Stjärne, Maria K.*†; Fritzell, Johan*; De Leon, Antonio Ponce§¶; Hallqvist, Johan†‡; for the SHEEP Study Group
Whether there is a contextual effect of a neighborhood's social and economic resources on the risk of coronary heart disease (CHD) has not yet been settled. However, living in socioeconomically disadvantaged neighborhoods has been associated with greater risk of myocardial infarction specifically,1–3 as well as a higher prevalence, and mortality from overall CHD.4–6 There is no consensus in the literature regarding what features of the context might have etiologic implications for cardiovascular outcomes, although it has been suggested7 that, at the neighborhood level, material and infrastructure resources make stronger contributions than aspects of collective social functioning.
Similarly, in the literature on the health effects of income inequality, there is an extensive debate regarding the mechanisms behind and the possible contextual effects of such inequality.8–12 One hypothesis is that the effect of income inequality derives from the curvilinear relation between income and health,13 with a decreasing benefit of individual health-related resources and material living standards.9,10 It is also suggested that psychologic mechanisms such as perceived inequality (ie, the making of social comparisons) contribute.14,15 Another hypothesis is that income inequality creates societies that are hazardous to people's health due to underinvestment in human capital16 and disruption of the grounds for solidarity and social cohesion, which diminish social capital.17 It is claimed that these influences have an impact exceeding the effect that personal income itself exerts on individual health.8
The health effect of income inequality weakens when moving from higher levels of aggregation (such as nations) to lower levels (such as neighborhoods),18,19 and even the direction of the effect might change.20 There are several reasons for such an effect of scale. If the underlying processes are primarily political, as has been suggested, the effects of income inequality will most likely be found when comparing states or nations. Nations that have great disparities within their borders tend to have less extensive redistributive systems and lower levels of human capital and social security investments. Most studies conducted outside the United States compare income inequality across smaller units such as regions, municipalities, and parishes.21,22 It is unclear, however, whether these units have enough political autonomy to influence the processes behind income inequality. Alternatively, income dispersion at the regional level may be seen merely as a result of the labor market's geographic structure.
Within the smallest units such as neighborhoods, the direction of the association is not evident. Income inequality within a neighborhood is rather a measure of socioeconomic heterogeneity, mirroring a mixture of residents with divergent social and economic characteristics. The residential composition of a neighborhood is partly a result of housing and zoning strategies but is also driven by segregation processes in a wider sense. Furthermore, it has been suggested that economic heterogeneity in urban communities can have beneficial effects. One hypothesis is that poor people benefit from sharing neighborhoods with more affluent families. A certain proportion of middle and upper class people in urban neighborhoods may be necessary to sustain basic institutions.23
Sweden has a relatively low level of income inequality, with a Gini coefficient (defined below) of 0.22 at the beginning of the 1990s and 0.25 at the end.24 However, there is considerable spatial socioeconomic segregation in the larger urban regions, mostly from concentrations of high income earners. Furthermore, the economic recession of the early 1990s dramatically increased unemployment rates and increased the geographic concentration of poverty.25
Pathways connecting socioeconomic circumstances in neighborhoods to the development of CHD have been suggested. There is a consistent and strong relation between individual socioeconomic position and risk of CHD.26,27 Because individual socioeconomic position influences the type of neighborhood a person chooses to, or has the ability to, live in, it is treated as a confounder in most studies. However, it has also been argued that socioeconomic context influences a person's chances of a favorable career28,29; accordingly, it is plausible that an individual's socioeconomic position partly mediates and partly confounds the relation between socioeconomic context and occurrence of disease. It has also been shown that the neighborhood socioeconomic context affects affluent and poor residents differentially.30
We focus on 2 social aspects of the residential environment and analyze their effects on incidence of myocardial infarction (MI). The first is level of economic resources, as measured by median income; the second is degree of socioeconomic homogeneity, as measured by income distribution. The level of economic resources is considered as a property of the neighborhood social environment; however, if substantial contrasts are present, it may also be a sign of residential segregation within an urban region. Socioeconomic heterogeneity in a neighborhood implies a mix of people of different social backgrounds and economic resources, whereas a high degree of socioeconomic homogeneity in neighborhoods is an expression of spatial social polarization. Figure 1 illustrates a conceptual model of different types of socioeconomic contexts, defined by combinations of levels of income and income dispersion in neighborhoods.
The research questions are illustrated in Figure 2, which is a modified version of a model developed by Blakely and Woodward.31 The first 3 research questions concern the presence of a contextual effect. After adjustment for individual social characteristics reflecting the social selection of people into different contexts, (1) Is residence in a low-income neighborhood associated with an increased incidence of MI? (2) Is residence in a neighborhood with a high degree of income homogeneity likewise associated with an increased incidence of MI? and (3) Do both of these associations remain after mutual adjustment? Fourth, we examine whether there is synergy between neighborhood income level and degree of income homogeneity with neighborhoods classified as outlined in Figure 1. Fifth, we analyze whether the individual economic situation makes people more vulnerable to the effect of a low-income context as an illustration of other important mechanisms. (6) The last figure illustrates another important mechanism in which established biomedical risk factors and health behaviors are involved as mediators of the contextual effect. We do not address this question here.
The Stockholm Heart Epidemiology Program (SHEEP) is a population-based case–control study of the causes of myocardial infarction.32 The study comprised all nonfatal (n=1,643) and fatal (n = 603) first events of MI among Swedish citizens age 45 to 70 years, resident in Stockholm County during 1992–1993 for men and 1992–1994 for women. Cases were identified from the coronary and intensive care units at emergency hospitals in Stockholm County, from the Hospital Discharge Register for the county, or from death certificates from the National Cause of Death Register maintained by Statistics Sweden. Standardized diagnostic criteria for MI were applied. Cases were included at time of disease onset. Simultaneously, 1 control per case was randomly selected from the corresponding study base after stratification for age, sex, and hospital catchment area (10 areas). All controls were initially checked for previous MI and were alive when recruited, regardless of the vital status of the corresponding case. More referents than cases were finally included; in most instances, the referent was already included when the case chose not to participate; and in some, the referent chose not to participate at first, and another one was sampled, although they both decided to participate in the end. All subjects received a postal questionnaire covering a large set of potential risk factors. A telephone interview minimized partial nonparticipation. Information on fatal cases was obtained from close relatives. Furthermore, an extensive amount of register-based information on income, wealth, and family structure was linked to the study subjects.
The SHEEP study encompassed 2,246 cases and 3,206 controls. The nonparticipation rate among cases was 28% for women and 19% for men, whereas the corresponding figures for controls were 30% and 25%. Response rates did not differ by age or catchment area.32 Due to insufficient address information, 7% of subjects were not included in this study. After excluding rural inhabitants and subjects in sparsely populated small urban areas (4%), 3,610 individuals were included—485 female and 1,062 male cases, and 697 female and 1,366 male controls.
“Neighborhood” refers to a small residential area originally used as a census area and defined according to homogeneity of buildings and land use. These areas are now continually updated by the Office of Regional Planning and Urban Transportation although no longer used for census purposes.33 SHEEP participants’ most recent address was GIS-coded and assigned a neighborhood code. At the time of the study, Stockholm County comprised 1,132 small areas (neighborhoods). Rural areas were excluded from the analyses, because the processes of economic and residential segregation differ between rural and urban areas (urban area defined as 200 or more inhabitants or less than 200 m between buildings). We also excluded neighborhoods with fewer than 10 income earners (mostly industrial sites). In total, the analyses were based on 873 neighborhoods with a median population of 1,073 (standard deviation = 1,163).
To characterize the economic context in neighborhoods, we used the work and mortality database, a register-based total population cohort with baseline in 1990. The contextual measures are based on all individuals age 18 years or over. Equivalent disposable household income (in 1990) was used as the income measure. This measure of income includes earnings and income from self-employment for both spouses, as well as all cash benefits that the family receives, and subtracts taxes paid. To compare income among individuals from households of differing size and structure, we adjusted income according to an equivalence scale. The scale used was based on the norms for social assistance at that time.34,35
To assess the level of economic resources in neighborhoods, we calculated medians of the equivalent disposable income. Furthermore, we calculated Gini coefficients of this income measure to define the neighborhood's degree of economic homogeneity. The coefficient for each area36 was calculated as:
Equation (Uncited)Image Tools
where Y = cumulative proportion of disposable income, X = cumulative proportion of the population, and k = number of individuals.
The coefficient ranges from zero to 1, with zero representing total homogeneity and 1 maximum heterogeneity. If the Gini coefficient is zero, every person possesses an equal share of the total disposable income in the neighborhood.
Individual Social Characteristics
Individual income is measured as equivalent disposable household income at the year of inclusion on the basis of information derived from the income tax registry. Family socioeconomic position is based on occupation classified according to a system developed by Statistics Sweden37; it is determined by the dominance method, which considers information on occupation from both spouses and uses the one regarded as dominant with respect to influence on the attitudes, behaviors, ideology, and consumption patterns of the household in general (developed by Erikson38). For SHEEP participants, we use information from the questionnaire on latest occupation before inclusion in the study; and for their spouses, we use information from the 1990 census. Information on educational level and family status is taken from the questionnaire.
We calculated odds ratios as estimators of incidence rate ratios (IRRs) and their associated 95% confidence intervals (CIs) to measure the effect of contextual exposures on incidence of MI. Specifically, we fit multilevel random intercept logistic regression models with individuals treated as first-level units nested within neighborhoods treated as second-level units. We used penalized quasilikelihood procedures with second-order Taylor-series approximation in the final models, as suggested by Goldstein and Rasbash.39 The presence of overdispersion was evaluated. All the fitted models met a binomial distribution requirement.
The sample of controls was stratified by age, sex, and hospital-catchment area. All analyses were adjusted for age (in 5-year age groups) and stratified by sex. Because hospital-catchment area and residential area characteristics are correlated, the stratified sampling of controls would have introduced confounding, biasing relative risks toward unity. Therefore, all models included control weights to eliminate the effect of sampling by catchment area.
We included indicators of individual social position that operate in the segregation process as confounders, ie, as determinants of the selection of individuals into specific types of neighborhoods. Examples are disposable income, family socioeconomic group, educational level, labor market position, and cohabitation.
To evaluate both the within-ecologic level interaction and the cross-level interaction, we used Rothman's model40 for the analysis of biologic interaction or synergism. The model assesses whether there are cases occurring only in the presence of joint exposures, or in other words, if there is a departure from additivity of effects. To quantify interaction, we used the synergy (S) index with 95% CI.41
Equation (Uncited)Image Tools
where IRR denotes the incidence rate ratios, using IRR00 (unexposed) as denominator. IRR10 and IRR01 are the incidence rate ratios for those with 1 exposure, and IRR11 is for those with both exposures. The index denotes synergy if its value exceeds 1.0 and antagonism if it is less than 1.0.
The multilevel analyses were carried out using MlwiN version 2.0 (Multilevel Models Project, Institute of Education, University of London), whereas SAS version 8.2 (SAS Institute, Cary, NC) was used for the other analyses.
The median income level in the neighborhoods was 86,924 Swedish Kroner, and ranged from 60,680 SEK in the first percentile to 110,653 SEK in the 99th percentile. Socioeconomic heterogeneity in neighborhoods, as measured by the Gini coefficient, ranged from 0.14 in the first percentile to 0.43 in the 99th with a median of 0.20. The correlation between the contextual exposures was −0.12.
All social characteristics are unevenly distributed across area categories (quartiles of neighborhood income level), with the highest quartile having higher rates of highly educated, high-income earners and individuals in nonmanual occupations, and also more couples and higher employment levels (Table 1).
Table 2 shows a graded increase in incidence rates of first acute MI across quartiles of income level in neighborhoods for both women and men. The incidence rate ratio for women living in low-income neighborhoods, compared with those living in high-income neighborhoods, is 2.6 (95% CI = 1.8–3.7). Adjustment for sociodemographic composition of the area reduces the excess risk by 44%; that is, 44% of the excess risk is accounted for by the fact that individuals from lower social strata are more likely to live in low-income neighborhoods. The incidence rate ratio for men living in the lowest income quartile is 2.0 (1.6–2.6); adjustment for individual social characteristics reduces the excess risk by 48%, mainly because of the impacts of individual income and family socioeconomic group. Further adjustment for area income homogeneity had little effect on the result; in other words, we found no within-level confounding from income heterogeneity by area.
Table 3 provides estimates of the effect of neighborhood income homogeneity on risk of MI. We used the most heterogeneous quartile of areas as the reference group. For men, the crude models suggest slight beneficial effects of income heterogeneity. However, this pattern weakens after adjustment for individual social position, and we found no consistent contextual effect of income dispersion for women or for men.
Because our conceptual model (Fig. 1) allows for possible interaction between contextual dimensions, we calculated incidence rate ratios by category of neighborhood income level and dispersion. Table 4 displays adjusted incidence rate ratios with IRR00 as denominator and also stratum-specific synergy estimates. We found a prevailing effect of income level, although with some differences by sex. Among women, the results suggest a weak negative effect of income heterogeneity in mid- and low-income neighborhoods. By contrast, among men, income homogeneity has a negative effect in low-income neighborhoods, which suggests interaction in the highly exposed stratum (S22).
Due to the weak and inconsistent effects of neighborhood income dispersion, we focused on mechanisms linking neighborhood income level to incidence of MI. Figure 3 graphically explores the cross-level interaction between individual disposable income and neighborhood income level. Table 5 presents results from cross-level interaction analyses in which both individual disposable income and neighborhood income level are treated as dichotomous exposures with the median as cutoff point. For women, the effect of combined exposure exceeds the additive effect, giving a synergy index of 2.7 (95% CI = 1.0–7.0) in the age-adjusted model. In other words, women living on a low disposable income are more vulnerable to the effect of a low-income context. This is also illustrated by Figure 3, which shows a steeper decrease in MI incidence for women in the 2 lowest income quartiles as neighborhood income level increases. For men, the effect is additive (as illustrated in both Table 5 and Fig. 3).
We found an increased incidence of MI in low-income neighborhoods that was not due to individual social characteristics. These results are consistent with other studies of MI incidence1 and of CHD morbidity and mortality4–6 based on individual and neighborhood data. The implication is that, in addition to the social gradient in CHD found when defining social position on the basis of individual social characteristics such as occupation, education, or income, the general level of socioeconomic resources in a neighborhood contributes to a context that further strengthens the effects of social stratification.
Socioeconomic heterogeneity within a neighborhood seems to have less effect on MI. Our hypothesis was that heterogeneity of income in neighborhoods, given the national level of income heterogeneity, reduces the risk of MI. The hypothesis was based on the longstanding claim in social planning that it is an advantage to society if people from different walks of life share the same neighborhood. This may well be true, but we did not find consistent empiric evidence for an associate reduction in risk of MI.
Despite the fact that the degree of socioeconomic heterogeneity is measured as the income dispersion within neighborhoods, we would like to underline that the results, in our opinion, have limited application to the current debate on income inequality and population health. That hypothesis was originally phrased on a higher aggregation level,42 and its merit should not be valued on a much lower aggregation level such as neighborhoods.43
We looked at the interaction between neighborhood income level and personal economic situation as a potential part of the mechanism, and concluded that women in low-income families may be more vulnerable to low-income contexts. Among men, the effect was additive, although the lowest income group showed a greater decrease in MI compared with the highest income group in relation to a rise in neighborhood income level. The pattern of increased vulnerability among low-income earners has also been observed by Diez Roux and colleagues. Although their results were not stratified by sex, they showed a more than additive effect of low personal income and socioeconomic deprivation on risk of CHD.1
Several potential mechanisms in the context-disease relation are expected to act through established risk factors. Neighborhood socioeconomic characteristics have been linked to the prevalence of cardiovascular risk factors.5,44 The common interpretation is that the contextual effect on CHD is partly due to the various health behaviors that lead to coronary heart disease; that is, there is a greater prevalence of detrimental health behaviors in deprived areas. The adjustment for individual social characteristics takes into account health behaviors and biomedical risk factors that are unevenly distributed across social strata. However, it is hypothesized that health behaviors and pathophysiological processes are shaped by the context in which they occur, which would imply that regardless of the individual socioeconomic position, factors such as smoking, physical inactivity, and obesity, as well as biomedical risk factors such as insulin resistance will be more common in low-income neighborhoods. A study45 of the trend in coronary artery disease risk in young adults shows that living in an adverse socioeconomic context is related to the development of insulin resistance and metabolic syndrome. There are also studies46 linking residential environment (in terms of accessibility, cost, and esthetic quality) to health behaviors such as physical activity and dietary patterns, although the evidence so far is limited.
Strengths and Weaknesses of the Study
Our study is population-based with a high participation rate. Participation did not vary by hospital catchment area.32 This makes the current study less sensitive to selection bias due to a differential response rate associated with area-level deprivation, which can be a considerable problem in contextual studies.47
Furthermore, both the validity of MI diagnosis and the reliability of case identification were high.48 The use of incident first event of MI is especially relevant when evaluating the etiologic implications of contextual exposures. In particular, it reduces the risk of bias from health-related selection into low-income contexts. Also, because case-fatality rates have been found to be related to neighborhood deprivation,49 the inclusion of both fatal and nonfatal cases reduces influences from factors such as availability of acute care or distance to a hospital.
Assessment of the contextual exposures is based on information from the total population, which results in high precision.
A limiting factor is lack of information on residential mobility and previous contextual exposures. We used address information at time of enrollment to define neighborhood of residence, and we did not consider any time lag or accumulation of contextual exposures in the analyses. Variations in residential duration for both cases and controls might imply nondifferential misclassification of the contextual exposures, which would bias the relative risk ratios toward unity. However, 76% of the controls in the study had lived at the same address for 10 years or more.
The use of administrative boundaries when estimating contextual exposures has been criticized. Nevertheless, we think that the neighborhoods as defined for our study capture the spatial differentiation of socioeconomic recourses reasonably well. The boundaries are natural and are homogeneous regarding building structure and land use. When, as is intended, we proceed with social area analyses, our conclusions will not be drawn from a particular area but from groups of areas with a specific type of context. That is, adjoining areas with similar contexts, whose inhabitants do not necessarily perceive there to be a border between them, will be analyzed as belonging to the same group. At a higher level of aggregation (parishes, for example), information from sometimes contrasting small residential areas is merged, which hides important contrasts between neighborhoods and causes nondifferential misclassification of exposures.
This study has extensive information on individual-level confounders, which makes it possible to reduce the bias from omitted (individual-level) variables and enables analyses of mechanistic issues. There is no universal agreement on how to specify the first-level model correctly. Our approach was to adjust for individual and family socioeconomic characteristics to make individuals from different contexts more comparable. Family socioeconomic position made the greatest contribution. By using a family-based measure of the socioeconomic position, we improved control of confounding, especially for women. The final combination of social characteristics broadly captures aspects of social stratification and indicates the opportunity to acquire housing in a desirable area. The further inclusion of social characteristics (such as country of birth and labor market position) did not substantially alter the results.
This approach is of course a simplification of a complex reality and must be interpreted with caution. Due to the constant interplay between individuals and their context, a clear distinction between what is confounded by individual social position and health behaviors and what is mediated through these factors is not possible in these type of analyses.50 To some extent, social surroundings shape people's opportunities, preferences, and actions, and to some degree people will form their own social environment. The possibility of making causal inference from the neighborhood studies so far conducted has recently been questioned.51 In line with Diez-Roux,52 we would argue that it is still productive to use observational data and that the neighborhood effects are not by definition endogenous to the compositional characteristics of neighborhoods. The main question addressed in our study is whether specific aspects of the local social environment have an impact on MI incidence. Answering this question can be regarded as a first step toward attaining the ultimate goal of understanding the causal role played by context. However, that goal is far from reached, and as Diez Roux states, “associations … on neighborhood health effects are what they are: measures of conditional associations under certain assumptions.”52(p.1959) We would argue that income level in the neighborhood serves as a proxy for a range of circumstances affecting people's daily life that would have been missed if only social circumstances measured at the individual level were considered. To obtain further insight into the social etiology of MI, it is of great interest to assess the contribution made by both sources.
In conclusion, this study confirms that the socioeconomic context of neighborhoods has an effect on cardiovascular outcomes after adjustment for individual socioeconomic situation.
We thank Michael Lundberg for valuable help with calculations of the Gini coefficients and Fredrik Mattson and Rickard Ljung for skillful SAS programming.
1. Diez-Roux AV, Merkin SS, Arnett D, et al. Neighborhood of residence and incidence of coronary heart disease. N Engl J Med
2. Kölegård Stjärne M, Diderichsen F, Reuterwall C, et al. Socioeconomic context in area of living and risk of myocardial infarction: results from Stockholm Heart Epidemiology Program (SHEEP). J Epidemiol Community Health
3. Kölegård Stjärne M, Ponce de Leon A, Hallqvist J. Contextual effects of social fragmentation and material deprivation on risk of myocardial infarction: results from the Stockholm Heart Epidemiology Program (SHEEP). Int J Epidemiol
4. Davey Smith G, Hart C, Watt G, et al. Individual social class, area-based deprivation, cardiovascular disease risk factors, and mortality: the Renfrew and Paisley study. J Epidemiol Community Health
5. Diez-Roux AV, Nieto JF, Muntaner C, et al. Neighborhood environments and coronary heart disease: a multilevel analysis. Am J Epidemiol
6. Waitzman JN, Smith KR. Phantom of the area: poverty-area residence and mortality in the United States. Am J Public Health
7. Davey Smith G, Whitley E, Dorling D, et al. Area based measures of social and economic circumstances: cause specific mortality patterns depend on the choice of index. J Epidemiol Community Health
8. Subramanian SV, Kawachi I. Income inequality and health: what have we learned so far? Epidemiol Rev
9. Lynch J, Smith GD, Harper S, et al. Is income inequality a determinant of population health? Part 1. A systematic review. Milbank Q
10. Lynch J, Smith GD, Harper S, et al. Is income inequality a determinant of population health? Part 2. US national and regional trends in income inequality and age- and cause-specific mortality. Milbank Q
11. Wagstaff A, van Doorslaer E. Income inequality and health: what does the literature tell us? Annu Rev Public Health.
12. Wilkinson R. Unhealthy Societies. The Afflictions of Inequality.
London: Routledge; 1996.
13. Gravelle H. How much of the relation between population mortality and unequal distribution of income is a statistical artefact? BMJ
14. Deaton A. Health, inequality and economic development. Journal of Economic Literature
15. Åberg Yngve M. Resources and Relative Deprivation. Analysing Mechanisms Behind Income, Inequality and Ill-Health
. Stockholm: Almqvist & Wiksell International; 2005.
16. Kaplan GA, Pamuk ER, Lynch JW, et al. Inequality in income and mortality in the United States: analysis of mortality and potential pathways. BMJ
17. Kawachi I, Kennedy BP. Health and social cohesion: why care about income inequality? BMJ
18. Soobader MJ, LeClere FB. Aggregation and the measurement of income inequality: effects on morbidity. Soc Sci Med
19. Blakely TA, Lochner K, Kawachi I. Metropolitan area income inequality and self-rated health—a multi-level study. Soc Sci Med
20. Hou F, Myles J. Neighborhood Inequality, Relative Deprivation and Self-Perceived Health Status.
Analytical studies branch research paper series no. 228. Ottawa, Canada; 2004.
21. Gerdtham UG, Johannesson M. Absolute income, relative income, income inequality and mortality. Journal of Human Resources
22. Osler M, Prescott E, Gronbaek M, Christensen U, Due P, Engholm G. Income inequality, individual income, and mortality in Danish adults: analysis of pooled data from two cohort studies. BMJ
23. Wilson WJ. The Truly Disadvantaged: The Inner City, the Underclass, and Public Policy.
Chicago: University of Chicago Press; 1987.
24. Fritzell J. Inkomstfördelningens trender under 1990-talet. In: Fritzell J, Palme J, eds. Välfärdens finansiering och fördelning.
Stockholm: SOU; 2001;57:131–189.
25. Andersson R. Etnisk och socioekonomisk segregation i Sverige 1990–1998. In: Fritzell J, ed. Välfärdens förutsättningar
. Arbetsmarknad, demografi och segregation.
Stockholm: SOU; 2000;37:223–267.
26. Hallqvist J, Lundberg M, Diderichsen F, et al. Socioeconomic differences in risk of myocardial infarction 1971–1994 in Sweden: time trends, relative risks and population attributable risks. Int J Epidemiol
27. Peltonen M, Rosen M, Lundberg V, et al. Social patterning of myocardial infarction and stroke in Sweden: incidence and survival. Am J Epidemiol
28. Robert S. Socioeconomic position and health: the independent contribution of community socioeconomic context. Annu Rev Sociol
29. Andersson R. Socio-spatial dynamics: ethnic divisions of mobility and housing in post Palme Sweden. Urban Studies
30. Stafford M, Marmot M. Neighborhood deprivation and health: does it affect us all equally? Int J Epidemiol
31. Blakely T, Woodward A. Ecological effects in multi-level studies. J Epidemiol Community Health
32. Reuterwall C, Hallqvist J, Ahlbom A, et al. Higher relative but lower absolute risks of myocardial infarction in women than in men for some major risk factors in the SHEEP study. J Intern Med
33. Regionplane och trafikkontoret Basområden, Områdesindelningen i Stockholms län
34. Statistics Sweden. Bakgrundsfakta till arbetsmarknads och utbildningsstatistiken.
Louise Örebro; 2001:2.
35. Fritzell J. Icke av marknaden allena: Inkomstfördelningen i Sverige.
Stockholm: Almqvist & Wiksell International; 1991.
36. Brown MC. Using Gini-style indices to evaluate the spatial patterns of health practitioners: theoretical considerations and an application based on Alberta data. Soc Sci Med
37. Statistics Sweden. Swedish Socioeconomic Classification. Reports on Statistical Co-ordination.
38. Erikson R. Social class of men, women and families. Sociology
39. Goldstein H, Rasbash J. Improved approximations for multilevel models with binary responses. J Royal Stat Soc
40. Rothman K, Greenland S. Modern Epidemiology
. Philadelphia: Lippincott-Raven; 1998.
41. Rothman K. The estimation of synergy or antagonism. Am J Epidemiol
42. Wilkinson RG. Income distribution and life expectancy. BMJ
43. Fritzell J. Commentary: PN form, comparability, and levels in the income and health relationship. Int J Epidemiol
. In press.
44. Hart C, Ecob R, Smith GD. People places and coronary heart disease risk factors: a multilevel analysis of the Scottish heart health study archive. Soc Sci Med
45. Diez Roux AV, Jacobs DR, Kiefe CI. Neighborhood characteristics and components of the insulin resistance syndrome in young adults: the Coronary Artery Risk Development In young Adults (CARDIA) study. Diabetes Care.
46. Diez Roux AV. Residential environments and cardiovascular risk. J Urban Health.
47. Leiland A, Finlayson A, Clark D, et al. Assessing the representativeness of health surveys [Abstract]. Eur J Public Health S
48. Hammar N, Nerbrand C, Ahlmark G, et al. Identification of cases of myocardial infarction: hospital discharge data and mortality data compared to myocardial infarction community registers. Int J Epidemiol
49. Macintyre K, Stewart S, Chalmers J. Relation between socioeconomic deprivation and death from a first myocardial infarction in Scotland: population based analysis. BMJ
50. Macintyre S, Ellaway A, Kawachi I, et al. Neighborhoods and Health
. New York: Oxford University Press; 2003.
51. Oakes JM. The (mis)estimation of neighbourhood effects: casual inference for a practicable social epidemiology. Soc Sci Med
52. Diez Roux AV. Estimating neighbourhood health effects: the challenges of casual inference in a complex world. Soc Sci Med.
© 2006 Lippincott Williams & Wilkins, Inc.