Uncertainties about the interpretation of patterns of self-reported symptoms among communities exposed to polluting industry have been recognized for some time. 1,2 The underlying difficulty is that there are two alternative explanations for an observed association between exposure and symptoms. Associations may be found as a result of a true biological effect; exposure to a hazard leads to more symptoms. Alternatively, increased symptoms may occur because the population in question is aware of the hazard; this awareness may lead to concern about the possibility of illness and thus predispose individuals to reporting more ill health. We describe this sequence of events as “awareness bias,” which we consider to be the propensity to report more illness and symptoms as a result of proximity to a potential hazard, in the absence of a measurable biological effect. Awareness bias therefore gives rise to a problem of interpretation, which necessitates disentangling these two kinds of explanation. As is the case with recall bias, another form of differential reporting, 3–6 arguments for awareness bias are so intuitively convincing that the need for data is often forgotten.
The purpose of this study is to explore, with empirical data, the arguments for and against awareness bias as an explanation for environmental associations. As it would be extremely difficult to gauge subjects’ level of awareness about a hazard directly, we relied on proximity to the source as a proxy for awareness (proximity to the source is itself a surrogate for proximity to emissions). Although it is unlikely that people who live in the vicinity of heavy industry are unaware of it, or its emissions, it is less clear whether they are worried by them. Our indicator of concern was responses to questions about stress and anxiety. We examined the relation among proximity to industry, the presence or absence of “worry” about pollution, and symptom reporting, using two studies carried out in Northeast England. 7–9 Both were undertaken amidst huge controversies about the likelihood of health effects from local heavy industry. At the time of each study, no link between exposure and ill health had been proven. What made a comparison between them interesting with regard to awareness bias was that self-reported data indicating a link between exposure to emissions and ill health emerged from one study (the Monkton study) but not the other (the Teesside study). In an editorial about the Monkton study, Malmberg 10 concluded that “overall, the arguments for refuting awareness bias are not entirely convincing” and suggested that the next step was to “devise epidemiologic methods to study subtle health effects that remove any uncertainty about awareness bias.”
The specific questions we examined in the light of Malmberg’s comment were as follows: (1) Are people living close to industry more likely to worry about pollution? (2) Are “worriers” different from “nonworriers” in terms of social characteristics and health-related behavior? (3) Is there a relation among proximity to industry, worry, and health status?
This study is therefore an exploration of the hypothesis that proximity to industry leads to increased reporting of illness through its association with worry (without any “true” biological effect) and addresses the challenge implicit in Malmberg’s comment cited above. 10
The Monkton Coking Works and Teesside Environmental Epidemiology Studies
The Monkton study centered on Tyneside’s last coking works. 7 It arose from community concerns about the potential health effects of coking pollution on nearby residents. We found overwhelming concern among those living closest to the works about pollution and its potential health affects. 11 No discernible impact from exposure to pollution was found on mortality, cancer, birth weight, stillbirth, and lung function. 7 Nevertheless, an excess of some respiratory problems and other illnesses among those living closest to the works, based not only on self-reports but also on general practice consultation data, was best explained, we argued, by exposure to emissions.
The Teesside study centered on long-standing concerns about poor health and pollution from one of Western Europe’s biggest steel and petrochemical complexes. The research took place across this conurbation of more than 400,000 people and compared a number of neighborhoods at varying distances from industry. It developed less from concerns of residents but more from family doctors’ concerns about the health problems of their patients. 12 Studies have shown, however, that a majority of Teesside residents felt that industrial pollution was a problem, 13 particularly for those living close to industry. 14,15 We also know that many of the poorer areas of Teesside experienced particularly high levels of premature mortality, in comparison with equally poor parts of a town 25 km north. 16 In contrast to the Monkton study, the Teesside study provided evidence of mortality being associated with exposure to pollution. Lung cancer in women living closest to industry on Teesside was found to be consistently higher than in women living farther away, 17 and mortality from respiratory disease in men and women was higher among those living nearer to industry. 9 Nevertheless, no evidence of excess self-reported respiratory ill health or general practitioner consultations among the communities living closest to industry was found. 9
Method of Original Studies
A self-completion postal questionnaire was sent to a stratified random sample of residents in each of the study areas. 7–9 The Monkton study respondents lived in one of three areas: an inner area, closest to the works; an outer area, adjoining, but at a greater distance from the works; and a control area, 6–10 km from the works (see Figure 1). In Teesside, residents were grouped into three zones, A, B, and C, varying in proximity to industry (near, intermediate, and farther, respectively (see Figure 2), with zone C being the referent group. Respondents in the different study areas were closely matched on socioeconomic, occupational, and health-related behavior characteristics. 7–9
;)
FIGURE 1: Areas in the Monkton postal questionnaire survey.
FIGURE 2: Areas in the Teesside postal questionnaire survey.
A number of questions were asked about illnesses and symptoms; demographic, socioeconomic, and occupational characteristics; and health-related behaviors. The response rate in the Monkton study was 68%, and that in the Teesside study was 61%. The questionnaires in the two studies were similar. For the purposes of this study, we used only questions that were identical in the two studies.
The hypothesis tested in both studies was whether there was a relation between proximity to industry and ill health. If such a relation were to be found, the direction of an ill health gradient would be, for Monkton, inner > outer > control, and for Teesside, zone A > zone B > zone C.
Analysis for This Study
We divided respondents into three worry groups on the basis of replies to the question, “Have any of these things been a cause of worry or stress for you in the last year?” The subsequent list of 17 items included industrial air pollution, housing problems, money problems, unemployment, and diet and smoking habits. Respondents were divided into the following groups: those who recorded no worries; those whose worries excluded industry-related categories; and those whose worries included one or more industry-related categories (see Table 1). Information was available on age, sex, socioeconomic deprivation, damp housing (for example, mold on walls), smoking, alcohol consumption, and exercise levels as potential confounders associated both with the “worry” category and self-reported health.
Table 1: Percentages and Numbers of Subjects by Worry Category and Area a. Monkton Study (Surveyed in 1990)
Self-reported illnesses and symptoms were chosen for further analysis if they fulfilled the following three criteria: (1) that the prevalence followed the pattern hypothesized at the outset of the study, ie, higher in the area closest to industry and lowest in the area furthest away; (2) that there was identical wording of the questions in both the Monkton and Teesside studies; and (3) that the illness or symptom could be plausibly associated with either allergic or other acute effects of air pollution. 18,19 A number of illnesses and symptoms fell into this category: allergies, hay fever, headache, sinus illness, summer cough, and summer phlegm. These were cross-tabulated by area and worry group. The purpose of this analysis in the Monkton dataset was to explore what happened to the “ill health gradient” demonstrated for the whole sample 7 when further analyzed by worry categories.
In the Teesside dataset, the purpose was to see whether such gradients that were not apparent in the overall analysis 9 appeared in further analysis. We also included five illnesses/symptoms that could not plausibly be associated with the effects of air pollution. These were arthritis, back trouble, hemorrhoids, indigestion, and painful joints.
Finally, we used multiple logistic regression to examine the effects of area of residence (the proxy for proximity to industry) and worry on the illnesses and symptoms while allowing for the effects of other variables likely to influence illness or symptom prevalence. All of the variables were entered into each model in one step. Results for area and worry are reported as odds ratios with 95% confidence intervals. Separate odds ratios are reported for inner vs outer areas and for inner vs control areas (A vs B and A vs C, respectively, for the Teesside study). Similarly, separate odds ratios are given for “non-pollution worriers”vs “nonworriers” and for “pollution worriers”vs “nonworriers.” The other variables in the model were sex, age (in four categories, 16–24, 25–44, 45–64, and 65+ years), smoking (never-smoker, ex-smoker, current smoker), alcohol consumption (never, not now, less than once a week, once or twice a week, most days), deprivation (on a scale of 0–4 indicating how many of the following factors were present: unemployment, no car, non-owner-occupied household, overcrowding), damp housing (absent or present), and exercise (never or rarely, once or twice a month, once or twice a week, most days).
Results
Table 1a shows that in the Monkton study those living closest to the coking works were much more likely to be concerned about industry-related issues and less concerned about other issues than those in the outer and control areas. The differences between the areas in Teesside were less pronounced (Table 1b), but adults living closest to industry were more likely to report industry-related worries than residents in other areas.
Table 2 shows how a number of factors with an influence on health varied by “worry group.” In Monkton (Table 2a), those in the “no-worry” group had a mean age similar to that of the other groups but were less likely to smoke cigarettes or to live in damp accommodation and were slightly less likely to take alcohol regularly. Among the Teesside group (Table 2b), those in the no-worry group had a higher mean age, experienced the least deprivation and damp housing, and were less likely to smoke than the other groups.
Table 2: Selected Characteristics of Respondents a. Monkton Study
We further analyzed these health-related variables by worry group and proximity to industry and essentially found the same pattern as shown in Table 2 irrespective of how close to industry respondents lived. For example, the lowest proportion of smokers was consistently in the no-worry group for the inner, outer, and control groups of the Monkton study and zones A, B, and C in the Teesside study; the highest proportion of respondents living in damp accommodation was in the industry-related worry group for the inner, outer, and control groups of the Monkton study and zones A, B, and C in the Teesside study.
Table 3a introduces evidence on health status and shows that illness prevalence rates in Monkton respondents overall were consistently highest in the group that recorded industry-related worries and lowest in the group that recorded no worries, irrespective of whether or not the illness was plausibly associated with air pollution. The pattern within each of the worry categories differed, however. For allergies, sinus illness, summer cough, and summer phlegm, among nonworriers, the highest prevalence was found in the inner area, with an illness gradient in the predicted direction (inner > outer > control) (albeit with very small numbers). For hay fever and headache, the highest prevalence was also found in the inner area. For arthritis, back trouble, hemorrhoids, indigestion, and painful joints, no gradient was apparent among nonworriers. For these five, with the exception of hemorrhoids, the highest prevalence was in the control area.
Table 3A: Illness and Symptom Prevalence Rates by Worry/Nonworry Category: Monkton Study
Table 3B: Illness and Symptom Prevalence Rates by Worry/Non-Worry Category, Teesside Study
Prevalence rates for those with non-industry worries were compared between those for nonworriers and those for industry-related worriers, and no gradient was apparent. Among those with industry-related worries, with the exception of summer cough, summer phlegm, and indigestion, prevalence rates were not highest in the inner area and there was no gradient in the hypothesized direction. For the Teesside sample, there was no noteworthy ill health gradient among illnesses plausibly related to air pollution (Table 3b). As with the Monkton study, the highest prevalence rates were observed for those with industry-related worries where a slight gradient was apparent for summer cough, summer phlegm, and back trouble, and there was a distinct gradient for indigestion. Again, as for Monkton, prevalence rates for those with non-industry worries lay between the no-worry and industry-related worry groups, and no gradient was apparent. Prevalence rates were lowest for those with no worries in each of the zones A, B, and C. Gradients (A > B > C) were observed in this group for allergies, arthritis, and painful joints.
Tables 4a and 4b show the results of the multiple logistic regression for the effects of area (proximity to industry) and worry on our set of illnesses and symptoms. The model allows for the effects of other variables likely to influence illness or symptom prevalence. In the Monkton study (Table 4a), when worry was excluded from the model (column A), the odds ratios for the area comparison inner area vs control were greater than 1, except for back trouble and hemorrhoids. This finding indicates that those living closest to industry were more likely to report these illnesses and symptoms when all of the other factors were taken into account. The picture is altered when worry was included in the model (Table 4a, column B). The odds ratios for area were then very close to or below 1.
Table 4A: Logistic Regression* Analysis of Selected Illnesses; Excluding Worry (A) and Including Worry (B) and Showing Effects of Proximity to Industry: a. Monkton Study
Table 4B: Logistic Regression* Analysis of Selected Illnesses; Excluding Worry (A) and Including Worry (B) and Showing Effects of Proximity to Industry: Teesside Study
Similarly, for worries unrelated to industrial pollution, the odds ratios were close to 1, suggesting that such worries did not affect illness prevalence. For worries related to industrial pollution, however, odds ratios exceeded 1.50, indicating that those who worried about pollution were more likely to report these illnesses compared with nonworriers. This was the case for all illnesses.
In Teesside, as shown in Table 4b column A, when worry was excluded from the model, the odds ratios for area were close to 1. With worry in the model, odds ratios for worries excluding industrial air pollution were also close to 1, but odds ratios for worries including industrial air pollution were greater than 1.50 for all illnesses except arthritis and hemorrhoids.
Discussion
Disentangling the relation among proximity to industry, worry, and self-reported illness is complex. Our findings do not provide a single definitive answer to this problem. Our findings were also subject to the usual limitations of postal questionnaire surveys, particularly the lack of information from nonresponders. Nevertheless, comparison between self-reported socioeconomic and 1991 census data indicated that our samples were, in this respect, representative of the populations from which they were drawn. 7–9,17 Our clearest finding was that worry was a better predictor of reported illness than proximity to industry. Differences appeared, however, when we stratified for worry category; worries about industry had a stronger association with illness than nonindustry worries.
Most straightforward and unsurprising is the answer to our first question, whether people living close to industry were more likely to worry about pollution. The extent of industry-related worries differed considerably depending on proximity to industry: those living closer reported more industry-related worries, and those living at a distance reported more non-industry-related worries or, indeed, no worries, in line with previous reports. 11 It also appeared that worrying about industrial pollution was associated with worrying less about other factors, for example, smoking and alcohol consumption. 11 This finding, too, is consistent with other studies. 20
Our second question concerned possible differences between worriers and nonworriers on a number of social and health-related characteristics. The general pattern suggested that the no-worry group experienced fewer of the factors related to ill health, such as smoking, alcohol consumption, and damp housing, irrespective of proximity to industry.
These observations have clear implications for research on the impact of industrial pollution on the health of people living nearby. Industries invoke worry among those who live nearby, and those who worry most are more likely to experience greater deprivation and to be smokers than those who do not.
Our third question concerned the relation between proximity to industry, worry and self-reported illness, and the role of awareness bias. We found a consistent association between worry and ill health. For all illnesses and symptoms examined, the lowest prevalence rates were among the group with no worries, higher prevalence rates were found among those with non-industry worries, and prevalence rates were higher again among those with industry worries (Tables 3a and 3b). This result was found in both Monkton and Teesside and could not be explained by a higher mean age and more men in the no-worry group (Tables 2a and 2b) or proximity to pollution sources. Worry, particularly industry worries, had a particular influence on self-reported illnesses and symptoms, both related and unrelated to the effects of air pollution. Worriers reported more illness irrespective of where they lived; nonworriers only reported more ill health if they lived near industry (in Monkton, when the illness was plausibly related to air pollution).
In the Monkton study, the overall association between proximity of residence and ill health was observed most clearly in the no-worry group for those illnesses and symptoms that were plausibly associated with air pollution. Therefore, among nonworriers in Monkton, but not in Teesside, for a number of specific illnesses and symptoms, a gradient of ill health was observed. This analysis accords with the recommendation of Ozonoff et al21 that studying those who do not believe that an exposure has made them ill tests the awareness bias hypothesis. The overall association persisted for most of the symptoms plausibly associated with air pollution when we adjusted for a wide range of potential confounding factors (Table 4a, column A) but disappeared when worry was included in the model (Table 4a, column B). In contrast, no association was found between area and illness in the Teesside study (Table 4b, column A), but an association was found between illness and worry about industrial pollution (Table 4b, column B). This finding implies that worry about industrial pollution is a consistent predictor of self-reported ill health irrespective of whether the illness is plausibly associated with air pollution or not.
Our analysis shows that worry has a greater influence on self-reported ill health than residential proximity to industry. This finding is in line with work that indicates that judgments of physical health are influenced by states such as anxiety or depression and vice versa. 22 Nevertheless, it does not help us to decide whether or not people are reporting more ill health because of their awareness of the hazard or whether living close to industry has led to biological illness directly (exposure to air pollution) or indirectly (anxiety over the hazard leading to ill health).
The strongest evidence we have to counter the awareness bias hypothesis as an explanation for the health gradients observed in the Monkton study is the absence of a gradient among those who worried about industry. This finding was coupled with the persistence of a gradient among the no-worry group for the illnesses and symptoms plausibly related to air pollution, but not among the other illnesses studied. There were small numbers in this group, but the association persisted when adjusting for confounders (Table 4a, column A).
The comparison between the Monkton and Teesside studies is important, particularly because different conclusions were reached about the association between proximity to industry and self-reported morbidity. The analysis presented here, however, showed few differences between the two studies in the relations between worry and ill health. One difference was the much smaller proportion of respondents in Teesside living close to industry who reported industry-related worries compared with the Monkton study (37%vs 73%). A number of circumstantial factors may account for this difference, including the following: Teesside’s longer historical links between its industries and residents; greater employment ties; a tendency for health concerns on Teesside to be raised by health professionals; and the epidemiologic study instigated predominantly by concerns of local government, health officials, and researchers. 12 Conversely, the Monkton study arose almost entirely because of local concerns, there was no strong employment link between the industry and residents, and there was a history of antagonism between local residents and the industry. 11 Crucially, public involvement in the controversy about the health impact of industrial pollution was much greater in Monkton. Nonetheless, the issue on Teesside had been given a high profile in the media. As we have seen, in both Teesside and Monkton, worry about industrial pollution was a better predictor of self-reported illness than proximity to industry. Nevertheless, the absence of a clear and consistent gradient among the industry-worry groups in either study argues against awareness bias.
In both studies, irrespective of residential proximity to industry, people who worried more reported more illness. It does not appear that proximity alone (our proxy for awareness) leads to more self-reported illness, a finding that indicates that the gradients found in the Monkton study 7 are not simply a consequence of awareness bias.
Our interpretation finds support from some other environmental epidemiologic studies. 21,23,24 Roht et al24 found that those expressing concern over a hazard were more likely to report symptoms and concluded, “the direction of causation is unclear, as a respondent’s symptoms may have led to the opinion that the waste disposal site affected the environment rather than the person’s opinion leading to the reporting of symptoms” (p 430). Ozonoff et al21 found an association between people who believed that the air or water made them ill and higher symptom prevalence. They concluded, “While this could be interpreted as strong evidence of recall bias, a true biological effect could also have brought respondents to the conclusion that the air and water was [sic] responsible for their feelings of ill health” (p 593). They concluded that recall bias alone did not account for their findings.
One problem with an approach like that of Ozonoff et al, 21 which excludes data from those who worry about the hazard, is that it potentially omits a high proportion of the sample: 73% and 37% in Monkton and Teesside, respectively. We do not, therefore, advocate the inclusion only of those who live close to industry, and are not worried by it, despite the fact that the pattern of ill health among those who are not concerned about the hazard provides useful information. 21,23,24
Acknowledging that living close to a source of pollution invokes stress 25,26 leaves exposed communities open to the charge of manifesting their anxiety as physical problems, or somatizing. 27 The limited evidence suggests that exposed and control populations exhibit similar levels of, for example, hypochondriasis 24 and somatizing. 28 Yet our studies show a huge disparity in a more direct measure of awareness, anxiety about industry. Collecting data on psychological traits in study and control populations, as advocated by Dales et al, 1 invites the danger that self-reported physical problems could become attributed to psychological traits within the study population when they are truly a manifestation of exposure to pollution. Even if self-reported health problems were attributed solely to somatization, an important problem remains if this is an indirect consequence of residential proximity to industry. Furthermore, from a sociological perspective, the knowledge that subjects have would be regarded simply as a fact of life and a precondition for successful research. 11,12 In emphasizing this point, we do not advocate taking personal testimony unquestioningly at face value, but wish to stress that the central issue is one of interpretation. Our solution has been to use several datasets, for example, self-reports, doctor’s records, routine mortality and morbidity data, and exposure measurements. 7,9 If the aim is to remove any uncertainty about awareness bias, self-reported data should not be used alone.
How generalizable are these findings from Northeast England? Although environmental disputes involving communities and industries each have their own characteristics, we have found, like others, 29,30 that environmental health issues can generate considerable anxiety. We would suggest that it is important to establish the degree of concern. The particular methods used need to take account of each specific context, especially social and cultural factors, and we would not advocate wholesale replication of the questions we used. Nevertheless, concerns about and proximity to the hazard need to be considered when interpreting self-reported data.
Acknowledgments
We thank Susanne Young and Carole Frazer for secretarial assistance; Ann Rooke for map drawing; and Richard Edwards, Denise Howel, Tom Webster, and two anonymous referees for valuable comments.
References
1. Dales R, Spitzer W, Suissa S, Schecter MT, Tousignant P, Steinmetz N. Respiratory health of a population living downwind from natural gas refineries. Am Rev Respir Dis 1989; 139:595–600.
2. Wagner R. What symptom and disease prevalence studies can and cannot do. Epidemiology 1998; 9:S96.
3. Lippman A, Mackenzie SG. What is “recall bias” and does it exist? In: M. Marois, ed. Progress in Clinical and Biological Research Vol 163. Prevention of Physical and Mental Congenital Defects: Part C: Basic and Medical Science, Education, and Future Strategies. New York: Alan R. Liss, 1985.
4. Raphael K. Recall bias: a proposal for assessment and control. Int J Epidemiol 1987; 16:167–170.
5. Coughlin SS. Recall bias in epidemiologic studies. J Clin Epidemiol 1990; 43:87–91.
6. Raphael KG, Cloitre M. Does mood-congruence or causal search govern recall bias? A test of life event recall. J Clin Epidemiol 1994; 47:555–564.
7. Bhopal RS, Phillimore P, Moffatt S, Foy C. Is living near a coking works harmful to health? J Epidemiol Community Health 1994; 48:237–247.
8. Teesside Environmental Epidemiology Study Group. Health, illness and the environment in Teesside and Sunderland. Newcastle upon Tyne: University of Newcastle upon Tyne; 1995.
9. Bhopal R, Moffatt S, Phillimore P, Pless-Mulloli T, Foy C, Dunn C, Tate J. Does living close to a constellation of industries impair health? Occup Environ Med 1998; 55:812–822.
10. Malmberg P. Coking to death? Lancet 1994; 344:632.
11. Moffatt S, Phillimore P, Bhopal R, Foy C. “If this is what it’s doing to our washing, what is it doing to our lungs?”
Industrial pollution and public understanding in North-East England. Soc Sci Med 1995; 41:883–891.
12. Phillimore P, Moffatt S. Discounted knowledge-local experience, environmental pollution and health. In: Popay J, Williams G, eds. Researching the People’s Health. London: Routledge; 1994; 134–153.
13. Gavin HF. Perceptions of Air Quality in Middlesbrough: Analysis of a Survey by Middlesbrough Borough Council. Teesside: University of Teesside, 1996.
14. Bush J, Moffatt S, Dunn C. Perceptions of risks to health from industrial air pollution: a high priority? In: Sjoberg B-MD, ed. Paper presented at Annual Meeting of the Society for Risk Analysis–Europe, 1997. Stockholm, 1997;284–287.
15. Moffatt S, Bush J, Prince H, Dunn C, Howel D. Community views on ill-health, air quality and “expert” information: the effects of proximity to heavy industry and poverty. Epidemiology 1998; 9:S143.
16. Phillimore PR, Morris D. Discrepant legacies: premature mortality in two industrial towns. Soc Sci Med 1991; 33:139–152.
17. Pless-Mulloli T, Phillimore P, Moffatt S, Bhopal R, Foy C, Dunn C, Tate J. Lung cancer, proximity to industry and poverty in Northeast England. Environ Health Perspect 1998; 106:189–196.
18. Primatesta P, Bost L, Dond W. Respiratory symptoms and atopic conditions. ch. 2. In: Health Survey for England. London: HMSO, 1996; 10–20.
19. Schwartz J, Dockery DW, Neas LM, Wypij D, Ware JH, Spengler JD, Koutrakis P, Speizer FE, Ferris BG Jr. Acute effects of summer air pollution on respiratory symptom reporting in children. Am J Respir Crit Care Med 1994; 150:1234–1242.
20. The Royal Society. Risk: Analysis, Perception and Management. London: The Royal Society, 1992.
21. Ozonoff D, Colten ME, Cupples A, Heeren T, Schatzkin A, Mangione T, Dresner M, Colton T. Health problems reported by residents of a neighborhood contaminated by a hazardous waste facility. Am J Ind Med 1987; 11:581–597.
22. Tessler R, Mechanic D. Psychological distress and perceived health status. J Health Soc Behav 1978; 19:254–260.
23. Hertzmann C, Hayes M, Singer J, Highland J. Upper Ottawa Street landfill site health study. Environ Health Perspect 1987; 75:173–195.
24. Roht LH, Vernon SW, Weir FW, Pier SM, Sullivan P, Reed LJ. Community exposure to hazardous waste disposal sites: assessing reporting bias. Am J Epidemiol 1985; 122:418–433.
25. Eyles J, Taylor SM, Johnson N, Baxter J. Worrying about waste: living close to solid waste disposal facilities in southern Ontario. Soc Sci Med 1993; 37:805–812.
26. Hallman WK, Wandersman A. Attribution of responsibility and individual and collective coping with environmental threats. J Soc Issues 1992; 48:101–118.
27. Dayal HH, Li Y, Dayal V, Mittal CK, Snodgrass W. Use of medication data to validate an association in community-based symptom prevalence studies. Arch Environ Health 1994; 49:93–97.
28. Horowitz J, Stefanko M. Toxic waste: behavioral effects of an environmental stressor. Behav Med 1989; 15:23–28.
29. Brown P. Popular epidemiology, toxic waste and social movements. In: Gabe J, ed. Sociology of Health and Illness. Oxford: Blackwell, 1995.
30. Fitchen JM, Heath JS, Fessenden-Raden J. Risk perception in community context: a case study. In: Johnson BB, Covello VT, eds. The Social and Cultural Construction of Risk. Boston: Reidel, 1987; 31–54.
31. Martin C, Platt S, Hunt S. Housing conditions and ill-health. BMJ 1987; 294:1125–1127.
