METHODS

Data for this study came from a cross sectional health survey, designed to assess determinants of health at the local level, conducted in four neighbourhoods in Hamilton, Ontario. A detailed description of the neighbourhood selection process can be found in Luginaah et al.²⁴ In summary, neighbourhoods were selected through a combination of statistical methods that used socioeconomic and demographic data extracted at the census tract level from the 1996 census of Canada in conjunction with smoking data from a random survey of adults. Principal component analysis, local indicators of spatial association (LISA), and geographical information systems were used to identify neighbourhoods representing clusters of 17 socioeconomic and demographic determinants of health and related risk factors. This analysis was coupled with qualitative interviews with prominent local decision makers and key informants, which identified similar areas of interest for study. The goal was to select neighbourhoods with a range of characteristics representing salient hypotheses from the population health literature. Interestingly, the selected neighbourhoods displayed various combinations of economic wealth and social diversity, including high versus low income and high versus low social and demographic diversity (for example, lack or presence of recent immigrants, visible minorities, etc). The neighbourhoods selected (see figure 1) are referred to as Chedoke-Kirkendall (high income and high diversity), the Downtown Core (low income and high diversity), Northeast Industrial (low income and low diversity), and the Southwest Mountain (high income and low diversity) (see table 1 for specific neighbourhood characteristics).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Study neighbourhoods in Hamilton, Ontario Canada.

A telephone survey (n = 1504) was administered to a random sample of about 300 selected people in each neighbourhood and to an additional 300 respondents randomly chosen from across the city (called the “Old City” in figure 1 because of a subsequent boundary change that occurred after our study began) excluding the four neighbourhoods between November 2001 and April 2002. The survey was completed by people aged 18 years and older and has an overall response rate of 60%. The survey contained a range of questions designed to capture residents’ perceptions of their neighbourhoods, social and community networks, health status and behaviours, use of healthcare services, and demographic and socioeconomic characteristics. Approval for this research was granted by the university ethics committee.

Logistic regression analysis was used to examine the relation between three health outcomes, including: self assessed health status (fair/poor versus good/very good/excellent), chronic conditions (one or more), emotional health (using the GHQ-20),^25,26 and perceptions of neighbourhood environments. A backward conditional regression model with a p⩽0.1 (to be inclusive) was conducted for each health outcome and included a range of individual factors. The use of stepwise regression for neighbourhood analysis has been validated by other studies.^27,28 Independent variables included in all initial backward conditional regression models represented a range of factors including length of residence, condition of dwelling, a neighbourhood of residence dummy variable (Old City is the reference category), as well as a range of demographic and socioeconomic factors.

A minority of respondents refused to report income (17%). Subsequent analysis revealed that older women with low education and poor health status were more likely to refuse this question. To include more cases in the final analysis and overcome this systematic bias, values were interpolated using a hot-deck imputation approach²⁹ and Stata 7 software. The hot-deck approach stratifies the sample based on gender, age, education and employment status, factors that were associated in our sample with lower reported income among those who did report. For every missing value, one case with the same characteristics for the stratification variables is chosen randomly and the associated income value of the “donor” substituted for the missing income data. The interpolated values are slightly lower than the sample mean but this is expected given the characteristics associated with non-reporting income discussed above. This method is similar to imputation used in related studies.¹⁶

In addition, individual risk factors were included in the models, including three dichotomised variables that compare smokers with non-smokers, people who have exercised in the past three months with those that have not, and overweight compared with non-overweight people, as measured by body mass index (an index of weight to height).

Those independent variables at or near conventional significance of p⩽0.05 in each of the backward stepwise regression models were then selected and entered manually into a multiple regression model until a parsimonious model was achieved. As a result, while all variables were entered in the initial stepwise regression model, the results presented in table 4 are only for those variables included in the parsimonious model for each health outcome. Given the a priori importance attached to neighbourhood of residence, the neighbourhood variable was forced into each model.

In much of the research on neighbourhood and health, respondents are asked to rate the extent to which a list of pre-defined neighbourhood characteristics are problems in their neighbourhood. For example, Ellaway et al¹⁷ measure respondents’ perceptions of neighbourhood by asking them to assess 13 socioenvironmental problems (for example, vandalism, nuisance from dogs, burglaries, lack of recreational facilities). Others have used similar instruments.^16–18 In contrast, in the Hamilton survey respondents were allowed to respond to two open ended questions designed to measure likes and dislikes about their neighbourhood. Only the first mention was recorded. Respondents were also asked to suggest the most important thing that could be done in their neighbourhood to improve or maintain their health. Respondents could provide up to three mentions for this question, however, because of the low number of responses for the second and third mentions, only the first mention is included in the analysis. Finally, respondents were asked to rate their overall neighbourhood satisfaction on a four point scale ranging from very satisfied to very dissatisfied. Although we allowed for more scope in the open ended responses, we followed the method used by Ellaway et al¹⁷ of classifying responses obtained for each of three open ended questions into categories that were either “social” (for example, lack of safety, friendliness of neighbours) or “physical” (for example, green space, pollution). Reliability in coding was ensured by dual coding of the responses by more than one researcher. Table 2A illustrates the typical responses assigned to each category and table 2B shows selected examples of social and physical neighbourhood characteristics. The following section reports results of descriptive analyses and multivariate logistic regressions.

RESULTS

Significant differences in health status across the four neighbourhoods are apparent, with twice as many respondents in the Downtown and Northeast neighbourhoods reporting fair/poor health as compared with the Chedoke-Kirkendall and Mountain neighbourhoods (table 3). While little variation in the reporting of chronic conditions exists, a much higher percentage of residents in the Downtown Core and the Northeast neighbourhoods scored above the cut off point (⩾4) on the GHQ-20, indicating a probable case of emotional distress. Almost 50% of Mountain respondents reported no dislikes about their neighbourhood, much higher than the other three neighbourhoods (table 3). Similarly, the highest percentage of respondents reporting no potential neighbourhood improvements for improving health was also in the Mountain neighbourhood (see table 3).

A particular focus of this work was to differentiate between social and physical aspects of the neighbourhood. For almost all indicators, physical aspects of neighbourhood were mentioned with greater frequency than social aspects (table 3). Social dislikes were reported with the greatest frequency by respondents in the Downtown neighbourhood (35%). Physical dislikes were reported with the greatest frequency in the Chedoke-Kirkendall (52%) and Northeast (52%) neighbourhoods, both areas with proximate industrial land use. Almost 20% of Downtown residents report dissatisfaction as compared with less than 5% of Mountain and Chedoke-Kirkendall respondents. With respect to suggestions for neighbourhood improvements that may affect health and wellbeing, most were physical (table 3), with the highest levels of reporting in the Chedoke-Kirkendall and Northeast neighbourhoods. Although the level of reporting of social improvements was much lower, the highest frequency of reporting was in the Downtown neighbourhood (20%). For example, “get rid of prostitutes” was a frequent response.

Table 4 gives the results of logistic regression models for each of the three health outcomes. Consistent with past research on self assessed health status,^30–32 low household income, and being overweight are associated with increased odds of being in poor health (table 4). Three of the neighbourhood variables are significantly related to self assessed health status. Respondents living in the Northeast neighbourhood are twice as likely to report fair/poor health (OR 2.07, 95% CI 1.23 to 3.49) as compared with people living in the “Old City” area. People who report that they want to improve the social aspects of their neighbourhood have increased odds of reporting fair/poor health (OR 1.80, 95% CI 1.05 to 3.16) relative to those who report no improvements. In contrast, people who report that they like physical characteristics of their neighbourhood are less likely to rate their health as fair/poor (OR 0.50, 95% CI 0.28 to 0.90) as compared with those who report that they like nothing about their neighbourhood.

With respect to emotional health, women, smokers, and those with an annual household income below the LICO have significantly increased odds of scoring above the GHQ-20 cut off point than do men, non-smokers, and those above the LICO. Respondents who have lived in their neighbourhood for less than five years and those living in homes in need of major repairs are more likely to score above the GHQ-20 cut off point than those who have resided in their neighbourhood for more than five years and those living in homes not requiring major repairs. Similarly, people who are dissatisfied with their neighbourhood as a place to live have increased odds of scoring above the cut off point (OR 2.39, 95% CI 1.40 to 4.09) than those who are satisfied, while those who like their neighbourhood’s physical aspects have decreased odds (OR 0.45, 95% CI 0.26 to 0.80) as compared with those who report no neighbourhood likes.

The model for chronic conditions shows that the odds of reporting one or more chronic conditions increases with age, again consistent with past findings^33,34 (table 4). Low education levels and unemployment are both associated with increased odds of reporting one or more chronic conditions. Respondents who are dissatisfied with their neighbourhood (OR 1.66, 95% CI 1.05 to 2.64) or report that they dislike physical aspects of their neighbourhood (OR 1.56, 95% CI 1.19 to 2.05) both have increased odds of reporting one or more chronic conditions. This observation may be linked to arguments suggesting that people with physical impairments (that is, long term physical conditions) may rate their neighbourhoods as more problematic than less impaired people because of mobility restrictions.¹⁸

DISCUSSION

The results indicate that significant differences in health status across the four neighbourhoods are apparent in self assessed health status and emotional distress, but this is not the case for chronic conditions. Neighbourhoods with lower SES reported poorer health and more emotional distress. Perceptions of the physical environment seem more important than those of social concerns in all neighbourhoods. For all three health outcomes, individual risk factors followed expectations, with measures of SES, age, and lifestyle all significantly associating with poor health outcomes. Physical environmental problems were significantly associated with chronic conditions while reporting physical likes were associated with both lower odds of reporting poor self assessed health status and lower odds of scoring above the GHQ cut off point.

Our results show the importance of individual perceptions of neighbourhoods for shaping health. The neighbourhood of residence dummy variable was entered into all three models but was only significant for self assessed health. This suggests that perceptions about the neighbourhoods in which people live are just as important for health as the neighbourhoods themselves.* Much of the research linking neighbourhood perceptions to health has focused on neighbourhood problems.^16–18 In our analysis, we not only test for the health inhibiting (that is, dissatisfaction, dislikes and suggestions for improvement) aspects of physical and social neighbourhood environments but also the health promoting (that is, likes) aspects. Our results suggested a differentiated relation between health inhibiting and health promoting aspects and each of the three health outcomes. Neighbourhood dissatisfaction and physical dislikes were associated with increased odds of reporting chronic conditions while physical likes were associated with decreased odds of fair/poor health and scoring above the GHQ-20 cut off point. Finally, respondents who stated that the social aspects of their neighbourhood needed improving were more likely to report fair/poor health. Overall, however, physical environmental concerns dominated social ones as determinants of health. In discussing the importance of perception, a few methodological and interpretive problems deserve mention.

Firstly, the findings showed significant differences in both health status and perceptions of social and physical dimensions of neighbourhood among the four study neighbourhoods (table 3). It may be the case that social and physical characteristics have differential levels of importance for health in different neighbourhoods. It is possible that physical aspects might be a strong predictor of health in the Northeast Industrial neighbourhood due to its proximity to the steel industry and the comparatively high pollution levels there.³⁵ In contrast, social aspects might have a stronger effect in the Downtown Core where there are higher rates of crime, homelessness, drug trafficking, and prostitution. These questions await further research.

Another limitation may be our definition of neighbourhood. While much research has used administrative districts, such as census tracts,¹¹ this study sought to define the boundaries of neighbourhoods through spatial clusterings of specific demographic and socioeconomic features with the input of key decision makers in the city.²⁴ Had the boundaries been drawn differently, relations between neighbourhood and health might be different. Furthermore, as Caughy et al¹⁴ note, specification of neighbourhoods for research purposes does not always match perceptions of neighbourhood boundaries of residents.^36,37

The use of open ended questions may have introduced some bias in the models with the tendency of some respondents to report always positively or negatively. However, we chose this method for measuring neighbourhood perceptions (in contrast with pre-defined lists of characteristics) as it allows respondents to identify those characteristics most important to them.

Furthermore, we cannot conclude if perceptions of social and physical neighbourhood characteristics reflect the local environment in which people live or individual attributes (for example, state of mind, health status). That is, the relation between perceptions of neighbourhood characteristics and health are not necessarily causal. Ellaway et al¹⁷ argue that results like these show that either people in poor health are more negative about their neighbourhoods or people with a pessimistic view are more likely to report both their health and neighbourhoods are poor. In addition, as Diez Roux³⁹ notes, health and neighbourhood characteristics, such as physical and social environments, may influence each other.^22,23

Despite these limitations such findings support the suggestions of Macintyre et al³⁹ that “a more differentiated picture has tended to emerge, in which rather than there being one single, universal ‘area effect on health’ there appear to be some area effects on some health outcomes, in some population groups, and in some types of areas” (page 128). While it is not possible to identify which specific social and physical attributes of neighbourhoods are health promoting and which are health inhibiting, the research demonstrates the importance of neighbourhood perceptions for health. The dominance of physical concerns over social ones may have resulted from the industrial nature of Hamilton, but this finding highlights the need for further research to identify factors driving perceptions of place as a determinant of health.