reported health. Even after controlling for
individual—level variables (socio-economic characteris-
tics and health-related behaviours) residence in a low
social capital area was still associated with an excess risk
of reporting fair or poor health. Kawachi, Kennedy and
Wilkinson (1999) demonstrated a strong correlation
between income inequality, crime and social trust at the
state-level, suggesting a link between income inequality
and social cohesion. Investigating the Russian mortality
crisis, Kennedy, Kawachi, and Brainerd (1998) found
that their various indices of social capital and social
cohesion were strongly associated with age-adjusted
mortality and life expectancy for both men and women.
Walberg et al. (1998) also used crime as an index of
social cohesion in their regression analysis of regional
variations in the fall of life expectancy which occurred
after the collapse of communism in Russia. Reductions
ARTICLE IN PRESS
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–12831268
in life expectancy were most closely associated with
labour turnover, and were greatest in regions where
crime levels were highest and where incomes were most
unequal. Blakely, Kennedy and Kawachi (2001),ina
multi-level study, explored the relationship between
voting rates and self-rated health in the USA. There
was no direct association between income inequality and
variations in voter turnout (suggesting that the connec-
tion between inequality and this dimension of social
capital was not clear) but there was a suggestion that
individuals living in states with low voter turnout had
increased odds of fair or poor self-rated health.
Subramanian, Kawachi, and Kennedy (2001) investi-
gated the health effects on individuals (measured in
terms of the probability of self-reported poor health) of
state-level income, income inequality, and social capital.
As absolute income increased, the probability of
reporting poor health decreased. There were modest
effects for income inequality for high-income groups but
not for other income groups. Finally, the probability of
reporting poor health increased significantly as state-
level social capital declined. The authors thus contend
that this study demonstrated an ‘independent effect of
social capital’ (Subramanian et al., 2001, p. 16).
Some more recent studies have explored relationships
at a smaller geographical scale. Lochner, Kawachi,
Brennan, and Buka (2003) provided a cross-sectional
analysis of the relationship between social capital and
mortality for 342 neighbourhood clusters in Chicago.
Higher levels of neighbourhood social capital were
associated with reduced mortality for Whites, even after
adjustment for neighbourhood deprivation. For Blacks,
however, the associations were less consistent and often
not statistically significant.
Subramanian et al. (2001) found complex effects of
community-level social capital on the probability of
reporting self-rated poor health in 40 communities in the
USA. Higher levels of community social trust were
associated with a lower probability of reporting poor
health. Controlling for individual-level perceptions of
trust, however, rendered the main effect of community-
level social trust statistically insignificant. However,
there was a complex interaction effect: the health-
promoting effects of community-level social trust were
apparently greater for high-trust individuals. So, if
social capital does have beneficial effects, we cannot
assume that it is equally beneficial for all; it ‘may be
‘‘good medicine’’ only for for those who express high
levels of trust or who value trustworthiness in others’.
Furthermore, once allowance was made for the indivi-
dual compositional effects of socio-economic status,
‘communities do not make a difference to poor self-rated
health’ (Subramanian et al., 2001, p. S31).
The message of these studies is therefore somewhat
contradictory. Some strong claims have been made on
the basis of ecological analyses for large spatial units,
but whether these are units which exert a meaningful
influence on peoples lives is debatable. On the other
hand, at smaller spatial scales, there is a lack of
consistency in the results.
Our work was prompted by the relative absence of
similar work on the UK in general and at a small spatial
scale in particular. We were struck by an apparent
paradox. At the same time as influential studies of health
inequality were focussing on small areas, such as
electoral wards, work on social capital was being
conducted for very large spatial units. There was a need
for work which produced estimates of social capital for
small areas and then explored whether variations in
health outcomes were related to differential levels of
social capital.
We first describe how we devised small area estimates
of social capital, via a method which we characterise as
‘synthetic estimation’ of aspects of social capital. We
then describe a modelling exercise to explore the
relationship between individual and area characteristics
and health outcomes. The response variable used is the
probability that a respondent to the 1985 English Health
and Lifestyle Survey (HALS) was still alive in 1999. We
seek to explain this in terms of combinations of
individual socio-economic attributes, material circum-
stances of areas, and measures of social capital for small
areas. We find that our measures of social capital added
little to the explanatory power of our models and we
discuss some of the implications of these results.
Creating small-area indicators of social capital: direct
measurement and synthetic estimation
Many studies of social capital and health have
perforce had to measure their social capital indicators
for large spatial units. If we want to develop analyses
which are meaningful in relation to the contexts in which
people live their lives, however, it is clearly desirable to
produce measures of social capital for small areas. If we
focus on the structural component of social capital, then
there are several possibilities for constructing spatially—
disaggregated indicators of the proportion of the
population who engage in the kind of activities thought
to be conducive to the formation of social capital.
Some indicators of social capital (such as participa-
tion in associational life) are available from national
social surveys in England. However, at best the data are
only available at the regional level, and it is implausible
that there would be no subregional variations. We
therefore reviewed the availability and potential of
several measures either of engagement in the political
process or participation in altruistic or associational
activity. One possibility would be to explore the
distribution of members or supporters of organisa-
tions with a humanitarian (Amnesty), environmental
ARTICLE IN PRESS
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–1283 1269
(Greenpeace, Friends of the Earth) or social welfare
(Shelter) objective. Such organisations recruit and
campaign nationally (Jordan & Maloney, 1997) and
their lists of subscribers/donors are large (sometimes
several hundred thousand), permitting the construction
of stable ratios at a small-area level. Studies have
revealed interesting variations between places in the
distribution of members of such organisations (see, for
example, Cowell and Jehlicka (1995) on environmental
groups). However, if social capital is generated through
face-to-face interaction rather than the kind of ‘cheque-
book participation’ characteristic of such organisations,
it is questionable whether such measures would con-
stitute an appropriate index of social capital.
A second possibility was voter turnout, which has
been used in previous epidemiological studies; Whitley,
Gunnell, Dorling, and Davey-Smith (1999) used con-
stituency—level data on abstention rates in seeking to
account for spatial variations in suicide rates. However,
after adjusting for variations in an index of social
fragmentation, abstention rates were not significantly
associated with suicide rates. Indices of deprivation, in
this ecological study, were associated more strongly with
the pattern of suicide than were abstention rates.
We considered using ward-level data on participation
in local government elections. It is possible to calculate
turnout because the data consists of counts of voters and
of those on the electoral roll, but no standardisation for
age and sex composition is possible, and turnout is not
an accurate measure of engagement in the political
process because of non-registration (Pattie, Dorling,
Johnston & Rossiter, 1996, argue that non-registration
is closely correlated with turnout, however). Turnout in
local elections can also be highly variable, depending on
local circumstances (e.g. the presence of controversial
candidates), whether or not local elections are used by
voters to register what are in effect protest votes, and the
marginality or otherwise of seats. For all these reasons
we decided against using voter turnout to index civic
participation.
Finally, we considered the distribution of blood
donors, as this appeared a priori to pass most tests of
a small-area indicator of social capital, and some have
suggested that it might be used as a surrogate for social
capital (e.g. MacIntyre & Ellaway, 2003, p. 38). Titmuss
(1970, p. 13) regarded the arrangements for blood
donation and transfusion as a ‘social indicator which
‘within limits, is measurable and tells us something
about the quality of relationships and human values’.
Blood donation in the UK is usually an entirely
altruistic act, carried out by individuals without reward;
furthermore, subject to some constraints on health
grounds, the majority of the adult population can
donate blood and opportunities to donate are wide-
spread. However, for reasons explained elsewhere, we
are sceptical about the value of blood donation as an
index of social capital, though we did incorporate it in
our modelling exercise.
1
Given the limitations of such sources we adopted an
alternative approach which has been tried and tested
elsewhere (e.g. Twigg & Moon, 2002). We draw on
national surveys of England which ask questions about
activities which are believed to contribute to the
formation of social capital. These surveys are the
General Household Survey (GHS), the Survey of
English Housing (SEH) and the British Household
Panel Survey (BHPS). These datasets only permit a
disaggregation to the level of standard regions (N¼11);
while it would be possible to apply regional proportions
(e.g. the proportion of the middle class who volunteer)
from these surveys to local census data, this would make
the assumption that there were no within-region
variations. This is unlikely. However, we can circumvent
this problem by exploiting a property of the surveys,
namely that they rely on a clustered sampling design and
the surveys disclose whether or not individuals live in the
same primary sampling unit (PSU). We can therefore
take the PSU as an analogue for the local community
and, by aggregating the data for the individuals in each
PSU, it is possible to produce data for areas within
regions. This enables us to model within-region varia-
tion, in the probability of engaging in a given activity, as
a function not only of the individuals’ characteristics
and the region in which they live, but also as a function
of the characteristics of the local community, to which
the PSU provides an approximation.
The estimation process works as follows. First, multi-
level models are constructed which predict an aspect of
behaviour in terms of a combination of individual and
area circumstances and the interactions between them.
The resultant coefficients are then linked to the census
data. This requires a recognition that individual
probabilities are for particular types of individuals in
ARTICLE IN PRESS
1
These conclusions are based on an extensive multi-level
statistical analysis of the distribution of c. 1.8 Mn blood donors
in England whose age and sex were known, and who had given
blood in the 12 months prior to the date of supply of the data
by the National Blood Service to us in October 1999, and who
were aged between 25 and 64. We found strong associations
between blood donation and ecological measures of depriva-
tion. For all age groups, donation rates fell as the level of
deprivation in a neighbourhood rose, but the decline was clearly
steeper for older age groups. There were associations between
standardised donorship rates and demographic classifications
such as that developed by the Office for National Statistics
(ONS), but these were broadly consistent with comparative
levels of deprivation. There were also variations between the
regional management units of the National Blood Service.
Rather like voter turnout, therefore, blood donation rates are
shaped heavily by local contextual influences. The results of our
modelling exercise are discussed in Jones, Mohan, Barnard and
Twigg (2004).
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–12831270
areas with particular characteristics. With census data
on the number of individuals of each type in each area,
and the ecological characteristics of the area, we can
combine the multilevel regression estimates of these
individual, area and cross-level effects with census data,
to generate local predictions. These predictions are also
adjusted to take account of regional residuals that are
generated within the models. To clarify, we are not
applying coefficients to census data for the PSUs, but to
census data for electoral wards.
This process was repeated for a range of indicators
from various surveys (see Table 1) thereby generating
estimates of several possible surrogates for social capital.
These indicators were validated against direct measures
from other studies where appropriate and feasible and
the results of the validation give us confidence in our
methodology. The approach is described in more detail
elsewhere (Twigg & Moon, 2002;Mohan, Barnard,
Jones & Twigg, 2004, Chapter 4; both of these include a
flowchart giving details of the stages of the estimation
process). We now show how the synthetic estimates were
applied in practice in a modelling exercise to investigate
links between social capital, place and health.
Modelling the relationship between social capital, place
and health
Our aim was to explore the ecological influence of
social capital on individual health outcome. We under-
took this task by using the Health and Lifestyle Survey
(HALS) which is a comprehensive study of the health of
the adult national UK population (Cox, 1988). The
original sample from England, Wales and Scotland of
9003 respondents were initially interviewed in 1984/1985
and the original respondents were ‘flagged’ to provide
the subsequent date and cause of death (Cox et al.
(2001)). The present analysis is based on the English
sample of 7578 individuals followed to 2001; we do not
include 239 individuals who have been lost to follow-up.
We evaluated the importance of the synthetic estimates
of social capital in affecting whether a person lived or
died, taking account of individual demographic, lifestyle
and social circumstances, as well as ecological measures
of deprivation, and we also incorporated our standar-
dised blood donation rate.
The HALS survey investigated various individual
aspects of social capital as well as the socio-economic
circumstances of the respondents. We were supplied
with identification details of the respondent’s ward of
residence, which allowed us to attach the areal indicators
of social capital (as generated through the synthetic
estimation process) to individual survey data.
We report on a number of different model sequences
and discuss the conceptual justification for each in terms
of the causal pathway between social capital and
population health. In the first sequence of modelling
the approach is largely exploratory and the modelling is
undertaken as follows:
The first stage assesses the influence of individual
measures of social capital (i.e. derived from HALS)
on the risk of dying after controlling for a number of
individual confounders (i.e. age, sex, tenure, social
class and health-related behaviours);
Second, the effect of areal (i.e. ecological) indicators
of social capital on the risk of death is assessed (after
controlling for the individual confounders used in
Stage 1);
ARTICLE IN PRESS
Table 1
The social capital indicators for which multilevel synthetic estimates have been produced
Survey Question topic (indicator of social capital)
GHS Participating in any voluntary activity over last year (non-political or trade-union)
Participating in voluntary activity for 11 days or more over last year (non-political or trade-union) (‘core’ volunteering)
BHPS Active in political party, trades union or an environmental group—Political Activity
Active in 2 or more social activities (including parents’ association, tenants’ group, religious group, voluntary group,
other community group. social or sports club, women’s institute)—Social Activity
Active in 2 or more altruistic activities (including tenants’ group, religious group, voluntary group, other community
group and women’s institute)—Altruistic Activity
Thinks that local friends are important
Feels that they belong to the neighbourhood
Willing to work with others to improve the neighbourhood
Talks regularly to neighbours
Frequently meets people locally
Voted in the last general election
SEH Feels that the local area is friendly
Feels that the local area has ‘community sprit’
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–1283 1271
Third, the importance of areal measures of material
deprivation is contrasted against the relative influence
of the area social capital indicators.
Finally, cross-level interactions are explored to test,
for example, whether the effect of social capital is
different for different social classes, tenure groups or
age groups
This approach reports the overall impact of areal
social capital once all individual and household vari-
ables have been taken into account, and it reports how
the effect of social capital may be attenuated once areal
deprivation enters the equation. However, this largely
exploratory approach does not test for potential
mediating pathways between, for example, individual
health-related behaviours and area social capital as
suggested by Lindstrom, Merlo and Ostergren (2003).
We therefore present a second set of modelling results
where the impact of area social capital is investigated,
conditional only on individual age and sex; this can be
thought of as summarising the maximum possible effects
of social capital. We then introduce individual health-
related behaviours to investigate possible mediating
effects. Similarly, we report the changes in social capital
model coefficients between the ‘maximum effect’ model
and a model conditional on individual material circum-
stances (i.e. after the inclusion of class and tenure).
A multilevel logistic modelling approach was used to
analyse the simultaneous influence of both individual
and ward level characteristics on the risk of dying. The
7578 respondents were nested within 396 electoral wards
and 198 Parliamentary constituencies and several suites
of models using MLwiN (Goldstein, Rasbash, Plewis, &
et al., 1998) were developed to investigate the research
objectives listed above. Although the analysis focuses on
individual and ward level characteristics, the geographi-
cal level of constituency is retained in the models to
reflect the clustered sample design of HALS. During
sample selection, parliamentary constituencies were
stratified according to population size and the 198
constituencies were selected with probability propor-
tional to size of electorate. Two wards were then selected
from each constituency, again with probability propor-
tional to electorate size, resulting in a balanced multi-
level design at the highest levels.
In the logistic models, parameters were estimated
using second-order Taylor expansion with predictive
quasi-likelihood (PQL). This estimation procedure is
considered superior to first or second order marginal
quasi-likelihood (MQL) if the number of observations
within clusters is not always large (see Goldstein, 1995,
Chapter 7). While the quality of parameter estimates
may be more accurate still for small clusters using
Markov chain Monte Carlo (MCMC) methods (Gilks,
Richardson & Spiegelhalter, 1996), this method is
computationally intensive. We did however check our
estimates using MCMC methods but no major differ-
ences were revealed when compared to second order
PQL results.
The results of the first sequence of models are shown
in Table 2. The logit values are given with their
associated pvalues and in Model A, the logit value of
2.763 is given for the constant in an age- and sex-
adjusted base model. Here the constant refers to the
stereotypical respondent who is female and was aged 46
in 1984. By taking the antilogit of this value, it is
estimated that the probability of dying during the period
since the original HALS for this stereotypical respon-
dent is 0.059 (5.9%). In Model B, we see the effects of
adding in the controls of low social class (made up of
social classes III (manual), IV and V) and public rented
accommodation. The constant in the model now refers
to a female, aged 46, from either social class I or II, who
lives in either owner occupied housing or private rented
accommodation. For ease of interpretation the logits are
also expressed as odds ratios (OR) along with their 95%
confidence intervals.
2
In Model B we see that both
public renting and low social class significantly increase
the odds of dying. The effect for public renting is greater
than that of social class. The risk of death for the
stereotypical respondent (defined above) is estimated at
around 4.0%; however, if this stereotypical individual
happens to be low social class then the risk is
approximately 5.9%, and if she lives in public sector
accommodation it increases to 9.9%.
Model C includes health-related behaviours, regular
tobacco smoking, unsafe alcohol consumption, un-
healthy diet and lack of exercise reported at the
individual level (definitions of these indicators are given
in the original Health and Lifestyle Survey: Cox et al.,
1988). The results of Model C are all significant apart
from unsafe alcohol consumption but it is interesting to
note that a significant and positive increase in risk is
given for the group whose alcohol consumption details
are unknown. The constant value indicates that the risk
of dying is 2.3% for a female respondent of average age,
who is from social class I, II or III non-manual, is an
owner-occupier, does not smoke, consumes safe
amounts of alcohol, eats a healthy diet and does not
smoke. The risk increases to 2.8% for a similar
individual but whose alcohol consumption status is
unknown. Of all the health related behaviours, smoking
has the greatest impact; the odds of dying are increased
ARTICLE IN PRESS
2
The odds ratio is a way of comparing whether the
probability of a certain event is the same for two groups. An
odds ratio of 1 implies that the event is equally likely between
the groups. An odds ratio greater than one implies that the
event is more likely (in comparison to the base group) and an
odds ratio less than one implies that the event is less likely. An
odds ratio of 2.0, for example, increases the risk of dying two-
fold.
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–12831272
ARTICLE IN PRESS
Table 2
Modelling the effects of social capital on death using HALS survey data
Model Logit (p) Odds Ratio (OR) (95%
Confidence Intervals (CI))
A(Base model:.age and sex)
Constant=2.763 (risk=5.93%)
B(A plus social class and tenure)
Constant=3.164 (risk=4.05%)
Social class IIIm,IV,V 0.389 (o0.000) 1.46 (1.22–1.79)
Public renting 0.951 (o0.000) 2.59 (1.86–3.60)
C. (B plus HRBs)
Constant=3.742 (risk=2.32%)
Social class 0.275 (0.006) 1.32 (1.08–1.60)
Public renting 0.805 (o0.000) 2.24 (1.60–3.13)
Smoking 0.550 (o0.000) 1.73 (1.45–2.07)
Alcohol 0.077 (0.557) 1.08 (0.83–1.40)
Unknown alcohol 0.191 (0.032) 1.21 (1.02–1.44)
Diet 0.238 (0.005) 1.27 (1.07–1.50)
Exercise 0.472 (o0.000) 1.60 (1.33–1.93)
D. (C plus individual ‘Community’) 0.101 (0.257) 1.11 (0.93–1.32)
E. (C plus individual ‘Reliable friends) 0.053 (0.841) 1.05 (0.63–1.78)
F. (C plus individual ‘Loneliness’) 0.260 (0.072) 1.30 (0.98–1.72)
G(C plus areal deprivation)
(within the 40% most deprived wards) 0.210 (0.020) 1.23 (1.04–1.46)
H(C plus voluntary activity)
ð24:30 30:20Þ
ð20:10 24:20Þ
ðo20Þ
% engaged in voluntary
activity
9
>
=
>
;
0.202 (0.087) 1.22 (0.97–1.26)
0.239 (0.044) 1.27 (1.01–1.60)
0.300 (0.014) 1.35 (1.06–1.71)
I(C plus core volunteering)
ð15:04 18:42Þ
ð11:62 15:03Þ
ðo11:61Þ
% who are ‘core’
volunteers
9
>
=
>
;
0.176 (0.138) 1.19 (0.95–1.50)
0.312 (0.009) 1.37 (1.08–1.73)
0.268 (0.030) 1.31 (1.03–1.67)
J(C plus social activity)
ð7:75 10:18Þ
ð6:01 7:74Þ
ðo6:0Þ
% engaged in social
activity
9
>
=
>
;
0.196 (0.098) 1.22 (0.97–1.53)
0.312 (0.009) 1.37 (1.08–1.72)
0.307 (0.011) 1.36 (1.07–1.73)
K(C plus altruistic activity)
ð3:08 4:50Þ
ð2:44 3:07Þ
ðo2:43Þ
% engaged in altruistic
activity
9
>
=
>
;
0.150 (0.202) 1.16 (0.92–1.46)
0.330 (0.006) 1.39 (1.10–1.76)
0.242 (0.048) 1.27 (1.00–1.57)
L(C plus political activity)
ð5:33 6:20Þ
ð4:61 5:32Þ
ðo4:60Þ
% engaged in political
activity
9
>
=
>
;
0.060 (0.608) 1.06 (0.84–1.34)
0.210 (0.075) 1.23 (0.98–1.55)
0.242 (0.040) 1.27 (1.01–1.60)
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–1283 1273
to 1.73 for a regular smoker. The odds for individuals
classed as inactive are slightly less at 1.60 and for those
consuming an unhealthy diet they are given as 1.27. Low
social class is similar in impact to an unhealthy diet but,
as in Model B, the greatest increase in risk is associated
with living in public rented accommodation. Here the
odds are given as 2.24, representing a risk of 5.0% for
someone living in such accommodation (but otherwise
having the characteristics of the base-category indivi-
dual) compared to the average respondent.
ARTICLE IN PRESS
Table 2 (continued )
Model Logit (p) Odds Ratio (OR) (95%
Confidence Intervals (CI))
M(C plus voted in last election)
ð79:12 80:68Þ
ð77:01 79:11Þ
ðo77:00Þ
% who voted in last
election
9
>
=
>
;
0.083 (0.460) 1.09 (0.87–1.35)
0.112 (0.330) 1.12 (0.89–1.40
0.026 (0.827) 1.03 (0.81–1.29)
N(C plus local friends important)
ð62:73 65:36Þ
ð59:86 62:72Þ
ðo59:85Þ
% who think local friends
are important
9
>
=
>
;
0.75 (0.496) 0.93 (0.75–1.15)
0.097 (0.386) 1.10 (0.88–1.37)
0.186 (0.108) 1.20 (0.96–1.51)
O(C plus belong to neigh’hood)
ð68:43 70:74Þ
ð65:53 68:42Þ
ðo65:52Þ
% who belong to the
neighbourhood
9
>
=
>
;
0.085 (0.454) 1.09 (0.87–1.36)
0.079 (0.494) 1.08 (0.86–1.36)
0.071 (0.557) 0.93 (0.73–1.18)
P(C plus work to improve n’hood)
ð76:36 79:32Þ
ð74:20 76:35Þ
ðo74:19Þ
% who would work to
improve the
neighbourhood
9
>
=
>
;
0.097 (0.402) 1.10 (0.88–1.38)
0.226 (0.051) 1.25 (1.00–1.57)
0.084 (0.486) 1.09 (0.86–1.38)
Q(C plus talks to neighbours)
ð70:38 72:65Þ
ð67:98 70:37Þ
oð67:97Þ
% who talk to neigbours
9
>
=
>
;
0.075 (0.504) 1.08 (0.87–1.34)
0.159 (0.163) 1.17 (0.94–1.47)
0.041 (0.720) 1.04 (0.83–1.30)
R(Model C plus freq’ly meets locals)
ð63:87 69:58Þ
ð59:97 63:86Þ
ðo59:96Þ
% who frequently meet
local people
9
>
=
>
;
0.028(0.808) 0.97 (0.78–1.22)
0.118(0.136) 0.89 (0.71–1.12)
0.226(0.068) 0.80 (0.63–1.02)
S(Model C feels local area friendly)
ð93:69 94:77Þ
ð92:23 93:68Þ
ðo92:22Þ
% who feel the local area
is friendly
9
>
=
>
;
0.125 (0.250) 1.13 (0.91–1.40)
0.106 (0.350) 0.90 (0.72–1.12)
0.169 (0.146) 0.84 (0.67–1.06)
T(Model C plus blood donation)
ð99:58 113:95Þ
ð77:25 96:57Þ
ðo77:24Þ
Standardised blood
donorship ratio
9
>
=
>
;
0.097 (0.408) 0.91 (0.72–1.14)
0.004 (0.975) 0.99 (0.80–1.25)
0.049 (0.677) 1.05 (0.83–1.32)
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–12831274
Before moving on to explore the impact of ecological
measures of deprivation and our social capital indica-
tors, we investigated the impact of individual aspects of
social capital as measured by HALS. Respondents are
asked whether they feel ‘part of their community’,
whether they have ‘people that they can rely upon no
matter what happens’, and how often they ‘feel lonely’.
Each of these variables was included in a base model
that is also adjusted for class, tenure and health-related
behaviours (Models D–F). None of the three individual-
based social capital questions from the HALS achieve
conventional levels of statistical significance but the
variable related to loneliness has an associated pvalue of
0.072. Here the relative odds of dying are increased to
1.30 for those who state that they are ‘often’ or ‘always’
lonely, compared to those who are ‘never’ or only
‘sometimes’ lonely. It could be argued that this variable
is really a surrogate for an individual’s access to social
support rather than stocks of social capital, and this
reflects some of the terminological confusion surround-
ing the concept of social capital.
Moving on to investigate the influence of ecological
indicators of deprivation and estimates of social capital,
each areal measure was added separately to a base
model containing age, gender, tenure, class and health-
related behaviours. Whilst Model B investigated the
impact of individual material circumstances via social
class and housing tenure, Model G assesses whether
there is evidence for a separate, independent effect for
material deprivation measured at the ward level, using
the Carstairs deprivation score (Carstairs & Morris,
1991). In the model, a dichotomy is used which
categorises wards according to whether or not they are
among the most-deprived 40% wards in the country.
Background exploratory work showed that this dichot-
omy was efficient in capturing the influence of ecological
measures of deprivation. Again we note that residing in
one of these deprived wards significantly increases the
odds of dying. The risk of dying for the base category
individual (an average aged woman, who owns her own
house, from social class I, II or III non-manual who lives
a healthy lifestyle) who resides in a non-deprived ward is
approximately 2.2%. However if the ward is amongst
the 40% most deprived then the risk is increased to
2.7%, representing an odds ratio of 1.23 in contrast to
the base category.
The results of adding in all other social capital
estimates are shown in Models H–T. All synthetic
estimates of social capital listed in Table 1 have been
included apart from ‘community spirit’; this indicator
was excluded because the initial multilevel model of this
variable used in the synthetic estimation process did not
have sufficient explanatory power to generate worth-
while predictions. The estimates of social capital have
been included in the models in the form of ordinal data
based on approximate quartile cut-off points. Quartiles
were used, rather than the actual scores, because (a) they
pick up any non-linear relationships between social
capital and mortality, (b) a classification system based
on equal counts provided us with equal reliability across
the resultant standard errors; and (c) the indicators are
synthetic estimates rather than direct measures, we did
not want to place too much emphasis on the actual score
of individual wards, and so we explored the effects via
these relatively broad groupings of social capital
measures. Details of these quartile-based class intervals
for each social capital indicator are given in the Table.
Results for the lowest three quartiles are shown in Table
2(and also in Tables 3 and 4) and are contrasted against
the stereotypical individual living in an area with the
highest estimated levels of social capital (Thus, for
voluntary activity, the contrast is against wards in which
the estimate of the proportion of the population
involved in voluntary activity is at least 30.2%).
Also, in respect of blood donation, we did consider
whether or not we should remove age and sex from the
model incorporating this term, as the blood donation
rate is already standardised for age and sex. However
there was no significant difference between models
incorporating age, sex and blood donation, and those
simply incorporating blood donation, and so we have
simply reported the former (Table 2, model T).
For both volunteering and ‘core’ volunteering, there
appears to be a greater risk of dying associated with
lower levels of activity (Models H and I). All categories
are significant apart from the third quartiles in both sets
of voluntary models and for core volunteering the
gradient across the quartiles does not appear linear.
Here the largest odds are given for the second quartile
(1.37). A similar relationship holds for social, altruistic,
and political activities, whereby the probability of dying
is increased as these activities decrease at an ecological
level. Again the third quartiles are not significantly
different from the base categories, which represent the
highest levels of areal participation (Models J–L).
Models M–O indicate that there is very little significant
association between risk of death and the levels of voter
turnout for the last general election; the numbers who
think that local friends are important; and feeling part of
the local neighbourhood. None of the terms are
significant for voting in the last general election and
neither does there appear to be a consistent trend in the
odds ratios across the quartiles (Model M). Although
not significant at conventional levels (p¼0:108), there
may be an effect for living in an area that has lowest
levels of people who consider local friends to be
important (Model N). Here the odds of dying are
increased to 1.20 (confidence interval=0.961.51).
Whilst we acknowledge that most of the terms are
insignificant in Models N and O, it is interesting that the
overall trend in the odds ratios for these two varia-
bles are in opposite directions. For example, as the
ARTICLE IN PRESS
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–1283 1275
proportion considering local friends to be important
decreases, then the probability of death increases.
However for feeling part of the neighbourhood, the
opposite is true and the odds decrease as the ecological
percentage of this variable also decreases. Again, the
relationship between ‘willing to work with others to
improve the neighbourhood’ (Model P) and ‘talking
regularly to neighbours’ (Model Q) is largely insignif-
icant and in both models the relationship is not
consistent across the quartiles. In contrast there appears
to be a consistent relationship across the quartiles for
ecological percentages of ‘frequently meeting people
locally’ but the direction does not give support to the
positive impact of social capital on health; the trend
indicates that the risk of dying decreases as the
percentages of this variable also decreases (Model R).
It should be noted here however, that it is only the
lowest quartile that is statistically significant at conven-
tional levels. None of the terms are significant in Model
S (which focuses on the overall friendliness of an area)
and again the direction of the relationship is not
consistent across the gradient of the quartiles. Finally,
standardised blood donation rates are investigated in
Model T (with age and sex in the model) and the results
suggest that there is no substantive relationship between
altruistic activity (as captured through blood donation)
and individual health outcome as surveyed via mortality
in the HALS.
In Models H–T described above (i.e. the areal social
capital models), deprivation has been excluded as a
term. When deprivation is included as a dummy in the
modelling procedure (i.e. whether a ward is amongst the
40% most deprived wards or not, according to Carstairs
score), the effect is for the previous significant findings
for the relationships between the probability of dying
and the measures of volunteering and social and
altruistic activities to become insignificant. Carstairs,
as a term, is also rendered insignificant. However, in
such models, individual social class and tenure maintain
their statistical significance. When deprivation is in-
cluded in the political activity model, deprivation
remains significant alongside class and tenure but the
relationship between political activity and health does
not maintain its significance. When deprivation is
introduced into the remaining models (i.e. Models
M–T), all social capital terms remain non-significant,
whilst Carstairs is significant for ‘voting in the last
general election’, ‘belonging to neighbourhood’, ‘work-
ing with others to improve neighbourhood’, ‘talking
regularly to neighbours’, ‘feels that the local area is
friendly’ and the standardised blood donation rate
(Model T—with age and sex in the model). It is
important to note, that during model construction for
the generation of synthetic estimates, some of the
ecological variables included in the models were similar
to the individual components of the Carstairs index (i.e.
unemployed males, overcrowding, non-car ownership
and low social class). In effect this means that these
variables are being entered twice, albeit with different
weight, into the equation and this may result in a
lessening of the overall effect for the Carstairs index of
deprivation. This suggests that, where it is significant,
there is a real and independent effect for ‘place’
deprivation. The final stage of modelling was to test
for possible cross-level interactions between the effect of
social capital and different class, tenure and age groups.
None of these interactions were statistically significant
(data not shown; details available from authors).
These results do not present a very strong case for the
influence of areal social capital on health when
individual material circumstances, health-related beha-
viour and area deprivation are all taken into account.
However it is possible that both individual material
circumstances and/or health-related behaviours are the
mediating pathway between social capital and health.
We can investigate this supposition by removing these
variables from the models and reporting the social capital
effects on a base model that contains age and sex only
(Table 3) and comparing these results with elements from
Table 2. We can also compare the results with a sequence
of models that contain age, sex, health-related behaviours
and area social capital (Table 4) and models that contain
age, sex, class, tenure and area social capital.
Table 3, in essence, represents the maximum social
capital effects due to the fact that the models are only
age-sex standardised. There is a clear, statistically
significant gradient for volunteering, social activity,
altruistic activity, political activity and voting behaviour
with lower levels of activity increasing the risk of
mortality in each case (Models B–G). Whilst the
gradient is also present for Models H, and N (the
importance of local friends and blood donation), the
coefficients are not always significant. If the social
capital and health link were mediated by health related
behaviours then we would expect the gradients exhibited
in Table 3 to diminish or disappear once individual
health related behaviours are included in the models.
The results shown in Table 4 indicate that although the
gradients remain (and the coefficients are statistically
significant) for volunteering, social and political activity,
they are less strong than in Table 3. Gradients also
remain for altruistic, political, and voting activity,
‘thinking that local friends are important’, ‘belonging
to the neighbourhood’ (Models E–I) and blood dona-
tion (model N) but terms are not always statistically
significant. Table 4 also suggests that the effect of the
inclusion of area social capital measures on the health-
related behaviour coefficients is minimal. In the base
model (Model A) summarising the effects of health-
related behaviours only (after controlling for age and
sex), smoking, unknown alcohol consumption, diet and
exercise are all significant and remain so in all of the
ARTICLE IN PRESS
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–12831276
ARTICLE IN PRESS
Table 3
Social capital and health: maximum effects
Model Logit (p) OR (95% CI)
A(Base model:.age and sex)
B(base plus voluntary activity)
ð24:30 30:20Þ
ð20:10 24:20Þ
ðo20Þ
% involved in voluntary
activity
9
>
=
>
;
0.304 (0.009) 1.36 (1.08–1.70)
0.454 (o0.000) 1.58 (1.25–1.98)
0.722 (o0.000) 2.06 (1.64–2.59)
C(base plus core volunteering)
ð15:04 18:42Þ
ð11:62 15:03Þ
ðo11:61Þ
% who are ‘core’
volunteers
9
>
=
>
;
0.301 (0.010) 1.35 (1.08–1.70)
0.513 (o0.000) 1.67 (1.33–2.10)
0.728 (o0.000) 2.07 (1.65–2.60)
D(base plus social activity)
ð7:75 10:18Þ
ð6:01 7:74Þ
ðo6:0Þ
% engaged in social
activity
9
>
=
>
;
0.283 (0.015) 1.33 (1.06–1.67)
0.506 (o0.000) 1.66 (1.32–2.09)
0.712 (o0.000) 2.04 (1.62–2.56)
E(base plus altruistic activity)
ð3:08 4:50Þ
ð2:44 3:07Þ
ðo2:43Þ
% engaged in altruistic
activity
9
>
=
>
;
0.257 (0.028) 1.29 (1.03–1.74)
0.572 (o0.000) 1.77 (1.41–2.23)
0.639 (o0.000) 1.89 (1.50–2.39)
F(base plus political
activity)
ð5:33 6:20Þ
ð4:61 5:32Þ
oð4:60Þ
% engaged in political
activity
9
>
=
>
;
0.131 (0.283) 1.14 (0.90–1.45)
0.309 (0.011) 1.36 (1.07–1.73)
0.377 (0.002) 1.46 (1.15–1.86)
G(base plus voted in last election)
ð79:12 80:68Þ
ð77:01 79:11Þ
ðo77:00Þ
% who voted in last
election
9
>
=
>
;
0.126 (0.283) 1.13 (0.90–1.43)
0.258 (0.031) 1.29 (1.02–1.64)
0.256 (0.039) 1.29 (1.01–1.65)
H(base plus local friends important)
ð62:73 65:36Þ
ð59:86 62:72Þ
ðo59:85Þ
% who think local friends
are important
9
>
=
>
;
0.064 (0.579) 0.94 (0.75–1.18)
0.090 (0.449) 1.09 (0.87–1.38)
0.313 (0.012) 1.37 (1.07–1.75)
I(base plus belong to neigh’hood)
ð68:43 70:74Þ
ð65:53 68:42Þ
ðo65:52Þ
% who feel they belong to
the neighbourhood
9
>
=
>
;
0.249 (0.033) 1.28 (1.02–1.61)
0.277 (0.020) 1.32 (1.04–1.66)
0.243 (0.053) 1.24 (1.00–1.63)
J(base plus work to improve n’hood)
ð76:36 79:32Þ
ð74:20 76:35Þ
ðo74:19Þ
% who would work to
improve the
neighbourhood
9
>
=
>
;
0.185 (0.117) 1.20 (0.96–1.52)
0.446 (o0.000) 1.56 (1.24–1.97)
0.398 (0.001) 1.49 (1.17–1.89)
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–1283 1277
subsequent models. Furthermore, their coefficients are
similar across all of the social capital models (Models
B–N). All of this suggests that there may be some
evidence for health-related behaviour having a mediat-
ing effect on the relationship between area social capital
and health as the impact of social capital is attenuated
slightly once health-related behaviours are included in
the models. The relationship between health-related
behaviour and health remains, irrespective of social
capital.
A similar exercise was undertaken using individual
material circumstances (class and tenure) and the
significant gradients shown in the ‘maximum effects’
model (Table 3) remain only for the two measures of
volunteering once class and tenure are entered into the
model (results not shown). Again the gradients for these
two aspects of voluntary activity were reduced. Whilst
gradients (in the expected direction) remain for social
activity, altruistic activity, political activity and the
‘importance of local friends’, one or more of the terms in
the gradient are insignificant. Similarly an inverse
gradient remains for ‘frequently meeting locals’ and
‘feels that the local area is friendly’ but none of the terms
are significant. The impacts of class and tenure remain
fairly constant across all models. Again, this suggests
that the relationship between area social capital and
health may be mediated by individual material circum-
stances.
Conclusions
We have modelled individual and ecological data
simultaneously to account for variations in individual
mortality in a follow-up study, and in this respect our
work contrasts with earlier cross-sectional and aggregate
studies. We have been able to assess the effect of
individual and ecological measures of social capital in
models which also contain demographic, health-related
behaviour and social-structural variables at the indivi-
dual level. Our work is also novel in that we have
produced estimates of levels of social capital for small
areas (electoral wards).
The modelling exercise has indicated that our validated
estimates of social capital do not account substantially for
variations in mortality, and even in those instances where a
modest effect is found, the impact is attenuated when a
measure of material deprivation is included in the model.
The effect of social capital was also lessened when health-
related behaviours or material circumstances were included
in age–sex adjusted models, suggesting that these factors
may be part of a possible mediating pathway between
social capital and health. It is, however, possible that
mortality is simply not sensitive enough, as an indicator of
individual health, to detect real and valid impacts of social
capital.
The largely negative results could be because the
models are based on estimates, not on observed values,
ARTICLE IN PRESS
Table 3 (continued )
Model Logit (p) OR (95% CI)
K(base plus talks to neighbours)
ð70:38 72:65Þ
ð67:98 70:37Þ
oð67:97Þ
% who talk to neighbours
9
>
=
>
;
0.029 (0.813) 0.97 (0.77–1.23)
0.085 (0.486) 1.09 (0.86–1.39)
0.007 (0.950) 1.01 (0.79–1.29)
L(base plus freq’ly meets locals)
ð63:87 69:58Þ
ð59:97 63:86Þ
ðo59:96Þ
% who frequently meet
locals
9
>
=
>
;
0.281 (0.013) 0.76 (0.61–0.94)
0.455 (o0.000) 0.63 (0.51–0.79)
0.648 (o0.000) 0.52 (0.41–0.66)
M(base plus feels local area friendly)
ð93:69 94:77Þ
ð92:23 93:68Þ
ðo92:22Þ
% who feel the local area
is friendly
9
>
=
>
;
0.135 (0.253) 1.14 (0.91–1.44)
0.155 (0.209) 0.86 (0.67–1.09)
0.095 (0.452) 1.91 (0.71–1.17)
N(base plus blood donation)
ð99:58 113:95Þ
ð77:25 96:57Þ
ðo77:24Þ
Standardised blood
donorship ratio
9
>
=
>
;
0.062 (0.612) 0.94 (0.74–1.19)
0.059 (0.808) 1.06 (0.84–1.34)
0.257 (0.612) 1.29 (1.02–1.65)
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–12831278
ARTICLE IN PRESS
Table 4
Modelling the effects of social capital and health-related behaviours on mortality
Model Logit (p) OR (95%CI)
A(Base model: age, sex and HRBs)
Smoking 0.653 (o0.000) 1.92 (1.61–2.29)
Alcohol 0.079 (0.550) 1.08 (0.84–1.40)
Unknown alcohol 0.254 (0.004) 1.29 (1.08–1.53)
Diet 0.296 (o0.000) 1.34 (1.14–1.59)
Exercise 0.504 (o0.000) 1.66 (1.37–2.00)
B(A plus Voluntary activity)
Smoking 0.612 (o0.000) 1.84 (1.55–2.20)
Alcohol 0.062 (0.635) 1.06 (0.82–1.38)
Unknown alcohol 0.231 (0.009) 1.26 (1.06–1.50)
Diet 0.256 (0.003) 1.29(1.09–1.53)
Exercise 0.487 (o0.000) 1.63 (1.35–1.96)
(24.30 – 30.20) 0.241 (0.042) 1.27 (1.01–1.60)
(20.10 – 24.20) 0.345 (0.004) 1.41(1.12–1.78)
(o20) 0.518 (o0.000) 1.67 (1.33–2.11)
C(A plus core volunteering)
Smoking 0.609 (o0.000) 1.84 (1.54–2.19)
Alcohol 0.066 (0.613) 1.07 (0.83–1.38)
Unknown alcohol 0.229 (0.009) 1.26 (1.06–1.50)
Diet 0.263 (0.002) 1.30 (1.10–1.54)
Exercise 0.483 (o0.000) 1.62 (1.35–1.95)
(15.04 – 18.42) 0.231 (0.051) 1.26 (1.00–1.59)
(11.62 – 15.03) 0.413 (o0.000) 1.51 (1.20–1.91)
(o11.61) 0.520 (o0.000) 1.68 (1.34–2.12)
D(A plus social activity)
Smoking 0.611 (o0.000) 1.84 (1.55–2.20)
Alcohol 0.077 (0.557) 1.08 (0.84–1.40)
Unknown alcohol 0.235 (0.008) 1.26 (1.06–1.50)
Diet 0.266 (0.002) 1.30 (1.11–1.54)
Exercise 0.486 (o0.000) 1.63 (1.35–1.96)
ð7:75 10:18Þ
ð6:01 7:74Þ
ðo6:0Þ
% engaged in social
activity
9
>
=
>
;
0.247 (0.037) 1.28 (1.02–1.61)
0.436 (o0.000) 1.55 (1.23–1.95)
0.517 (o0.000) 1.68 (1.33–2.11)
E(A plus altruistic activity)
Smoking 0.617 (o0.000) 1.85 (1.55–2.21)
Alcohol 0.073 (0.576) 1.08 (0.83–1.39)
Unknown alcohol 0.227 (0.010) 1.26 (1.05–1.49)
Diet 0.268 (0.002) 1.31 (1.11–1.54)
Exercise 0.441 (o0.000) 1.55 (1.23–1.96)
ð3:08 4:50Þ
ð2:44 3:07Þ
ðo2:43Þ
% engaged in altruistic
activity
9
>
=
>
;
0.493 (o0.000) 1.64 (1.36–1.97)
0.200 (0.091) 1.22 (0.97–1.54)
0.472 (o0.000) 1.60 (1.27–2.02)
F(A plus political activity)
Smoking 0.644 (o0.000) 1.90 (1.60–2.27)
Alcohol 0.076 (0.564) 1.08 (0.83–1.40)
Unknown alcohol 0.243 (0.006) 1.28 (1.07–1.52)
Diet 0.283 (o0.000) 1.33 (1.12–1.57)
Exercise 0.500 (o0.000) 1.65 (1.37–1.99)
ð5:33 6:20Þ
ð4:61 5:32Þ
ðo4:60Þ
% engaged in political
activity
9
>
=
>
;
0.081 (0.505) 1.08 (0.86–1.37)
0.234 (0.053) 1.26 (1.00–1.60)
0.255 (0.035) 1.29 (1.02–1.64)
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–1283 1279
ARTICLE IN PRESS
Table 4 (continued )
Model Logit (p) OR (95%CI)
G(A plus voted in last election)
Smoking 0.645 (o0.000) 1.91 (1.60–2.27)
Alcohol 0.079 (0.546) 1.08 (0.84–1.40)
Unknown alcohol 0.245 (0.006) 1.28 (1.07–1.52)
Diet 0.298 (o0.000) 1.35 (1.14–1.59)
Exercise 0.504 (o0.000) 1.66 (1.37–1.99)
ð79:12 80:68Þ
ð77:01 79:11Þ
ðo77:00Þ
% who voted in last
election
9
>
=
>
;
0.117 (0.309) 1.12 (0.90–1.41)
0.210 (0.074) 1.23 (0.98–1.55)
0.161 (0.183) 1.74 (0.93–1.49)
H(A plus local friends important)
Smoking 0.642 (o0.000) 1.90 (1.59–2.26)
Alcohol 0.074 (0.576) 1.08 (0.83–1.39)
Unknown alcohol 0.240 (0.007) 1.27 (1.07–1.51)
Diet 0.299 (o0.000) 1.35 (1.14–1.59)
Exercise 0.501 (o0.000) 1.65 (1.37–1.99)
ð62:73 65:36Þ
ð59:86 62:72Þ
ðo59:85Þ
% who think local friends
are important
9
>
=
>
;
0.061 (0.591) 0.94 (0.75–1.18)
0.083 (0.476) 1.09 (0.86–1.37)
0.237 (0.050) 1.27 (1.00–1.61)
I(A plus belong to neigh’hood)
Smoking 0.647 (o0.000) 1.91 (1.60–2.28)
Alcohol 0.074 (0.573) 1.08 (0.83–1.39)
Unknown alcohol 0.249 (0.005) 1.28 (1.08–1.53)
Diet 0.293 (o0.000) 1.34 (1.14–1.58)
Exercise 0.503 (o0.000) 1.65 (1.37–1.99)
ð68:43 70:74Þ
ð65:53 68:42Þ
ðo65:52Þ
% who feel they ‘belong
to the neighbourhood’
9
>
=
>
;
0.186 (0.106) 1.21 (0.96–1.51)
0.212 (0.071) 1.24 (0.98–1.55)
0.117 (0.339) 1.12 (0.88–1.43)
J(A plus work to improve n’hood)
Smoking 0.633 (o0.000) 1.88 (1.58–2.24)
Alcohol 0.076 (0.564) 1.08 (0.83–1.40)
Unknown alcohol 0.243 (0.006) 1.27 (1.07–1.52)
Diet 0.288 (o0.000) 1.33 (1.13–1.57)
Exercise 0.497 (o0.000) 1.64 (1.36–1.98)
ð76:36 79:32Þ
ð74:20 76:35Þ
ðo74:19Þ
% who would work to
improve the
neighbourhood
9
>
=
>
;
0.161 (0.170) 1.17 (0.93–1.48)
0.352 (0.003) 1.42 (1.13–1.79)
0.244 (0.045) 1.26 (1.01–1.62)
K(base plus talks to neighbours)
Smoking 0.655 (o0.000) 1.93 (1.62–2.29)
Alcohol 0.077 (0.560) 1.08 (0.83–1.40)
Unknown alcohol 0.253 (0.004) 1.29 (1.08–1.53)
Diet 0.296 (o0.000) 1.34 (1.14–1.59)
Exercise 0.504 (o0.000) 1.66 (1.37–2.00)
ð70:38 72:65Þ
ð67:98 70:37Þ
oð67:97Þ
% who talk to neighbours
9
>
=
>
;
0.004 (0.975) 1.00 (0.80–1.26)
0.114 (0.335) 1.12 (0.89–1.41)
0.007 (0.950) 1.01 (0.80–1.27)
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–12831280
but we have validated our data against real-world
observations and therefore have confidence in them, at
least as indicators of relative levels of social capital (see
Mohan et al., 2004, Chapter 5). Bearing that in mind, the
modelling exercises have indicated that there is no real
evidence for a strong consistent relationship between
health outcome and ecological measures of social capital.
There is a great deal of collinearity between deprivation
and the measures of social capital and it is impossible to
unpack their relative effects in a combined model.
Furthermore, the direction of the relationship between
the social capital indicators and health is not always
consistent; in other words, high levels of social capital do
not always generate positive health advantages. Through-
out all the models, the effects of individual social class and
tenure remain statistically significant with tenure having
the greatest impact on health outcome. This casts doubt on
whether social capital has an ecological influence on health
outcomes over and above that of material conditions, at
least at this spatial scale.
There could be other reasons for these inconclusive
results. One possibility, as Blaxter and Poland (2003)
suggest, is that the indicators are estimates of the
propensity for individuals to engage in particular
behaviours, and are not, therefore, true contextual
constructs. Such a charge could be levelled against
almost any quantitative index derived in this way, but
the alternative (direct observation of some kind) is
prohibitively expensive. Another possibility is that the
use of ward-level measures mis-specifies relationships
because wards are administrative areas, not true
‘communities’ which have meaning for individuals.
Further tests at a range of spatial scales might therefore
be desirable. A third possibility is that there are effects
of social capital on health related behaviours but they
are small and not detected through our response
variable (the probability of dying). A fourth possibility
is that there are other aspects of social capital, which
have not been considered here, which may have benefits
for individuals. In this paper we have only focussed on
ARTICLE IN PRESS
Table 4 (continued )
Model Logit (p) OR (95%CI)
L(A plus freq’ly meets locals)
Smoking 0.614 (o0.000) 1.85 (1.55–2.20)
Alcohol 0.063 (0.630) 1.07 (0.82–1.38)
Unknown alcohol 0.237 (0.007) 1.27 (1.07–1.51)
Diet 0.274 (0.001) 1.32 (1.11–1.55)
Exercise 0.492 (o0.000) 1.64 (1.36–1.97)
ð63:87 69:58Þ
ð59:97 63:86Þ
ðo59:96Þ
% who frequently meet
local people
9
>
=
>
;
0.174 (0.123) 0.84 (0.67–1.04)
0.322 (0.005) 0.72 (0.58–0.91)
0.461 (o0.000) 0.63 (0.50–0.80)
M(A plus feels local area friendly)
Smoking 0.654 (o0.000) 1.92 (1.61–2.29)
Alcohol 0.082 (0.531) 1.09 (0.84–1.40)
Unknown alcohol 0.256 (0.004) 1.29 (1.09–1.54)
Diet 0.291 (o0.000) 1.34 (1.13–1.58)
Exercise 0.501 (o0.000) 1.65 (1.37–1.99)
ð93:69 94:77Þ
ð92:23 93:68Þ
ðo92:22Þ
% who feel local area is
friendly
9
>
=
>
;
0.141 (0.218) 1.15 (0.92–1.44)
0.113 (0.346) 0.89 (0.71–1.13)
0.099 (0.417) 0.91 (0.71–1.15)
N(A plus blood donation)
Smoking 0.647 (o0.000) 1.91 (1.60–2.28)
Alcohol 0.075 (0.570) 1.08 (0.83–1.39)
Unknown alcohol 0.249 (0.005) 1.28 (1.08–1.53)
Diet 0.292 (o0.000) 1.34 (1.13–1.58)
Exercise 0.500 (o0.000) 1.65 (1.37–1.99)
ð99:58 113:95Þ
ð77:25 96:57Þ
ðo77:24Þ
Standardised blood
donorship ratio
9
>
=
>
;
0.089 (0.459) 0.91 (0.72–1.16)
0.005 (0.964) 1.01 (0.80–1.27)
0.164 (0.175) 1.18 (0.93–1.49)
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–1283 1281
only one aspect of social capital—that inspired by
Putnam’s work on civic participation—and it is possible
that the formulations of Coleman and Bourdieu, on
social relations and social networks, respectively, may
provide a way forward. Having said that, those concepts
have proved difficult to operationalise and measure
(though see Gatrell, Popay & Thomas, 2004).
However, an alternative suggestion can be made. In
this study we have not found conclusive evidence in
support of social capital as a contextual construct which
has an influence on health. In other work at the level of
standard regions (Mohan et al., 2004, Chapter 2)we
have similarly failed to find such connections. Given this
we would argue that there is a case for scepticism as to
whether social capital (as measured in this research)
adequately captures aspects of differentiation between
communities which contribute to health variations.
While we might not go so far as Foley and Edwards
(1999) when they suggest that the time has come to
disinvest in social capital, we suggest that our work adds
weight to a growing body of research (e.g. Pearce &
Davey-Smith, 2003) which challenges the explanatory
power of social capital (vis-a
`-vis material circumstances)
and which is therefore sceptical about whether demon-
strable health benefits will be obtained from investing in
social capital.
Acknowledgements
Thanks to the Health Development Agency for
funding the work on which this paper was based under
their ‘Social capital for health’ programme, and to S. V.
Subramanian and Sarah Curtis for their detailed and
constructive comments on an earlier version.
References
Anderson, E. (1990). Streetwise: Race, class and change in an
urban community. Chicago: University of Chicago Press.
Blakely, T. B., Kennedy, B., & Kawachi, I. (2001). Socio-
economic inequality in voting participation and self-rated
health. American Journal of Public Health,91, 99–104.
Blaxter, M., & Poland, F. (2003). Moving beyond the survey in
exploring social capital. In C. Swann, & A. Morgan (Eds.).,
Social capital for health: Insights from qualitative research
(pp. 87–107). London: Health Development Agency.
Campbell, B. (1993). Goliath: Britain’s dangerous places.
London: Methuen.
Carstairs, V., & Morris, R. (1991). Deprivation and health in
Scotland. Aberdeen: University Press.
Coulthard, M., Walker, A., & Morgan, A. (2002). People’s
perceptions of their neighbourhood and community involve-
ment: Results from the social capital module of the General
Household Survey 2000. London: TSO.
Cox, B.D. (1988). Health and Lifestyle Survey, 1984–5
computer file Colchester: ESRC Data Archive.
Cox, B., et al. (2001). Health and Lifestyle (HALS) Deaths
datafile: Second revision, January 2001. Essex University:
Data Archive.
Cowell, R., & Jehlicka, P. (1995). Backyard and biosphere: The
spatial distribution of support for English and Welsh
environmental organisations. Area,27, 110–117.
Foley, M., & Edwards, B. (1999). Is it time to disinvest in social
capital? Journal of Public Policy,19, 141–173.
Fukuyama, F. (1995). Trust: The social virtues and the creation
of prosperity. Harmondsworth: Penguin.
Gatrell, A., Popay, J., & Thomas, C. (2004). Mapping the
determinants of health inequalities in social space: can
Bourdieu help us? Health and Place, forthcoming.
Gilks, W. R., Richardson, S., & Spiegelhalter, D. J. (1996).
Markov chain Monte Carlo in Practice. London: Chapman
& Hall.
Goldstein, H. (1995). Multilevel statistical models. London:
Edward Arnold.
Goldstein, H., Rasbash, J., Plewis, I., et al. (1998). A User’s
Guide to MlwiN. Institute of Education, University of
London.
Hall, P. A. (1999). Social capital in Britain. British Journal of
Political Science,29, 417–461.
Jones, K., & Moon, G. (1993). Medical geography:
Taking place seriously. Progress in Human Geography,17,
515–524.
Jones, K., Mohan, J., Barnard, S., & Twigg, L. (2004).
Mapping the gift relationship: An analysis of the spatial
distribution of blood donors in England. Mimeo, Institute
for the Geography of Health, University of Portsmouth.
Jordan, G., & Maloney, W. (1997). The protest business.
Manchester: Manchester University Press.
Kawachi, I., & Berkman, L. (Eds.). (2003). Neighbourhoods and
health. Oxford: Oxford University Press.
Kawachi, I., Kennedy, B. P., & Glass, R. (1999). Social capital
and self-rated health: A contextual analysis. American
Journal of Public Health,89, 1187–1193.
Kawachi, I., Kennedy, B. P., Lochner, K., & Prothrow-Stith,
D. (1997). Social capital, income inequality and mortality.
American Journal of Public Health,87, 1491–1498.
Kawachi, I., Kennedy, B. P., & Wilkinson, R. G. (1999). Crime,
social disorganisation and relative deprivation. Social
Science & Medicine,48, 719–731.
Kendall, J., & Knapp, M. (1996). The voluntary sector in the
UK. Manchester: Manchester University Press.
Kennedy, B. P., Kawachi, I., & Brainerd, E. (1998). The role of
social capital in the Russian mortality crisis. World
Development,26(11), 2029–2043.
Krishna, A., Shrader, E. (2000). Cross-cultural measures of
social capital: A tool and results from Indian and Panama.
(Social Capital Initiative Working Paper 21). Washington
DC: World Bank.
Lindstrom, M., Merlo, J., & Ostergren, P.-O. (2003). Social
capital and sense of insecurity in the neighbourhood: A
population-based analysis in Malmo, Sweden. Social
Science & Medicine,56, 111–1120.
Lochner, K., Kawachi, I., Brennan, R., & Buka, S. (2003).
Social capital and neighbourhood mortality rates in
Chicago. Social Science & Medicine,56, 1797–1805.
ARTICLE IN PRESS
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–12831282
Macinko, J., & Starfield, J. (2001). The utility of social capital
in research on health determinants. The Milbank Quarterly,
79, 387–427.
MacIntyre, S., & Ellaway, A. (2003). Neighbourhoods and
health: An overview. In I. Kawachi, & L. Berkman (Eds.).,
Neighbourhoods and health (pp. 20–42). 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, 213–234.
Maloney, W., Smith, G., & Stoker, G. (2000). Social capital and
urban governance: Adding a more contextualised top-down
perspective. Political Studies, 48.
Miller, W., Timpson, A., & Lessnoff, M. (1996). Political
culture in contemporary Britain. Oxford: Clarendon Press.
Mohan, J., & Mohan, G. (2002). Placing social capital. Progress
in Human Geography,26, 191–210.
Mohan, J., Barnard, S., Jones, K., & Twigg, L. (2004). Social
capital, geography and health. London: Health Development
Agency.
Parry, G., Moser, G., & Day, N. (1992). Political participation
in Britain. Cambridge: Cambridge University Press.
Pattie, C., Dorling, D., Johnston, R., & Rossiter, D. (1996).
Electoral registration, population mobility and the demo-
cratic franchise: The geography of postal voters, overseas
voters and missing voters in Great Britain. International
Journal of Population Geography,2, 239–259.
Pattie, C., Seyd, P., & Whiteley, P. (2004). Atomised citizens.
Oxford: Oxford University Press.
Pearce, N., & Davey-Smith, G. (2003). In social capital the key
to inequalities in health? American Journal of Public Health,
93, 122–129.
Pickett, K., & Pearl, M. (2001). Multilevel analyses of
neighbourhood socioeconomic context and health out-
comes: A critical review. Journal of Epidemiology and
Community Health,55, 111–122.
Putnam, R. (1993). Making democracy work: Civic traditions in
modern Italy. New Jersey: Princeton University Press.
Putnam, R. (2002). Democracy in flux: The evolution of social
capital in contemporary society. New York: Oxford Uni-
versity Press.
Robert, S. (1999). Socioeconomic position and health: The
independent contribution of community socioeconomic
context. Annual Review of Sociology, 489–516.
Salamon, L. (1995). Partners in public service. Baltimore: John
Hopkins University Press.
Sampson, R. J., Morenoff, J. D., & Earls, F. (1999). Beyond
social capital: Spatial dynamics of collective efficacy for
children. American Sociological Review,64, 633–660.
Sampson, R., & Raudenbush, S. (1999). Systematic social
observation of public spaces: A new look at disorder in
urban neighbourhoods. American Journal of Sociology,105,
603–651.
Subramanian, S. V., Kawachi, I., & Kennedy, B. (2001). Does the
state you live in make a difference? Multilevel analysis of self-
ratedhealthintheUS.Social Science & Medicine,53, 9–19.
Subramanian, S. V., Kim, D., & Kawachi, I. (2002). Social trust
and self-related health in US communities: A multilevel
analysis. Journal of Urban Health,79, 521–534.
Szreter, S., & Woolcock, M. (2004). Health by association?
Social capital, social theory and the political economy of
public health. International Journal of Epidemiology,33,
650–667.
Titmuss, R. (1970). The gift relationship: From human blood to
social policy. Harmondsworth: Penguin.
Twigg, L., & Moon, G. (2002). Predicting small area health-
related behaviour: A comparison of multilevel synthetic
estimation and local survey data. Social Science & Medicine,
54(6), 931–937.
Verba, S., Schlozman, K., & Brady, H. (1996). Voice and
equality: Civic voluntarism in American politics. Boston:
Harvard University Press.
Walberg, P., et al. (1998). Economic change, crime and
mortality crisis in Russia: Regional analysis. British Medical
Journal,317, 312–318.
Whitley, E., Gunnell, D., Dorling, D., & Davey-Smith, G.
(1999). Ecological study of social fragmentation, poverty,
and suicide. British Medical Journal,319, 1034–1037.
Wilkinson, R. G. (1996). Unhealthy societies: The afflictions of
inequality. London: Routledge.
Wilkinson, R. G. (2001). Mind the gap: Hierarchies, health and
human evolution. London: Weiderfeld and Nicolson.
Williams, C. (2002). Harnessing voluntary work: A fourth
sector approach. Policy Studies,23, 247–268.
Wilson, W. J. (1987). The truly disadvantaged: The inner city,
the underclass and the public policy. Chicago: University of
Chicago Press.
Wolpert, J. (1990). Generosity and civic commitment: The local
public and voluntary sector. In R. Bennett (Ed.)., Decen-
tralisation, local government and markets (pp. 172–193).
Oxford: Oxford University Press.
ARTICLE IN PRESS
J. Mohan et al. / Social Science & Medicine 60 (2005) 1267–1283 1283
