Solid evidence has established that air pollution adversely affects human health.1 In the general population, groups with lower socioeconomic status (SES) may simultaneously be more exposed to air pollution and more susceptible to its deleterious effects.2,3 Through these 2 mechanisms, air pollution appears more likely to affect deprived than privileged populations.2 Accurate estimation of the contribution of environmental exposures to social inequalities in health, especially the public health impact of air pollution, is currently a major research area3,4 and requires a better understanding of the interaction between SES and air pollution.2 Numerous ecologic studies of environmental equity (or environmental justice) have investigated this topic and assessed exposure and socioeconomic characteristics by using data collected on a geographic scale (eg, census tracts, census blocks, zip codes, counties, and metropolitan areas). Most conclude that groups with a low SES tend to experience more exposure to air pollutants and toxicants, due especially to the proximity of their homes to various pollution sources (including high-traffic roads, industrial facilities, and waste disposal sites).5–7 Nonetheless, some studies report no association,8,9 whereas others find inverse associations between SES and pollution, with the most privileged groups being more exposed.10
The lack of methodologic consensus on how to investigate environmental inequity makes these studies difficult to compare.11,12 Bowen11 recently reviewed the empirical research on environmental justice and concluded that published studies fail to consider important methodological issues about choice of geographic units, analytical methods, and exposure estimates and therefore produce results that are mostly unreliable and impossible to generalize. Another problem inherent in ecologic studies is spatial autocorrelation, which expresses the nonindependence of geographic observations.12 With very rare exceptions,13,14 studies of environmental equity do not consider this phenomenon, although it is essential to avoid violating the hypotheses that underlie the application of statistical models, and may thus lead to erroneous conclusions about associations.
In this study, conducted in the Strasbourg metropolitan area (Alsace region in northeastern France), we investigated the associations between exposure to nitrogen dioxide (NO2) and SES in residential neighborhoods. To assess the effect of spatial autocorrelation on these associations, we successively applied 2 types of regression models—one that takes this autocorrelation into account and another that does not. NO2 was selected for this study because it is known to be a good tracer of urban air pollution generated by traffic and because its spatial heterogeneity is greater than that of other air pollutants.15 It is also a pollution indicator for which exposure varies substantially among socioeconomic groups16–19 and for which short- and long-term associations with several respiratory and cardiovascular health outcomes have been reported.20–22 Although this work does not directly explore the relationship among health, SES, and environmental exposure, its epidemiologic perspective allows discussion of the extent to which bias in the measurement of associations may be due to failure to consider spatial autocorrelation. For this purpose, we used data from an ecologic epidemiologic study investigating the relationship between myocardial infarction and short-term exposure to air pollution, while taking into account SES, measured on a small-area scale in the Strasbourg metropolitan area.23
Study Area and Spatial Scale
The Strasbourg metropolitan area is an urban area of 28 municipalities (316 km2) with a population of about 450,000 inhabitants. The spatial scale used is the French census block, a submunicipal division designed by the French Census Bureau. It corresponds to a residential neighborhood with an average of 2000 inhabitants and is the smallest geographic unit in France for which socioeconomic and demographic information is available from the national census. The division of neighborhoods into census blocks takes into account the physical obstacles that may break up urban landscapes (important traffic arteries, bodies of water, green spaces, etc.) and aims to maximize their homogeneity in population size, socioeconomic characteristics, and land use and zoning. French census block is, in terms of population size, intermediate between US census tracts (about 4000 inhabitants) and US census block groups (about 1000 inhabitants). According to Krieger et al,24 this may be the most relevant spatial scale for measuring socioeconomic inequalities and assessing socioeconomic gradients in health or environment exposure.
The Strasbourg metropolitan area is subdivided into 190 census blocks. Sixteen blocks had very small populations (<250 inhabitants) and were excluded from the study (0.8% of the total population).
Socioeconomic Status Indicator
To characterize accurately the SES of census blocks, we used a socioeconomic deprivation index built for the Strasbourg metropolitan area at the census block level, which is described in detail elsewhere.25 Briefly, this index was constructed by a principal component analysis from a selection of 19 socioeconomic and demographic variables from the 1999 national census. These were variables that reflected multiple dimensions of deprivation: income, educational level, job, housing characteristics, family and households, and immigration status (Table 1). This index has proved its validity to demonstrate socioeconomic gradients in the incidence of myocardial infarction25 and asthma attacks.26
Hourly NO2 concentrations during year 2000 (averaged into annual concentrations) were modeled at the census level by the local air monitoring association, a partner of this project, using the ADMS-Urban dispersion model (“Atmospheric Dispersion Modeling System”).27 This model can estimate concentrations of an array of key atmospheric pollutants, including particles (PM10, PM2.5, TSP), nitrogen oxides (NOx), carbon monoxide (CO), ozone (O3), and sulfur dioxide (SO2) in urban environments at diverse temporal (hourly, daily, annually) and spatial scales (cities, neighborhoods, streets, industrial areas). In a recent review, Jerrett et al15 demonstrated the effectiveness and reliability of this type of model for assessing air quality in health effects assessment research.
To compute hourly concentrations, the ADMS-Urban model requires several types of data, including emission inventories, meteorologic data, and background pollution concentrations. These data must be as detailed and specific as possible—both spatially (geocoded) and temporally (hourly or about hourly). Table 2 details the main features of each type of data.
Emission inventories integrate key sources of pollution emitted in the Strasbourg metropolitan area (road traffic, residential, and industrial emissions). Emissions were calculated for several sources by a variety of methods (Table 2). The local air monitoring association collects these emission data on an annual basis; it then relates temporal profiles (hourly and monthly) to the corresponding activities and variables to compute hourly emissions. Hourly meteorologic data come from the local Météo France weather station. Monitoring stations in the Alsace network collect hourly background pollution data for NO2. Two successive ADMS-Urban calibrations in the urban area and on the outskirts of Strasbourg enabled us to select the background pollution conditions most representative of those observed in the Strasbourg metropolitan area.
Using these data, the air monitoring association modeled NO2 levels at the census block level for 3 points per census block. These points were extracted from a modeling grid with spatial resolution of 200 × 200 m and were selected as representative of the ambient air quality within each census block. The very high correlation (0.99) observed between mean NO2 levels per census block obtained from the selected points and mean NO2 levels from the modeling grid confirms the relevance of the selected points.
Finally, we assessed the temporal validity of the ADMS-Urban predictions for each census block by comparing the annual mean NO2 concentrations measured in 2000 to those estimated by the model at the same locations. Five monitoring sites were available for this comparison for the urban area and 3 for the outskirts. The very high correlations for both areas (0.89 for the urban area and 0.78 for the outskirts) show the model's precision in predicting local NO2 variations.
Spatial autocorrelation of the distributions of NO2 levels and of the deprivation index was estimated by calculating the Moran index (I).31 This coefficient varies between −1 for a negative spatial autocorrelation and +1 for a positive spatial autocorrelation (ie, when the neighboring census blocks have dissimilar or similar values for the variable considered).
First, associations between NO2 levels and the deprivation index were assessed with an ordinary least squares (OLS) regression model 1.
Where in our framework y corresponds to NO2 levels, x to the deprivation index, β to the regression coefficient associated with the deprivation index and ε to model residuals assumed to be independently and identically distributed (iid).
If the Moran index shows a significant spatial autocorrelation in the residuals, a spatial regression model must be applied to avoid violating the application's assumptions. Simultaneous autoregressive models (SAR) such as “SAR lag” and “SAR error” models are among the most commonly used.32 We selected the best SAR model specification with the Lagrange multiplier test statistics developed by Anselin et al,33,34 which led us to choose an SAR lag model 2.
The SAR lag model is similar to an OLS model in which a spatially lagged dependent variable Wy is introduced to control for spatial autocorrelation.32W corresponds to a spatial weight matrix that defined the notion of neighborhood between geographic units, and ρ to a spatial autoregressive parameter that estimates the scale of interactions between the observations of the dependent variable.
A logarithmic transformation was applied to the distribution of NO2 concentrations to ensure a normal distribution of the series. The nonlinear relation between NO2 levels and the deprivation index (Fig. 1) led us to consider the deprivation index as a categorical variable, which we divided into 5 deprivation categories (approximately equivalent to quintiles). The first category comprised the least deprived census blocks and the fifth the most deprived (Table 1 and Fig. 2A).
To assess the influence of the choice of the spatial weight matrix W, we performed a sensitivity analysis with different definitions of neighborhood (contiguity and distance) (eAppendix 1, https://links.lww.com/A721). Because changing the matrix did not affect the results, we present here the results with the simplest matrix, a first-order contiguity matrix. It considers that 2 geographic units i and j are connected if they directly share a border or a vertex. Finally, we compared the goodness-of-fit of different regression models with the Akaike information criterion.
All these analyses were performed with SAS version 9.1 (SAS Institute, Cary, NC) and GeoDa version 0.9.5-i. (Spatial Analysis Laboratory, University of Illinois, Urbana-Champaign, IL). Maps were drawn with the Geographic Information System ArcView version 9.1 (ESRI, Redlands, CA).
The spatial distribution of the Strasbourg metropolitan area deprivation index shows a socioeconomic gradient from the most deprived census blocks in the urban center and inner suburbs to the most privileged blocks on the outskirts (Fig. 2A). The most privileged (categories 1 and 2) are characterized by the best living conditions (lowest percentages in variables describing socioeconomic deprivation and highest percentages in variables showing material advantages), whereas the most deprived (category 5) is characterized by greater socioeconomic deprivation and little material and social resources (highest percentages in variables describing deprivation and lowest percentages in variables related to positive living conditions) (Table 1). As Figure 2A suggests, the spatial distribution of the deprivation index (I = 0.54 [95% CI = 0.45–0.64]) shows a strong positive spatial autocorrelation.
The spatial distribution of NO2 levels in the census blocks shows a pollution gradient from the most polluted census blocks located in the urban center and especially its immediate ring to the least polluted census blocks distributed in a second circle of greater circumference (Fig. 2B). The most polluted urban census blocks are those situated near the principal high-traffic arteries (highways and all main state roads) and the main industries. The distribution of NO2 levels also shows a strong positive spatial autocorrelation (I = 0.79 [0.69–0.88]).
Table 3 reports the descriptive statistics of NO2 concentrations according to deprivation categories. As seen consistently in Figures 2A and B, the most privileged census blocks have the lowest NO2 concentrations. On the other hand, the most deprived do not have the highest NO2 concentrations, which are found in the blocks of categories 3 and 4 (the midlevel deprivation areas).
Table 4 presents the results of the OLS and the SAR models. In the OLS model, pollutant levels are positively related to the deprivation index. As in Table 3, this relation is not monotonic and linear, from the privileged to the deprived areas. Again, we find the strongest associations with NO2 levels in categories 3 and 4 (β for category 3 = 0.26 [0.22–0.30]; β for category 4 = 0.24 [0.20–0.28]). The positive spatial autocorrelation detected in the residuals of the OLS model (I residual = 0.40 [0.31–0.50]) justifies the application of the SAR model. The relation between pollution levels and the index remains positive and nonlinear in the SAR model, despite much lower regression coefficients. Categories 3 and 4 are always most strongly associated with NO2 levels (β for category 3 = 0.08 [0.06–0.11]; β for category 4 = 0.08 [0.05–0.10]). Introducing the spatially lagged variable into the model allows to control for the presence of spatial autocorrelation (I residual = −0.04 [−0.13 to 0.06]). Finally, its inclusion in the SAR model clearly improves the quality of adjustment compared with the ordinary least squares model (Akaike information criterion reduced from −360.83 to −515.49).
This study demonstrates socioeconomic disparities in traffic-related air pollution exposure in the Strasbourg metropolitan area, at the census block level. To date, this project is the first study of environmental equity in France. The agreement of our findings with previous American,5,7,35 Canadian,13,14,36 and European16–19,37 studies suggests that environmental inequality is an universal phenomenon whereby the most disadvantaged bear a disproportionate burden of environmental hazards. This work adds, however, new findings by showing the need to verify the application's assumptions of regression models, notably the shape of the relationship, and by highlighting the potential impact of the spatial autocorrelation on the association estimates.
In the Strasbourg metropolitan area, the relation between NO2 levels and the deprivation index is nonlinear; the midlevel deprivation areas are most exposed to traffic-related air pollution. The central position of these neighborhoods in the Strasbourg metropolitan area, in direct contact with the principal highways that surround the center city, may partially explain this finding. In a 2001 report on the influence of traffic on air quality, the local air monitoring association reported that traffic density increased as it approached the very center of the city, with more than 100,000 vehicles a day on some highways.38 Moreover, 46% of the entire road network in the Strasbourg metropolitan area recorded NO2 levels exceeding World Health Organization guidelines (40 μg/m3 as an annual mean).39
Many hypotheses have been advanced to explain environmental inequalities based on SES. In the Unites States and Canada, this phenomenon is attributed mainly to racial and ethnic segregation,5,7,35 employment status,8,14 housing market dynamics,14,40 and income.5,7,14 In Europe, socioeconomic disparities—notably those related to social and racial segregation—are less marked than in the United States, and social and economic resources (income, material living conditions, housing) explain mainly the environmental inequalities.16–18 However, in our context, these variables alone do not explain the greater exposure of the midlevel deprivation blocks, showing that factors related to the local urban design also play a role.
In Strasbourg, 75% of residents of midlevel deprivation areas do not own their own home, and 90% live in multiple dwelling units. Their income allows them to live in the city center, which is quite expensive, but not to have their own home and live on the outskirts of the metropolitan area, where environmental conditions are more favorable. People in the privileged neighborhoods may live at a reasonable distance from the city center and buy single-family homes (more than 60% of the households in category 1 live in houses). Living at a greater distance from possible pollution sources, they are therefore potentially less exposed to NO2 and other traffic-related pollutants than those in midlevel deprivation areas. Nonetheless, distance from the economic center may require category 1 residents to travel further and longer than those from categories 3 and 4, and by means other than public transportation, which is available only on a limited basis in suburban areas; nearly 50% of households of these privileged areas have 2 or more cars. This situation may result in these residents experiencing relatively high exposure to traffic pollution while commuting to and from work and probably enhances their true exposure level, relative to ambient air values. Finally, the most deprived census blocks have a median monthly income below the poverty line (€ 722 in 2000)41 and must therefore live in subsidized housing—public or private (75% live in public housing and 21% in overcrowded homes)—which are dispersed around the inner ring surrounding central Strasbourg, and thus at a reasonable distance from the principal traffic arteries (Figs. 2A, B).
This surprising finding (in contrast with previous environmental equity literature that usually finds the lowest socioeconomic groups most exposed to environmental risks) may be explained by the local make-up of the social classes and housing in the Strasbourg metropolitan area. The same associations were confirmed for the other air pollutants we tested (PM10, CO, and O3), regardless of whether spatial autocorrelation was taken into account (results presented and discussed in eAppendix 2, https://links.lww.com/A722).
The interpretation of our findings must consider some weaknesses, notably exposure assessment. First, the use of a dispersion model such as the ADMS-Urban model may be limited by the extensive amount of input data that are required. Moreover, each type of data presents its own degree of uncertainty in spatial estimation, which is difficult to assess. Although this cannot be taken into account in the modeling, it may induce substantial exposure error. Uncertainty may come from data sources, estimation methods, or measurement tools. Models such as land-use regression, which are less complex to implement and can provide reliable estimates of traffic-related air pollution,15 could have been a relevant alternative. Briggs42 recently indicated, however, that these models were not interchangeable and their use should depend on the study context and the data availability.
Second, our study is constructed on the crude hypothesis that all subjects living in a given census block are exposed to the same pollution levels; it does not consider time-activity patterns, which might vary among individuals and according to SES. These issues may engender substantial bias in estimating personal exposure. This error will likely be greater for people who travel farther from their census block of residence (the privileged) than for those with lower mobility, who live near their workplace (residents of the midlevel and most deprived census blocks). Also, this study does not consider indoor air quality, another important source of error in estimating personal exposure because people spend an average of 60% of their time at home.43 Residents of deprived areas live in old dilapidated homes with poor ventilation (nearly 60% of these homes were built at least 40 years ago) that may favor the concentration of indoor pollutants. They may been more highly exposed to indoor sources than those in the privileged areas who can afford to invest in air conditioning, ventilation, and thermal isolation.44
Beyond these issues, however, the main objective of this work was to investigate environmental inequity while assessing the impact of spatial autocorrelation. The strength of the association between NO2 levels and deprivation declined when we controlled for spatial autocorrelation (regression coefficients decreased by 60%–67%). The relationship nonetheless remained substantive and nonlinear, and the quality of model adjustment clearly improved. The few studies of environmental equity that have taken this phenomenon into account also show a substantial modification of results compared with those that did not control it.13,14 Jerrett et al14 showed that the socioeconomic variables used to explain the variation of the levels of total suspended particles estimated in the city of Hamilton (Canada) differed according to the multivariate regression model used. The quality of adjustment of the SAR models in their study was lower than that in the OLS models. Another study by the same team reached a similar conclusion.13 These results thus confirm the need to consider spatial autocorrelation in ecologic studies of environmental equity to prevent erroneous conclusions about associations. The divergences observed between our study and those of Jerrett and coworkers13,14 demonstrate that the impact of considering spatial autocorrelation cannot be generalized and may depend on the structure of the spatial relationships between variables across different locations.
Studies on assessment social inequalities in health should also to take into account spatial autocorrelation.45 The associations between particulate air pollution and mortality in the American Cancer Society cohort, for instance, also depend on subjects' education levels46 and are modified when the spatial autocorrelation in the data is considered.47,48
To conclude, our study, in accord with others, sends a warning to epidemiologists who may want to use ecologic studies to explore the effects of air pollution on social inequalities in health. Ignoring the spatial autocorrelation in these factors may produce biased and uncertain estimates and lead to erroneous conclusions. Spatial epidemiology methods are available to explore this issue. Among them, the Bayesian approach is currently the object of increasing interest,49 especially for studying the contribution of air pollution to social inequalities in health.50
1. Brunekreef B, Holgate ST. Air pollution and health. Lancet
2. O'Neill MS, Jerrett M, Kawachi I, et al. Health, wealth, and air pollution: advancing theory and methods. Environ Health Perspect
3. Sexton K, Gong H, Bailar JC, et al. Air pollution health risks: do class and race matter? Toxicol Ind Health
4. Evans GW, Kantrowitz E. Socioeconomic status and health: the potential role of environmental risk exposure. Annu Rev Public Health
5. Gunier RB, Hertz A, Von BJ, Reynolds P. Traffic density in California: socioeconomic and ethnic differences among potentially exposed children. J Expo Anal Environ Epidemiol
6. Neumann CM, Forman DL, Rothlein JE. Hazard screening of chemical releases and environmental equity analysis of populations proximate to toxic release inventory facilities in Oregon. Environ Health Perspect
7. Perlin SA, Sexton K, Wong DW. An examination of race and poverty for populations living near industrial sources of air pollution. J Expo Anal Environ Epidemiol
8. Anderton DL, Anderson AB, Oakes JM, Fraser MR. Environmental equity: the demographics of dumping. Demography
9. Bowen W, Salling M, Haynes K, et al. Toward environmental justice: spatial equity in Ohio and Cleveland. Ann Assoc Am Geogr
10. Perlin S, Setzer R, Creason J, et al. Distribution of industrial air emissions by income and race in the United States: an approach using the Toxic Release Inventory. Environ Sci Technol
11. Bowen W. An analytical review of environmental justice research: what do we really know? Environ Manag
12. Haynes K, Lall S, Trice M. Spatial issues in environmental equity. Int J Environ Tech Manag
13. Buzzelli M, Jerrett M, Burnett R, et al. Spatiotemporal perspectives on air pollution and environmental justice in Hamilton, Canada, 1985–1996. Ann Assoc Am Geogr
14. Jerrett J, Burnett R, Karanoglou P, et al. A GIS-environmental justice analysis of particulate air pollution in Hamilton, Canada. Environ Plan A
15. Jerrett M, Arain A, Kanaroglou P, et al. A review and evaluation of intraurban air pollution exposure models. J Expo Anal Environ Epidemiol
16. Brainard J, Jones A, Bateman I, et al. Modelling environmental equity: access to air quality in Birmingham, Eng. Environ Plan A
17. Chaix B, Gustafsson S, Jerrett M, et al. Children's exposure to nitrogen dioxide in Sweden: investigating environmental injustice in an egalitarian country. J Epidemiol Community Health
18. Mitchell G, Dorling D. An environmental justice analysis of British air quality. Environ Plan A
19. Stroh E, Oudin A, Gustafsson S, et al. Are associations between socio-economic characteristics and exposure to air pollution a question of study area size? An example from Scania, Sweden. Int J Health Geogr
20. Nordling E, Berglind N, Melen E, et al. Traffic-related air pollution and childhood respiratory symptoms, function and allergies. Epidemiology
21. Rosenlund M, Picciotto S, Forastiere F, Stafoggia M, Perucci CA. Traffic-related air pollution in relation to incidence and prognosis of coronary heart disease. Epidemiology
22. Zanobetti A, Schwartz J. Air pollution and emergency admissions in Boston, MA. J Epidemiol Community Health
23. Havard S, Deguen S, Segala C, et al. Presentation of the PAISIM collaborative project: relationship between short-term exposures to air pollution, socioeconomic inequalities in health and the onset of myocardial infarction. Epidemiology
. 2006;17(suppl):S255; abstract.
24. Krieger N, Chen JT, Waterman PD, Soobader MJ, Subramanian SV, Carson R. Choosing area based socioeconomic measures to monitor social inequalities in low birth weight and childhood lead poisoning: the Public Health Disparities Geocoding Project (US). J Epidemiol Community Health
25. Havard S, Deguen S, Bodin J, Louis K, Laurent O, Bard D. A small-area index of socioeconomic deprivation to capture health inequalities in France. Soc Sci Med
26. Laurent O, Filleul L, Havard S, Deguen S, Declercq C, Bard D. Asthma attacks and deprivation: gradients in use of mobile emergency medical services. J Epidemiol Community Health
27. McHugh C, Carruthers D, Edmunds H. ADMS-Urban: an air quality management system for traffic, domestic and industrial pollution. Int J Environ Pollution
28. Kouridis C, Ntziachristos L, Samaras Z. COPERT III—Computer programme to calculate emission from road transport. Transport. User Manual (Version 2.1). Copenhagen: European Environment Agency. Technical Report No 50; 2000. Available at: http://reports.eea.europa.eu/Technical_report_No_50/en/tech50.pdf
. Accessed July 2, 2008.
29. UNECE/EMEP Task Force. EMEP/CORINAIR Emission Inventory Guidebook, 2007. Copenhagen: European Environment Agency. Technical Report No 16; 2007. Available at: http://reports.eea.europa.eu/EMEPCORINAIR5/en/page002.html
. Accessed July 2, 2008.
30. Regional Board for Research and Environment. Ministère de l'Economie, de l'Industrie et de l'Emploi; 2008. Available at: http://www.alsace.drire.gouv.fr/
. Accessed July 2, 2008.
31. Cliff A, Ord J. Spatial Processes: Models and Applications
. London: Pion Limited; 1981.
32. Anselin L. Spatial Econometrics: Methods and Models
. Dordrecht, The Netherlands: Kluwer Academic Publishers; 1988.
33. Anselin L. Lagrange Multiplier test diagnostics for spatial dependence and spatial heterogeneity. Geogr Anal
34. Anselin L, Bera A, Florax R, et al. Simple diagnostic tests for spatial dependence. Reg Sci Urban Econ
35. Morello-Frosch R, Jesdale BM. Separate and unequal: residential segregation and estimated cancer risks associated with ambient air toxics in U. S. metropolitan areas. Environ Health Perspect
36. Finkelstein MM, Jerrett M, Sears MR. Environmental inequality and circulatory disease mortality gradients. J Epidemiol Community Health
37. Wheeler BW, Ben-Shlomo Y. Environmental equity, air quality, socioeconomic status, and respiratory health: a linkage analysis of routine data from the Health Survey for England. J Epidemiol Community Health
38. Association pour la Surveillance et l'Etude de la Pollution Atmosphérique en Alsace. Estimation de la qualité de l'air en proximité automobile sur la Communauté Urbaine de Strasbourg. Schilthigheim: ASPA 01102501-I-D; 2001. Available at: http://w3.atmo-alsace.net/
. Accessed October 15, 2007.
39. World Health Organization. Air Quality Guidelines for Europe. 2nd ed. Copenhagen, Denmark: Who Regional Office for Europe; 2000. Available at: http://www.euro.who.int/document/e71922.pdf
. Accessed March 26, 2008.
40. Been V, Gupta G. What's fairness go to do with it? Environmental justice and the siting of locally undesirable land uses. Cornell Law Rev
41. National Institute of the Statistic and the Economic Studies. La France en fait et chiffres; 2007. Available at: http://www.insee.fr/
. Accessed March 26, 2008.
42. Briggs D. The role of GIS: coping with space (and time) in air pollution exposure assessment. J Toxicol Environ Health A
43. ExpoFacts Sourcebook. The Expolis Study. 2008. Available at: http://cem.jrc.it/expofacts/
. Accessed March 26, 2008.
44. Janssen NA, Schwartz J, Zanobetti A, Suh HH. Air conditioning and source-specific particles as modifiers of the effect of PM(10) on hospital admissions for heart and lung disease. Environ Health Perspect
45. Lorant V, Thomas I, Deliège D, Tonglet R. Deprivation and mortality: the implications of spatial autocorrelation for health resources allocation. Soc Sci Med
46. Krewski D, Burnett R, Goldberg MS, et al. Reanalysis of the Harvard Six Cities Study and the American Cancer Society Study of Particulate Air Pollution and Mortality
. Cambridge, MA: Health Effects Institute; 2000.
47. Jerrett M, Burnett RT, Goldberg MS, et al. Spatial analysis for environmental health research: concepts, methods, and examples. J Toxicol Environ Health A
48. Krewski D, Burnett R, Jerrett M, et al. Mortality and long-term exposure to ambient air pollution: ongoing analyses based on the American Cancer Society cohort. J Toxicol Environ Health A
49. Graham P. Intelligent smoothing using hierarchical Bayesian models. Epidemiology
50. Naess O, Piro FN, Nafstad P, Smith GD, Leyland AH. Air pollution, social deprivation, and mortality: a multilevel cohort study. Epidemiology