Ethnicity, environment and health: putting ethnic inequalities in health in their place

Social Science & Medicine 55 (2002) 1647–1661

Ethnicity, environment and health: putting ethnic inequalities in health in their place Saffron Karlsena,*, James Y. Nazrooa, Rob Stephensonb a b

Department of Epidemiology and Public Health, University College London, 1-19 Torrington Place, London WC1E 6BT, UK Carolina Population Center, University square, CB# 8120, 123 West Franklin Street, Chapel Hill, North Carolina 27516, USA

Abstract We set out to explore the influence of environment on ethnic inequalities in health. Studies exploring the relationship between environment and health have tended to ignore the role of ethnicity, and the health impact of the residential concentration of ethnic minority groups in disadvantaged areas. Those that have explored the role of ethnicity tend to focus on the way in which residential concentration may promote a sense of community among ethnic minority groups, and, consequently, may be protective of health (the ‘ethnic density effect’). Again, they have tended to ignore the health impact of the concentration of ethnic minority groups in areas of social and economic disadvantage. We undertook a factor analysis to determine aspects of perception of ‘quality’ of the local environment, followed by multi-level analyses to explore the relationship between self-reported fair or poor health and individual- and ward-level characteristics among four ethnic groups (Caribbean, Indian, Pakistani and Bangladeshi, and white) in the UK. Results of the factor analysis suggested three underlying dimensions of perception of quality of the local area, related to the quality of the local environment, the provision of local amenities and local problems of crime and nuisance. These factors were entered into the multi-level models at level 2, along with indicators of ward-level ethnic density and Townsend’s deprivation score, with age, gender and household social class entered at level 1. In general, there was a residual random ward-level effect suggesting an area influence on self-assessed health. However, on the whole, none of the ward-level indicators showed any statistically significant association with self-assessed health, making it difficult to precisely determine the mechanisms operating. These findings suggest, though, that there is no ethnic density effect on selfassessed health for ethnic minority groups. r 2002 Elsevier Science Ltd. All rights reserved. Keywords: Ethnicity; Environment; Ethnic density; Health; Multi-level modelling; Social class, UK

Introduction Social inequalities in health have been described as resulting from the complex interplay of genetic, biological, social, environmental, cultural and behavioural factors (Department of Health, 1991,1999). But, in the exploration of ethnic health differentials, attention so far has focussed largely on the genetic, biological, cultural and behavioural influences, at the expense of wider *Corresponding author. Tel.: +44-20-7679-1713; fax: +4420-7813-0280. E-mail addresses: [email protected] (S. Karlsen), [email protected] (J.Y. Nazroo), [email protected] (R. Stephenson).

social and environmental factors (Bhopal, 1997; Sheldon & Parker, 1992). While there are now a large number of studies reporting associations between mortality and physical and psychological morbidity and environmental deprivation (e.g. Davey-Smith, Hart, Watt, Hole, & Hawthorne, 1998; Kennedy, Kawachi, Glass, & Prothrow-Stith, 1998; Ben-Shlomo, White, & Marmot, 1996; Kaplan, Pamuk, Lynch, Cohen, & Balfour, 1996; Kennedy, Kawachi, & Prothrow-Stith, 1996; Eames, Ben-Shlomo, & Marmot, 1993; see also Pickett & Pearl 2001), there remain few exploring the relationship between environmental disadvantage and poor health among ethnic minority groups. This paper will attempt to put ethnic inequalities in health ‘in their place’, through an exploration of associations between

0277-9536/02/$ - see front matter r 2002 Elsevier Science Ltd. All rights reserved. PII: S 0 2 7 7 - 9 5 3 6 ( 0 1 ) 0 0 2 9 7 - 0

1648

S. Karlsen et al. / Social Science & Medicine 55 (2002) 1647–1661

individual characteristics, area attributes and selfassessed fair or poor health. At the 1991 British Census, 5.5% of respondents described themselves as being of an ethnic group other than ‘white’ (Owen, 1994). Over three-fifths of these lived in the major urban areas of Greater London and the metropolitan counties of West Yorkshire and the West Midlands, where fewer than a fifth of the total white population lived. And more than half of people from ethnic minority groups lived in electoral wards1 where ethnic minority people made up over two-fifths of the population (Owen, 1994). Smaje (1995) shows how those areas with a high concentration of residents from ethnic minority groups also tend to be those that are more deprived according to environmental characteristics and service provision. In terms of the physical environment, a greater proportion of people from ethnic minority groups reside in urban and industrial areas, which, Smaje suggests, are also likely to be characterised by greater pollution and environmental toxicity than exists on average in the country (Smaje, 1995). A literature review of US based studies found that ethnic minority groups and those with a low social position were more likely to reside in areas near environmental hazards and with less likelihood of regulation, amelioration and clean up of those hazards, compared with other groups (Brown, 1995; see also Stretesky & Hogan, 1998). Such environmental hazards can put health at risk: for example, it has been argued that differences in housing conditions and local environments has lead to the increased risk of an elevated blood lead level among Black, compared with white, children in the US (Lanphear, Weitzman, & Eberly, 1996). It is possible that such a greater exposure to environmental hazards is also present among ethnic minority people in Britain. There are also a number of theoretical pathways that have been proposed as linking ethnic residential segregation (and geographical inequalities in income more generally) and poor health. Fiscella and Franks (1997) suggest four possible mechanisms which may operate. Firstly, any association between area and health may be simply confounded by other exogenous factors, for example perceived associations between mean income in an area and levels of mortality may in reality be the result of the relationship between family income, or racial discrimination, and both mean area income and mortality. Secondly, income inequality may be a marker of government under investment in human capital and other health promoting resources more generally (see Kawachi & Kennedy, 1997; Kawachi, Kennedy, Lochner, & Prothrow-Stith, 1997). This may 1 Electoral wards comprise the second level of Census geography in England and Wales (after enumeration district). In 1991, there were 9930 wards in England and Wales, each representing approximately 2000 households.

manifest itself in a number of ways, including limiting job opportunities and access to nutritious, affordable food. Kaplan et al. (1996) also found a strong relationship between under investment in education and income inequality. Thirdly, Fiscella and Franks (1997) suggest individual deprivation may be further aggravated, via the cognitive process, by perceived deprivation which leads to feelings of hopelessness and hostility and to health risk-taking. These feelings and behaviours may also be exacerbated by local characteristics: for example, Wilson’s (1991) discussion of ‘concentration effects’, whereby those residing in deprived neighbourhoods experience both the constraints imposed by wider society (unemployment etc), and the behaviours and truncated aspirations of others living in the neighbourhood. In this way, it may be suggested that the perceived deprivation felt by less, compared with more, wealthy individuals is exaggerated by the perceived deprivation sensed by others in their locality. Finally, economic segregation is seen as a measure of other social forces, primarily the extent of social cohesion (Fiscella & Franks, 1997; Kawachi & Kennedy, 1997). Social capital is seen as a public good created as a by-product of social relationships, the area equivalent of an individual’s social support networks. Disinvestment in social capital, it is argued, leads to a loss of social cohesion, community involvement and trust, which encourages hostility and suspicion and is detrimental to psychosocial well-being (Kawachi et al., 1997; Waitzman & Smith, 1998). Kawachi et al. (1997) found a strong relationship between levels of mortality in an area and various features of local social organisation, including civic participation and engagement, norms of reciprocity, and trust in others. It has been argued that ethnic residential concentration occurs as a result of discriminatory housing policy, economic restructuring and other external processes and constraints on patterns of residence (Smaje, 1995; see also Lakey, 1997; Waitzman & Smith, 1998; Hart, Kunitz, Sell, & Mukamel, 1998), all of which may be detrimental to health. But it is also argued that the residential concentration of ethnic minority groups enhances social cohesion, allowing the development of local economic opportunities, social support and other patterns of interaction that may be protective of health (Halpern & Nazroo, 2000; Smaje, 1995; Halpern, 1993). Hence, the residential concentration of ethnic minority groups has been seen both as a symptom of wider structural disadvantage and as a means by which the impact of this disadvantage may be limited. This has lead to a number of studies which have explored the supposed inverse relationship between the incidence of illness (or poor health in general) in a particular ethnic group and its size relative to the local population, known as the ‘ethnic density effect’ (Halpern & Nazroo, 2000; Neeleman & Wessely, 1999; Lackland Sam, 1998;

S. Karlsen et al. / Social Science & Medicine 55 (2002) 1647–1661

Smaje, 1995; Cochrane & Bal, 1988; Faris & Dunham, 1939). Smaje (1995) suggests that one’s ethnic community is the location for economic activity, kinship relations, social integration and religious worship. He argues that ethnic concentration allows individuals to develop positive roles and status not otherwise available, that it enhances social support and ‘buffers’ chronic stressors, such as racial harassment. It may also enable greater political mobilisation, enhance material opportunities and encourage healthy behaviour (Smaje, 1995). In this way, participation in a consciously realised ethnic community may promote health and well-being both directly and indirectly, through limiting the impact of socio-economic constraints and other forms of discrimination. But, one issue to be borne in mind here is how the inter-relationship between socio-economic and ethnic inequalities in health may disguise an ‘ethnic density effect’ (Sundquist, 1995; Nazroo, 1998). For example, the concentration of many ethnic minority people in socio-economic deprivation (Nazroo, 1998) means that those living in areas with a high density of ethnic minority groups are more likely to experience the negative health consequences of socio-economic disadvantage than their counterparts in areas with lower concentration. The interaction between these two characteristics could potentially disguise any variation between those living in high or low ethnic minority density areas: such that the poor health produced by socio-economic disadvantage is offset by the positive effects of high own ethnic density (i.e. an ‘ethnic density effect’), and vice versa. Not surprisingly, then, although most studies suggest a protective effect, evidence of an ‘ethnic density effect’ on health is contradictory. Lackland Sam (1998) found that for adolescents with an immigrant background in Norway living in an ethnically homogeneous neighbourhood was related to life satisfaction. Similarly, Neeleman and Wessely (1999) found an association between same ethnic group residential density and suicide among all ethnic groups (including white people) in South London, after adjusting for gender, age, socio-economic status and migration. Findings which were supported by Kelly et al’s. (in press) study of schizophrenia among non-white ethnic minority groups. Halpern and Nazroo (2000) found a modest association between local ethnic group concentration and levels of reported psychiatric symptoms, although the strength of this association varied across different ethnic groups. Smaje (1995) found some association between general and specific indicators of health, same ethnic group concentration and socio-economic status. However, in their study of inpatient admissions for schizophrenia among foreignborn immigrant groups in England in 1981, Cochrane and Bal (1988) found no relationship between schizophrenia and local same ethnic group density. And

1649

Neeleman, Wilson-Jones, and Wessely (2001) found an inverted U-shaped relationship between relative risks of deliberate self harm and the ethnic mix of the local area. What is missing in these studies, though, is any exploration of the health effects of other factors associated with ethnic residential concentration. An examination of these will allow us to further develop our understanding of both ethnic inequalities in health and how the local environment might affect health more generally.

Methods The Fourth National Survey of Ethnic Minorities (FNS) The FNS was undertaken in 1993 and 1994 by the Policy Studies Institute and Social and Community Planning Research (now the National Centre for Social Research). The FNS contained a nationally representative sample of 5196 people of Caribbean and Asian origin (Indian, Pakistani, Bangladeshi and Chinese) who were interviewed in detail, together with a comparison sample of 2867 white people. Respondents were allocated an ethnic group on the basis of their responses to a question on their family origins, a measure that had close correlation with a question very similar to that used in the 1991 British Census (Nazroo, 1997a). Interviewers were in most cases ethnically and language matched with respondents and interviews were carried out in the language(s) of the respondents’ choice. The sampling procedures were designed to select probability samples of both individuals and households. The areas used for sampling were selected on the basis of data from the 1991 Census on the ethnic minority population size in particular enumeration districts and electoral wards. This sampling method produced a sample that included respondents from areas with a low ethnic minority concentration, a population that has been ignored by other regional and national surveys of ethnic minority groups in the UK. Screening for ethnic minority respondents was carried out using focussed enumeration, with recruiters identifying households containing ethnic minority people by visiting, for example, every sixth address in a defined area and asking about the ethnic origin of those living at both the visited address and the five addresses on each side. In order to maximise the efficiency of the sampling process, in households containing ethnic minority people, two respondents were selected for interview whenever possible. White respondents were identified using a straightforward stratified sampling process, where areas, then addresses and then individuals within addresses were identified to be included in the study. Six weighting factors were applied to the data in order to deal with the complex sample design and to ensure

1650

S. Karlsen et al. / Social Science & Medicine 55 (2002) 1647–1661

that the survey sample represented the population under study as closely as possible. These accounted for differences in the probability of selection into the study, variations in the response rate by ethnicity, age and gender, and differences in the age profile of the sample compared to 1991 Census figures. For further details of the survey methodology and weighting see Smith and Prior (1996). The questionnaire covered a comprehensive range of information on both ethnicity and other aspects of the lives of ethnic minority people, including demographic and socio-economic factors. (For the demographic details of the sample and other findings from a preliminary analysis of the data, see Modood et al., 1997). The questionnaire also had detailed coverage of physical and mental health (Nazroo, 1997a, b). Here we focus on one outcome, self-assessed fair or poor health, which was rated as excellent, good, fair, poor or very poor. This analysis compares those describing their health as fair, poor or very poor with those describing their health as excellent or good. Preliminary analysis of the FNS data showed great similarity between Pakistani people and Bangladeshi people according to both health outcomes and sociodemographic profiles. To overcome the problem of small numbers, these groups were combined for this analysis. The sample for this analysis was: white respondents (N ¼ 2867); those of Caribbean descent (N ¼ 1205); those of Indian descent (N ¼ 2000); and those of Pakistani and Bangladeshi descent (N ¼ 1776). The

Chinese sample (N ¼ 214) was too small to be included in this analysis. This analysis used data based on the electoral ward. In 1991, there were 9930 wards in England and Wales, each representing approximately 2000 households. Fig. 1 shows the distribution of respondents within the wards used in this analysis. There were 250 wards included, each containing 30 respondents on average. Eleven of the nineteen wards containing fewer than ten respondents were in the London area. The rest were distributed throughout England and Wales. Measuring perceptions of the quality of the local area Respondents to the FNS were asked about a range of possible problems that might have occurred in their local areas, and to rate their local areas as good, poor, or neither good nor poor for a range of amenities. Factor analysis (Kim & Mueller, 1979) was used to identify underlying dimensions that represented respondents’ perceptions of the quality of their local environment. This technique is used to identify factors that can be used to represent the correlation among sets of interrelated variables. This was conducted for each ethnic group separately, then, because of the great similarity of these findings across ethnic groups, the final analysis was conducted for all ethnic groups combined (including whites). The final factors were therefore identical for each group included in subsequent analyses. The principal component method of factor extraction was

Fig. 1. Distribution of respondents within wards.

S. Karlsen et al. / Social Science & Medicine 55 (2002) 1647–1661

used. This produces factors in sequence according to the amount of the total sample variance they account for. The total variance explained by each factor is called the eigenvalue: this analysis reports only factors with an eigenvalue of 1 or over (Kim & Mueller, 1979). Principal component analysis was followed by oblique rotation to allow for correlation between the different factors identified. Cronbach’s alpha reliability coefficients were used to test for correlation between the variables clustering under each of the dimensions identified. Details of the questions included in the factor analysis are shown in the results section. Individual respondents were allocated a factor score for each of the area quality factors identified. These scores were averaged for respondents living in the same ward, allowing us to allocate a mean score to the wards from which respondents were sampled. This mean ward score was calculated for each factor and included as continuous variables (at level two) in the multi-level modelling process. These mean ward scores, and the Townsend deprivation score, were also divided into quintiles (based on the (more even) distribution of white respondents) for the bivariate analysis exploring the distribution of ward-level characteristics by different ethnic groups. Measuring area effects A multi-level modelling strategy was employed to explore the effects of electoral ward on the relationship between socio-economic and demographic factors and self-reported fair or poor health. Most studies that have attempted to explore associations between local same ethnic group density and other area effects, and the health of people from ethnic minority groups, have tended to use ordinary regression models, which cannot adequately take account of the clustering of individuallevel effects by area. Multi-level modelling allows us to determine more accurately the relationship between individual and area level effects on the relationship between ethnicity and health. This approach adopts a random effects model, which generalises the ordinary fixed-effects logistic (regression) model (which assumes all variables are independent) by assuming that the individual probabilities of reporting fair or poor health are equal to the fixed-effects model plus a random variation on the logit scale due to an unobserved, or unmeasured, ward-level effect. So, in a multi-level model, a significant random effects term illustrates that the odds of reporting a particular health status are correlated at the ward level, and that there is a degree of ward-level heterogeneity unaccounted for by the individual level fixed parameters in the model. The analysis reported here applies a two level random effects logistic regression model (with each individual respondent at level one and electoral ward at level two)

1651

to determine the factors influencing self-reported fair or poor health. The data were modelled for each ethnic group separately. The model uses a dichotomous variable, coded one for reporting fair, poor or very poor health, and zero for reporting good or excellent health. The random effects model can be written as follows: Logitðpij Þ ¼ x0ij b þ uj ; where pij is the probability of reporting fair or poor health for the ith respondent in the jth ward, x0ij is a vector of covariates corresponding to the ith respondent in the jth ward, b is a vector of unknown parameters, and uj is the random effect at the ward level. The distribution of the random effects is assumed to be normal, with mean zero and variance s2u : When su ¼ 0; the model reduces to the ordinary logistic model, indicating that there is no significant correlation in the risk of reporting fair or poor health between wards. The testing of the null hypothesis su ¼ 0 against the alternative hypothesis su > 0 is used to test the significance of random effects terms, using a modified likelihood ratio test. (For further details of multi-level modelling see Goldstein, 1995; Rice & Leyland, 1996). Despite the small number of respondents in some wards (described above), the extent to which this will produce an actual underestimation in the random effect variance in the multi-level model will also depend on the total number of wards included in the analysis. While there is no standard in respect to the minimum number of level two units required to significantly reduce the effect of small numbers within them, that each of these analyses include between 93 (in the white models) and 154 (in the Indian models) wards suggest any underestimation in the random effect variance will be minimal. The variables were entered into the model in sequence. Model zero included no independent variables, and can therefore give an indication of any initial statistically significant random ward-level effect, and how this is affected by the inclusion of the independent variables. Model one included individual-level indicators of age, gender and household social class only. Household social class was assigned using the head of household’s occupation. Where there was more than one working adult in the household, class was allocated on the basis of gender and age (e.g. men’s over women’s and father’s over son’s, where they were both below retirement age). Respondents were divided into non-manual and manual headed households and households with no full-time worker. Model two introduced a measure of ethnic group density in the ward, derived from the small area statistics of the 1991 British Census. For respondents from ethnic minority groups, this was expressed as the percentage of residents living in the respondent’s ward who were of the same ethnic group (coded: o5%; 5–15%; >15%). For

1652

S. Karlsen et al. / Social Science & Medicine 55 (2002) 1647–1661

white respondents this was expressed as the percentage of residents from any ethnic minority group living in the same ward as the respondent (coded: o2%; 2–5%; and >5%). Model three included Townsend’s deprivation score, which employs Census data on ward levels of unemployment, overcrowding, (lack of) owner occupied accommodation and (lack of) car ownership. The Townsend score is a summation of the standardised scores (z scores) for each variable, where scores greater than zero indicate greater levels of material deprivation. Model four introduced indicators summarising and averaging respondents’ views of the ‘quality’ of their local area, using continuous mean ward scores calculated from the factor analysis, as described above. Again, higher scores indicate more perceived problems in the local area.

Results Perceptions of the quality of the local area Factor analysis retaining all factors with an eigenvalue of 1 or over identified three dimensions of perceived quality of the local area. These dimensions have been given broad titles to aid the presentation of results as follows: Factor 1: Problems of crime and nuisance. The questions loading on this factor were: *

*

Would you say this area is ‘good’, ‘poor’, or ‘neither good nor poor’: * for safety on the streets? * for safety from burglaries? I am now going to read out a number of things that are sometimes problems where people live. Please tell me for each one whether or not it is a problem for you in this neighbourhood. Is that a serious problem or not? * Burglaries * Vandalism * Theft of cars * Troublesome teenagers * Troublesome children * Assaults * Harassment (Cronbach’s alpha=0.86) Factor 2: Lack of amenities. The questions loading on this factor were:

*

Would you say this area is ‘good’, ‘poor’, or ‘neither good nor poor’: * in its provision of shops? * for access to public transport facilities?

*

* * * *

in its provision of other leisure facilities, such as cinemas and theatres? for ease of getting to work? in its provision of places of worship? in its provision of schools? For access to parks and green areas?

(Cronbach’s alpha=0.61) Factor 3: Environmental problems. The questions loading on this factor were: *

I am now going to read out a number of things that are sometimes problems where people live. Please tell me for each one whether or not it is a problem for you in this neighbourhood. Is that a serious problem or not? * Run-down gardens * Run-down open spaces * Litter and rubbish * Vacant properties * Condition of paths, paving and roads * Dogs’ mess * Heavy traffic * Street parking * Infestations of vermin * Graffiti * Closeness of industry (Cronbach’s alpha=0.72)

The Cronbach’s alpha reliability coefficients suggest a high degree of correlation between the variables loading on each factor, and, therefore, low measurement error. Analysis of variance showed statistically significant variation by ethnic group in each of these three factor scores and the Townsend deprivation score (Table 1). Table 1 also presents the distribution of the different ethnic groups using the ward means, divided into quintiles, for each of the three dimensions of quality of the local area and the ward-level Townsend deprivation score. Forty-one per cent of white and 46% of Indian respondents lived in the two quintiles of wards reported to have the fewest problems with crime and nuisance, compared with around a quarter of Caribbean and Pakistani and Bangladeshi respondents. Conversely, 62% of Caribbean and 66% of Pakistani and Bangladeshi respondents lived in the two ward quintiles where the most problems with crime were reported. Between 66% and 76% of respondents from ethnic minority groups lived in the two ward quintiles with the fewest problems caused by a lack of local amenities, compared with 40% of white respondents. Over half of Indian respondents lived in the two ward quintiles with the fewest reported environmental problems, compared with around two-fifths of white, Caribbean and Pakistani and Bangladeshi respondents. A fifth of white respondents lived in the most deprived ward quintile according to the Townsend deprivation

1653

S. Karlsen et al. / Social Science & Medicine 55 (2002) 1647–1661

Table 1 Distribution of ethnic groups according to area, grouped by factor and Townsend scores in quintiles and mean factor and Townsend scores by ethnic group from analyses of variance Area scores

Ethnic group (%) Quintile

White

Caribbean

Indian

Pakistani and Bangladeshi

Problems with crime Lowest

Highest Mean scores (F ¼ 141:8; po0:001) Weighted count Unweighted count

19.7 20.7 21.5 18.9 19.3

12.7 13.2 12.0 23.9 38.2

23.9 22.3 11.5 20.2 22.1

15.1 12.7 6.5 23.6 42.1

13.1

13.9

12.9

14.2

2867 2867

1567 1205

2091 2000

1146 1776

Lack of amenities Lowest

Highest Mean scores (F ¼ 200:6; po0:001) Weighted count Unweighted count

20.2 20.3 18.7 20.0 21.8

41.0 25.3 10.1 11.6 13.0

43.3 23.3 9.0 9.6 14.8

48.5 27.1 8.8 7.8 7.8

10.7

10.0

9.9

9.7

2867 2867

1567 1205

2090 2000

1147 1776

Environmental problems Lowest

Highest Mean scores (F ¼ 111:9; po0:001) Weighted count Unweighted count

20.1 17.4 22.6 20.8 19.0

21.4 13.8 8.4 24.4 32.0

36.0 19.3 15.2 10.6 19.0

15.5 24.9 11.5 22.9 25.1

14.5

14.8

14.0

14.8

2867 2867

1567 1205

2091 2000

1146 1776

Townsend deprivation score Lowest

Highest Mean scores (F ¼ 859:6; po0:001) Weighted count Unweighted count

20.4 20.2 20.1 19.4 19.8

3.3 5.1 11.3 8.0 72.3

5.8 9.5 9.9 26.3 48.5

1.0 1.6 4.0 12.7 80.6

0.5

5.0

3.2

6.2

2867 2867

score, compared with 49% of Indian respondents, 72% of Caribbean respondents and 81% of Pakistani and Bangladeshi respondents.

1567 1205

2091 2000

1147 1776

An investigation of the Pearson product–moment correlation between the three factors and Townsend score suggested statistically significant, though some-

1654

S. Karlsen et al. / Social Science & Medicine 55 (2002) 1647–1661

times small, correlations between: local crime levels and environmental conditions (0.729, po0:001); local crime levels and Townsend score (0.455, po0:001); the availability of amenities locally and local environmental conditions (0.054, po0:001); and local environmental conditions and Townsend score (0.434, po0:001). There was an inverse relationship between the availability of amenities locally and Townsend score (0.264, po0:001). There was no statistically significant correlation between the availability of amenities locally and local crime levels (0.009, p ¼ 0:4). Multi-level modelling of self-reported fair or poor health White respondents Table 2 shows results of multi-level models zero to four for white respondents. Model zero, with no independent variables included, suggested a random ward-level effect of borderline statistical significance. This statistical significance was removed with the inclusion of variables exploring age, gender and household social class in model one. This model showed a significantly greater odds of self-reported fair or poor health with increasing age, among females and by household social class. Those from a household headed

by a manual worker were twice as likely to report fair or poor health and those from households with no full-time worker were almost two and a half times more likely to report fair or poor health, compared with those from households headed by someone in a non-manual occupation. These associations persisted throughout the modelling process. The proportion of local residents from an ethnic minority group showed no statistical relationship with health when it was included in model two, or in model three. Townsend score showed a statistically significant association with reporting fair or poor health when it was added in model three. However, the association between Townsend score and self-assessed health disappeared with the inclusion (in model four) of the factors exploring the quality of the local area. In model four age, being female, coming from a household with no full-time worker or a household headed by a manual worker, and coming from a ward where less than 5% of the local population is from an ethnic minority group, were significantly associated with greater odds of selfreported fair or poor health. In model four, those living in areas where fewer than 5% of the population were from ethnic minority groups had a one-third greater risk of reporting fair or poor health, compared with those in

Table 2 Odds ratios, with 95% confidence intervals, for the logistic model for self-reported fair or poor health. White respondents only Parameter

Model 0

Model 1

Model 2

Model 3

Model 4

Constant

0.46 (0.42–0.51)

0.16 (0.12–0.21)

0.16 (0.12–0.23)

0.13 (0.10–0.18)

0.02 (0.01–0.08)

F

1.01 (1.01–1.01)

1.01 (1.01–1.01)

1.01 (1.01–1.01)

1.01 (1.01–1.01)

F F

1.00 1.35 (1.14–1.60)

1.00 1.35 (1.14–1.60)

1.00 1.37 (1.15–1.62)

1.00 1.36 (1.15–1.61)

F F F

2.43 (1.43–4.14) 2.01 (1.67–2.43) 1.00

2.43 (1.43–4.13) 2.01 (1.66–2.43) 1.00

2.41 (1.42–4.09) 1.89 (1.56–2.28) 1.00

2.39 (1.41–4.07) 1.88 (1.55–2.27) 1.00

Density of ethnic minority groups Under 2% F Over 2%, but less than 5% F Over 5% F

F F F

0.93 (0.72–1.19) 1.02 (0.76–1.36) 1.00

1.17 (0.92–1.50) 1.24 (0.94–1.65) 1.00

1.33 (1.02–1.73) 1.32 (1.00–1.75) 1.00

Townsend Scale parameter

F

F

F

1.06 (1.03–1.08)

1.02 (0.99–1.06)

F

F

F

F

1.06 (0.99–1.12)

F

F

F

F

1.00 (0.94–1.07)

F

F

F

F

1.08 (0.98–1.18)

1.06 (1.00–1.13)

1.04 (0.99–1.10)

1.05 (0.99–1.11)

1.00 (0.96–1.04)

1.00 (1.00–1.00)

Age Scale parameter Gender Male Female Occupational class Non-working Manual Non-manual

Crime and nuisance in area Scale parameter Area amenities Scale parameter Area environment Scale parameter Random effect Scale parameter s

1655

S. Karlsen et al. / Social Science & Medicine 55 (2002) 1647–1661

areas where over 5% of the population were from ethnic minority groups.

Indian respondents Table 3 shows results of models zero to four for Indian respondents. Model one showed a statistically significant association between self-reported fair or poor health and age, gender and household social class. Being female was associated with a 40% greater risk of reporting fair or poor health, compared with being male. Those from households headed by a manual worker were 67% more likely to report fair or poor health and those from households with no full-time worker were over twice as likely to report fair or poor health, compared with those from households headed by someone in a non-manual occupation. These associations persisted after the inclusion of other indicators. Townsend’s area deprivation score also showed a statistically significant association with self-reported fair or poor health in model three, but this association disappeared after the inclusion of the three local area quality factors. There was a significant random wardlevel effect throughout the modelling process, although this was of only borderline statistical significance in

model zero. In the final model, age, gender, and household occupational class were all associated with self-reported fair or poor health, although the ward-level indicators (Townsend score, the local area quality factors and ethnic density) showed no association with health. Pakistani and Bangladeshi respondents Table 4 shows results of models zero to four for Pakistani and Bangladeshi respondents. All four models including independent variables showed a statistically significant association between self-reported fair or poor health and being female, age and coming from a household headed by someone in a manual, rather than a non-manual occupation. There was also a large statistically significant random ward-level effect throughout the modelling process. Townsend score, ethnic density and the area quality factors were not associated with self-assessed fair or poor health. Caribbean respondents Table 5 shows results of models zero to four for Caribbean respondents. All four models including independent variables showed statistically significant associations between self-reported fair or poor health

Table 3 Odds ratios, with 95% confidence intervals, for the logistic model for self-reported fair or poor health. Indian respondents only Parameter

Model 0

Model 1

Model 2

Model 3

Model 4

Constant

0.41 (0.37–0.46)

0.03 (0.02–0.04)

0.03 (0.02–0.04)

0.02 (0.01–0.04)

0.01 (0.00–0.02)

F

1.06 (1.05–1.06)

1.06 (1.05–1.06)

1.06 (1.05–1.06)

1.06 (1.05–1.06)

Age Scale parameter Gender Male Female Occupational class Non-working Manual Non-manual

F F

1.00 1.40 (1.13–1.74)

1.00 1.40 (1.13–1.74)

1.00 1.41 (1.13–1.75)

1.00 1.40 (1.13–1.73)

F F F

2.29 (1.13–4.01) 1.67 (1.33–2.08) 1.00

2.29 (1.31–4.01) 1.66 (1.33–2.08) 1.00

2.21 (1.26–3.90) 1.57 (1.25–1.97) 1.00

2.25 (1.28–2.97) 1.59 (1.27–2.00) 1.00

Density of own ethnic group Under 5% Over 5%, but less than 15% Over 15%

F F F

F F F

1.00 1.17 (0.84–1.62) 1.01 (0.72–1.42)

1.00 0.96 (0.68–1.34) 0.79 (0.56–1.14)

1.00 0.98 (0.78–1.24) 0.86 (0.59–1.24)

Townsend Scale parameter

F

F

F

1.06 (1.02–1.10)

1.04 (0.99–1.09)

F

F

F

F

1.05 (0.95–1.16)

Crime and nuisance in area Scale parameter Area amenities Scale parameter Area environment Scale parameter

F

F

F

F

1.04 (0.931.17)

F

F

F

F

1.04 (0.92–1.17)

Random effect Scale parameter s

1.10 (1.00–1.22)

1.17 (1.03–1.33)

1.18 (1.04–1.35)

1.15 (1.01–1.30)

1.15 (1.02–1.31)

1656

S. Karlsen et al. / Social Science & Medicine 55 (2002) 1647–1661

Table 4 Odds ratios, with 95% confidence intervals, for the logistic model for self-reported fair or poor health. Pakistani and Bangladeshi respondents only Parameter

Model 0

Model 1

Model 2

Model 3

Model 4

Constant

0.52 (0.45–0.61)

0.02 (0.02–0.04)

0.02 (0.01–0.03)

0.02 (0.01–0.04)

0.01 (0.00–0.04)

F

1.07 (1.06–1.08)

1.07 (1.06–1.08)

1.07 (1.06–1.08)

1.07 (1.06–1.08)

Age Scale parameter Gender Male Female Occupational class Non-working Manual Non-manual

F F

1.00 1.34 (1.06–1.68)

1.00 1.34 (1.07–1.68)

1.00 1.34 (1.07–1.68)

1.00 1.34 (1.07–1.69)

F F F

1.32 (0.77–2.26) 1.62 (1.25–2.12) 1.00

1.32 (0.77–2.26) 1.61 (1.23–2.11) 1.00

1.33 (0.77–2.29) 1.63 (1.24–2.13) 1.00

1.33 (0.77–2.29) 1.64 (1.25–2.15) 1.00

Density of own ethnic group Under 5% Over 5%, but less than 15% Over 15%

F F F

F F F

1.00 1.12 (0.69–1.81) 1.10 (0.71–1.71)

1.00 1.17 (0.70–1.94) 1.25 (0.70–2.21)

1.00 1.17 (0.70–1.94) 1.41 (0.77–2.58)

Townsend Scale parameter

F

F

F

0.98 (0.92–1.04)

0.96 (0.89–1.02)

F

F

F

F

1.02 (0.91–1.14)

Crime and nuisance in area Scale parameter Area amenities Scale parameter Area environment Scale parameter

F

F

F

F

1.02 (0.88–1.16)

F

F

F

F

1.08 (0.92–1.28)

Random effect Scale parameter s

1.39 (1.16–1.67)

1.60 (1.25–2.03)

1.63 (1.27–2.09)

1.64 (1.28–2.11)

1.61 (1.26–2.05)

and age, being female and household social class. Coming from a household headed by a manual worker produced an increased likelihood of reporting fair or poor health of over 60% and coming from a household with no full-time worker was associated with around a five times greater likelihood of reporting fair or poor health, compared with those from households headed by someone in a non-manual occupation. There was no statistically significant random ward-level effect in model zero. The random ward-level effect showed borderline significance in models one, two and three. This significant random ward-level effect disappeared with the inclusion of the local area quality factors in model four. However, in the final model measures of ethnic density, local area quality factors and the Townsend score were also not significantly associated with self-assessed fair or poor health.

Discussion Three factors emerged in the factor analysis of responses to questions on perceptions of the quality of the local area. These were: local problems with crime

and nuisance, a local lack of amenities and local environmental problems. These are consistent with the two dimensions of community quality determined by Molinari, Ahern and Hendryx (1998), which they called ‘community problems’ (which included problems of crime and nuisance and local lack of amenities) and ‘environmental problems’. These are also consistent with aspects of the local environment that others have suggested may influence health (Sooman & Macintyre, 1995; Macintyre, MacIver & Sooman, 1993; see also Smaje, 1995). Exploration of the correlation between the three factors and Townsend score suggested there would be some similarity in the ward scores across the different indicators. The exception to this was scores for the factor exploring the availability of local amenities, compared with those for local crime levels. These associations would suggest that there is some congruence between the more subjective assessments of area ‘quality’ (as measured by the factors) and the more objective assessment of area deprivation made using the Townsend score. When exploring the three dimensions of area quality and Townsend’s score for each ethnic group, Caribbean and Pakistani and Bangladeshi respondents were found

1657

S. Karlsen et al. / Social Science & Medicine 55 (2002) 1647–1661

Table 5 Odds ratios, with 95% confidence intervals, for the logistic model for self-reported fair or poor health. Caribbean respondents only Parameter

Model 0

Model 1

Model 2

Model 3

Model 4

Constant

0.63 (0.55–0.72)

0.04 (0.03–0.06)

0.04 (0.02–0.06)

0.04 (0.02–0.06)

0.03 (0.01–0.14)

F

1.05 (1.05–1.06)

1.05 (1.05–1.06)

1.05 (1.05–1.06)

1.05 (1.05–1.06)

Age Scale parameter Gender Male Female Occupational class Non-working Manual Non-manual

F F

1.00 1.54 (1.18–2.00)

1.00 1.52 (1.16–1.99)

1.00 1.52 (1.16–1.99)

1.00 1.53 (1.17–2.00)

F F F

5.48 (1.56–19.3) 1.67 (1.26–2.20) 1.00

5.05 (1.42–18.0) 1.64 (1.24–2.16) 1.00

5.00 (1.41–17.8) 1.62 (1.22–2.15) 1.00

4.68 (1.33–16.6) 1.63 (1.23–2.17) 1.00

Density of own ethnic group Under 5% Over 5%, but less than 15% Over 15%

F F F

F F F

1.00 0.99 (0.70–1.40) 1.46 (0.97–2.20)

1.00 0.94 (0.62–1.41) 1.34 (0.80–2.27)

1.00 0.88 (0.59–1.33) 1.30 (0.78–2.17)

Townsend Scale parameter

F

F

F

1.01 (0.96–1.07)

1.00 (0.95–1.06)

F

F

F

F

0.94 (0.86–1.04)

Crime and nuisance in area Scale parameter Area amenities Scale parameter Area environment Scale parameter

F

F

F

F

0.95 (0.84–1.08)

F

F

F

F

1.12 (0.99–1.26)

Random effect Scale parameter s

1.14 (0.99–1.30)

1.19 (1.00–1.42)

1.18 (1.00–1.39)

1.18 (1.00–1.40)

1.15 (0.98–1.34)

to be concentrated in the most disadvantaged wards, according to the factors summarising responses on local environmental problems, local problems of crime and nuisance and Townsend’s ward deprivation score. Indian respondents were also found to reside in relatively disadvantaged wards, according to Townsend’s score. In contrast, white respondents were more likely than respondents from ethnic minority groups to report that they lived in wards that were lacking amenities. In general, these findings are consistent with previous studies showing the residential concentration of people from ethnic minority groups in environmentally and economically deprived areas (Smaje, 1995; Owen, 1994). It might appear, however, that white respondents, who are traditionally perceived as residing in more prosperous areas, could experience shortages in the provision of amenities locally. But, viewed in comparison with the findings of Lakey (1997), who showed that, given similar circumstances, white respondents were more likely to express dissatisfaction with their local areas than people from ethnic minority groups, it is possible that this is a consequence of differences in expectations about, or desires for, local provision held by people from ethnic minority and white groups, rather

than actual differences in the local provision of amenities. Associations between the five area indicators (sameethnic group density, local environmental problems, local problems of crime and nuisance, a local lack of amenities and Townsend’s ward deprivation score) and self-reported fair or poor health did not appear to influence health, although it was with the inclusion of the area quality factors that the random ward-level effect lost any statistical significance among the Caribbean group. The Townsend deprivation score showed a statistically significant association with self-reported fair or poor health for white and Indian respondents, but on both occasions this association disappeared with the inclusion of the area quality factors. Levels of deprivation in the ward appeared to have little additional health effect for the least affluent ethnic groups (those from Caribbean and Pakistani and Bangladeshi groups). That any significant association between Townsend’s score and self-assessed fair or poor health was removed with the inclusion of the local area quality factors would suggest that the additional health effect associated with ward levels of deprivation for the more affluent ethnic groups (those from the white and

1658

S. Karlsen et al. / Social Science & Medicine 55 (2002) 1647–1661

Indian groups) may operate through characteristics of the local area. On the surface, this would suggest that individual characteristics are more influential than ward-level characteristics in terms of health (particularly among less affluent groups), a conclusion also reached by other studies (Fiscella & Franks, 1997). This is also supported by Wilkinson’s (1997) suggestion that the level of resource development may be more critical than resource distribution in smaller geographical units, such that increases in median levels of income may contribute more to health improvements among the very poor than reducing levels of inequality. It is interesting to explore these findings in relation to the various pathways proposed as operating between ethnic and socioeconomic residential segregation and health, discussed earlier. Beginning with the second of the pathways proposed, it may be argued that the area quality factors determined here (exploring a lack of local amenities, poor environmental circumstances and concerns about crime) could be useful, if not exhaustive, indicators of government under investment in human capital. That none of these indicators showed any statistical association with health, and were effective only to reduce the random ward-level effect from borderline to no statistical significance for the Caribbean group, would add weight to suggestions that other issues are important in the relationship between residential segregation and self-assessed health. Similarly, that there was no relationship between Townsend’s deprivation score and self-reported fair or poor health would suggest the impact of relative deprivation is likely to be minimal. However, while there was some consistency in the statistical associations found in the multi-level modelling, the variation in the extent and behaviour of the random ward-level effect would suggest both that the impact of area on health differs by ethnic group, and, at least for the South Asian groups, that there is some other aspect of the local area which is unaccounted for by these indicators, but which has a significant influence on health. One possibility is that while our area quality factors would seem to correspond with the possible mechanisms through which environment may influence health identified by others (Sooman & Macintyre, 1995; Macintyre et al., 1993), some of these mechanisms will have been inadequately accounted for by this analysis. For example, Macintyre et al. (1993) identify the following: the physical features of the environment (which are shared by all residents in the locality); the availability of healthy/unhealthy environments at home, at work and at play (opportunities which may or may not be taken up, so may not affect everyone in the area in the same way); the services provided, privately or publicly, to support people in their daily lives, including education, local amenities, community organisations

and health and welfare services; socio-cultural aspects of the neighbourhood, which include the historical and current characteristics of the community, norms and values, community integration, and perceived threats to personal safety and crime; and the reputation of a neighbourhood: including how it is perceived by residents and outsiders, including planners and providers; and how it effects the self esteem and morale of the residents and local geographical mobility. So while our measures would seem to be related to the physical features of the environment, some of the services provided locally and some socio-cultural aspects of the neighbourhood, they are unlikely to have fully accounted for the health impact of: the broader physical features of the environment (for example, environmental pollution); the quality of the environment ‘at home, at work and at play’; the local provision of educational and health and welfare services and community organisations; certain socio-cultural aspects of the environment, including historical characteristics, norms and values and community integration; and the reputation of the neighbourhood. The availability of measures which were also able to explore these aspects of the relationship between environment and health may have shed further light on the pathways operating to produce the relationship between area and health found here. The final pathway between area and health discussed earlier is the role of social cohesion. While we have not included indicators of, for example, community involvement and trust per se, theories of the ethnic density effect suggest that the residential concentration of ethnic groups allows the development of economic and social support, which both promote social cohesion and protect health (Halpern & Nazroo, 2000; Smaje, 1995; Halpern, 1993). But white respondents were the only group for whom there was an association between selfreported fair or poor health and ethnic group density. And this association between local ethnic minority density and health was curious as it suggested an increased risk of ill health for white respondents in wards with a lower concentration of ethnic minority groups, which would seem to contradict theories of the health effect of ethnic density described earlier (Halpern & Nazroo, 2000; Neeleman & Wessely, 1999; Lackland Sam, 1998; Smaje, 1995; Cochrane & Bal, 1988; Faris & Dunham, 1939). That this density effect was present after adjusting for both individual and area characteristics, it would appear that it is some other aspect of living in the same residential area as people from ethnic minority groups that predicts better health for white people. This may be to do with an improved comparative social status, perhaps as a result of perceived ‘racial’ status, or may be the result of gentrification, where more prosperous (and therefore more healthy and primarily white) people repopulate less prosperous areas, in which ethnic minority groups are concentrated. Whichever it

S. Karlsen et al. / Social Science & Medicine 55 (2002) 1647–1661

is, these findings suggest either that local same-ethnic density does not promote social cohesion, or that levels of social cohesion do not provide the link between areas and health. It is worthwhile comparing these findings with those of the analysis of FNS data undertaken by Halpern and Nazroo (2000), which found a small but statistically significant association between ward-level same ethnic group density and mental health. There may be a number of reasons for the apparent inconsistency between our findings and those of Halpern and Nazroo (2000). Most obviously, higher same ethnic density could be more protective of mental than physical health, that is the effect of ethnic density might vary for different health indicators, and this might go some way towards explaining the inconsistency in ethnic density effect found in other studies, described earlier. Or, the association found by Halpern and Nazroo (2000) may have been an artefact of the single level model used in their analysis, which did not allow for the geographical clustering of individuals. Halpern and Nazroo (2000), building on a review of the literature on ethnic group density effects (Halpern, 1993), also suggest that it is ethnic density at the local, rather than the regional or national, level that is critical for (in their case, mental) health, and as such electoral wards, the geographical level at which this analysis was conducted, may still be too large to recognise the effects of local ethnic density. Unfortunately, the FNS sample was insufficiently concentrated to allow us to use smaller areas in the multilevel models. There is currently no formal theoretical rationale for the choice of geographical unit to be used in analyses such as these (Waitzman & Smith, 1998), and the range of theoretical pathways through which area characteristics may influence health (discussed earlier) would suggest a range of appropriate units, of widely varying sizes (for further discussion of this see Fiscella & Franks, 1997). A further issue here may be that electoral wards are not defined with reference to any particular social or economic framework, and so the extent to which any positive findings can be applied to the theoretical pathways proposed may be questionable. It may also be that the inconsistent health impact of ethnic density found in ours and earlier studies may be due to the influence of other individual characteristics. For example, a key feature in the relationship between ethnicity, health and residential environment, including own ethnic group density, will be the way in which ethnic identity may influence the engendering of a sense of community. Communities are not natural entities, but are constructed by people around proximity and social identification. And this may be one explanation for the association between a lower sense of community and having more perceived problems with levels of choice, cost and satisfaction in people’s interactions with the

1659

health-care sector found by Ahern and Hendryx (1998) (see also Ahern, Hendryx, & Siddharthan, 1996). So, while a ‘strong’ ethnic identity may lead to a stronger sense of community that may be protective of health (through psychological benefits) for those living in an area with large numbers of people from a similar background, it may produce detrimental health effects for those living elsewhere as a result of feelings of exclusion or isolation. Also, interaction with other ethnic, religious and other social groups has been found to affect the development of ethnic identity (Jacobson, 1997; Verkuyten, 1997), which could suggest that the health importance of local ethnic density may also vary. Hall (1992) discusses the way in which globalisation and sustained migration has lead to the pluralisation (or translation) of cultures and identities into new, ‘hybrid’ forms, which could suggest the diminished importance of a same-ethnic community to identity, and perhaps health. He also suggests this ‘global homogenisation’ may be matched by a revival of ‘ethnicity’: that identities may react by becoming more traditional, ‘attempting to restore their former purity and recover the unities and certainties which are felt as being lost’ (Hall, 1992, p. 309), a trend which would suggest a greater role for the support provided by the local same-ethnic community. Allied to this, however, is the way that technological advances, particularly improvements in global communications, may mean that the need for a geographically local community has diminished more recently, as the ability to interact with more greatly dispersed people has grown. In this way, the establishment of social support networks may no longer be reliant on geographical proximity, even for those, as described above, who retain more traditional identities or find themselves in more hostile local situations. These possibilities would suggest that the role of local ethnic density may be changing and this may also go some way toward explaining the apparent contradictory findings of the different studies exploring its relationship with health. It is worth pointing to some limitations of the current analysis. The cross-sectional nature of the data means it is difficult to precisely determine the direction of the relationship between individual and neighbourhood characteristics and health experience. This is particularly problematic for the investigation of influences which may take some time to be realised, such as the health impact of characteristics of the local area. Also, the small numbers available meant it has been necessary to combine heterogeneous groups for the analysis. There are also small numbers of respondents in particular wards, and the geographical clustering of ethnic minority groups will mean there are also especially small numbers of respondents from some ethnic minority groups in particular wards. While the relatively large number of wards included in the analyses should

1660

S. Karlsen et al. / Social Science & Medicine 55 (2002) 1647–1661

have accounted for any underestimation in the random effect variance, the lack of statistically significant associations between our ward-level indicators and self-assessed fair or poor health could suggest measurement imprecision or the influence of the small numbers of respondents in some wards. Finally, if there was a categorical difference in the effect of ethnic density at a particular level, then our banding may not recognise this, which could explain the lack of association between ethnic density and health for the ethnic minority groups included in this analysis. If, however, ethnic density has a continuous effect on health, our banding should not disguise this.

Conclusion In our analysis individual level variables (age, gender and social class) were important predictors of selfassessed fair or poor health. There was also some statistically significant random area level effect for all the ethnic groups examined. For the Caribbean group, this area level effect was ‘explained’ by the inclusion of area level variables representing local same-ethnic group density, Townsend’s deprivation score and factors related to local environmental problems, local problems of crime and nuisance and a lack of local amenities. In addition to this, the local density of ethnic minority groups was associated with self-reported fair or poor health for the white group and Townsend’s deprivation score was associated with self-reported fair or poor health for the white and Indian group, in at least one stage of the modelling. However, on the whole, there were no statistically significant area level variables in the final models, suggesting a possible lack of measurement precision and making it difficult for us to precisely determine the mechanisms operating at an area level. Nevertheless, the evidence suggests that, apart from for white people, there is no ethnic density effect on selfassessed health.

Acknowledgements This paper draws on research funded by the ESRC (L128251019) under the Health Variations Programme. The data used are drawn from the Fourth National Survey of Ethnic Minorities and thanks are due to the funders of the survey (particularly the Department of Health), advisory groups, colleagues at the Policy Studies Institute and the National Centre for Social Research, and, most importantly, the thousands of respondents who gave their time. The authors would also like to thank Kwame McKenzie, Richard G. Wilkinson and the anonymous reviewers for their helpful comments.

References Ahern, M. M., & Hendryx, M. S. (1998). Community characteristics as predictors of perceived HMO quality. Health and Place, 4(2), 151–160. Ahern, M. M., Hendryx, M. S., & Siddharthan, K. (1996). The importance of sense of community on people’s perceptions of their health-care experiences. Medical Care, 34(9), 911–923. Ben-Shlomo, Y., White, I. R., & Marmot, M. (1996). Does the variation in the socioeconomic characteristics of an area affect mortality. British Medical Journal, 213, 1013–1014. Bhopal, R. (1997). Is research into ethnicity and health racist, unsound or important science? British Medical Journal, 314, 1751–1756. Brown, P. (1995). Race, class and environmental health: A review and systematization of the literature. Environmental Research, 69, 15–30. Cochrane, R., & Bal, S. S. (1988). Ethnic density is unrelated to incidence of schizophrenia. British Journal of Psychiatry, 153, 363–366. Davey-Smith, G., Hart, C., Watt, G., Hole, D., & Hawthorne, V. (1998). Individual social class, area-based deprivation, cardiovascular disease risk factors, and mortality: The Renfrew and Paisley study. Journal of Epidemiology and Community Health, 52, 399–405. Department of Health .(1991). The health of the nation: A consultative document for health in England. London: HMSO. Department of Health .(1999). Saving lives: our healthier nation Cm4386. London: HMSO. Eames, M., Ben-Shlomo, Y., & Marmot, M. G. (1993). Social deprivation and premature mortality: Regional comparison across England. British Medical Journal, 307, 1097–1102. Faris, R. E., & Dunham, H. W. (1939). Mental disorders in urban areas. Chicago: University of Chicago Press. Fiscella, F., & Franks, P. (1997). Poverty or income inequality as predictor of mortality: Longitudinal cohort study. British Medical Journal, 314, 1724. Goldstein, H. (1995). Kendall’s Library of Statistics 3: Multi-level statistical models. London: Arnold (also available at http://www.arnoldpublishers.com/support/goldstein.htm). Hall, S. (1992). The question of cultural identity. In S. Hall, D. Held & T. McGrew (Eds.). Modernity and its Futures (pp. 274–325). Milton Keynes: Open University Press. Halpern, D. (1993). Minorities and mental health. Social Science and Medicine, 36(5), 597–607. Halpern, D., & Nazroo, J. (2000). The ethnic density effect: Results from a national community survey of England and Wales. International Journal of Social Psychiatry, 46(1), 34– 46. Hart, K. D., Kunitz, S. J., Sell, R. R., & Mukamel, D. B. (1998). Metropolitan governance, residential segregation and mortality among African Americans. American Journal of Public Health, 88(3), 434–438. Jacobson, J. (1997). Religion and ethnicity: Dual and alternative sources of identity among young British Pakistanis. Ethnic and Racial Studies, 20(2), 238–256. Kaplan, G. A., Pamuk, E. R., Lynch, J. W., Cohen, R. D., & Balfour, J. L. (1996). Inequality in income and mortality in

S. Karlsen et al. / Social Science & Medicine 55 (2002) 1647–1661 the United States: Analysis of mortality and potential pathways. British Medical Journal, 312, 999–1003. Kawachi, I., & Kennedy, B. P. (1997). Socioeconomic determinants of health: Health and social cohesion: Why care about income inequality? British Medical Journal, 314, 1037. 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. Kelly, J., van Os, J., McKenzie, K., Allardyce, J., Goel, R., McCreadie, R.G., & Murray, R.M. (in press). The incidence of schizophrenia in ethnic minorities is highest in London neighbourhoods which are mainly white. British Medical Journal. Kennedy, B. P., Kawachi, I., Glass, R., & Prothrow-Stith, D. (1998). Income distribution, socioeconomic status and self rated health in the United States: Multilevel analysis. British Medical Journal, 317, 917–921. Kennedy, B. P., Kawachi, I., & Prothrow-Stith, D. (1996). Income distribution and mortality: Cross-sectional ecological study of the Robin Hood Index in the US. British Medical Journal, 312, 1004–1007. Kim, J. O., & Mueller, C. W. (1979). Factor analysis: Statistical methods and practical issues. London: Sage. Lackland Sam, D. (1998). Predicting life satisfaction among adolescents from immigrant families in Norway. Ethnicity and Health, 3(1/2), 5–18. Lakey, J. (1997). Neighbourhoods and housing. In T. Modood, R. Berthoud, J. Lakey, J. Nazroo, P. Smith, S. Virdee, & S. Beishon (Eds.), Ethnic minorities in Britain: Diversity and disadvantage (pp. 184–223). London: Policy Studies Institute. Lanphear, B. P., Weitzman, M., & Eberly, S. (1996). Racial differences in urban children’s environmental exposures to lead. American Journal of Public Health, 86(10), 1460–1463. Macintyre, S., MacIver, S., & Sooman, A. (1993). Area, class and health: Should we be focussing on places or people? Journal of Social Policy, 22(2), 213–234. Modood, T., Berthoud, R., Lakey, J., Nazroo, J., Smith, P., Virdee, S., & Beishon, S. (1997). Ethnic minorities in britain: Diversity and disadvantage. London: Policy Studies Institute. Molinari, C., Ahern, M., & Hendryx, M. (1998). The relationship of community quality to the health of women and men. Social Science and Medicine, 47(8), 1113–1120. Nazroo, J. Y. (1997a). The health of Britain’s ethnic minorities: Findings from a national survey. London: Policy Studies Institute. Nazroo, J. Y. (1997b). Ethnicity and mental health: Findings from a national community survey. London: Policy Studies Institute.

1661

Nazroo, J. Y. (1998). Genetic, cultural or socio-economic vulnerability? Explaining ethnic inequalities in health. Sociology of Health and Illness,, 20(5), 710–730. Neeleman, J., & Wessely, S. (1999). Ethnic minority suicide: A small area geographical study in South London. Psychological Medicine, 29(2), 429–436. Neeleman, J., Wilson-Jones, C., & Wessely, S. (2001). Ethnic density and deliberate self harm; a small area study in south east London. Journal of Epidemiology and Community Health, 55, 85–90. Owen, D. (1994). Spatial variations in ethnic minority group populations in Great Britain. Population Trends, 78, 23–33. Pickett, K. E., & Pearl, M. (2001). Multilevel analyses of neighbourhood socioeconomic context and health outcomes: A critical review. Journal of Epidemiology and Community Health, 55, 111–122. Rice, N., & Leyland, A. (1996). Multilevel models: Applications to health data. Journal of Health Services Research and Policy, 1(3), 154–164. Sheldon, T. A., & Parker, H. (1992). Race and ethnicity in health research. Journal of Public Health Medicine, 14(2), 104–110. Smaje, C. (1995). Ethnic residential concentration and health: Evidence for a positive effect? Policy and Politics, 23(3), 251–269. Smith, P., & Prior, G. (1997). The Fourth National Survey of Ethnic Minorities: Technical Report. London: Social and Community Planning Research. Sooman, A., & Macintyre, S. (1995). Health and perceptions of the local environment in socially contrasting neighbourhoods in Glasgow. Health and Place, 1(1), 15–26. Stretesky, P., & Hogan, M. J. (1998). Environmental justice: An analysis of Superfund sites in Florida. Social Problems, 45(2), 268–287. Sundquist, J. (1995). Ethnicity, social class and health. A population-based study on the influence of social factors in self-reported illness in 223 Latin American refugees, 333 Finnish and 126 South European labour migrants and 841 Swedish controls. Social Science and Medicine,, 40(6), 777–787. Verkuyten, M. (1997). Discourses of ethnic minority identity. British Journal of Social Psychology, 36, 565–586. Waitzman, N. J., & Smith, K. R. (1998). Separate but lethal: The effects of economic segregation on mortality in metropolitan America. The Milbank Quarterly, 76(3), 341– 373. Wilkinson, R. G. (1997). Comment: Income, inequality and social cohesion. American Journal of Public Health, 87, 1504–1506. Wilson, W. J. (1991). Studying inner-city social dislocations: The challenge of public agenda research. American Sociological Review, 56(1), 1–14.