There is growing awareness among health researchers and policy makers that the local environment is important for people's health. This has resulted in a large number of research initiatives examining the health effects associated with different social and physical aspects of the neighbourhood environment. According to Yen and Syme, this research can be divided into studies investigating the roles of (1) the quality of the local environment, (2) the social organisation of the local community, often reflected in social capital and/or social cohesion, and (3) the socio-economic status of the neighbourhood [1]. All three aspects of the neighbourhood environment appear to be associated with community health. Studies focussing on the quality of the local environment have shown that perceived neighbourhood problems and crime, the absence of green space, and poor access to amenities are linked with the reporting of poor health [2–10]. The importance of social capital and social cohesion for people's physical and psychological well-being is now well known in the social epidemiological literature [11–13]. Although there is still considerable disagreement about the definition, measurement, and workings of social capital and cohesion, various studies have shown that these complex social constructs are associated with a wide range of health outcomes, such as life-expectancy, mortality rates, and self rated health [14–19]. The group of studies investigating the health effects of community socio-economic status consistently report significant associations with a large number of health outcomes, generally showing that people living in deprived neighbourhoods have poorer health [1, 20].

A limitation of many studies investigating the health effects of social and physical aspects of the neighbourhood environment is that they fail to take into account differences in neighbourhood socio-economic status. The results of these studies are ambiguous considering that disadvantaged neighbourhoods may suffer from poorer physical and social infrastructures, [5] and that there is still a lack of clarity about the specific pathways that link area of residence to health [21]. It may be that, as suggested by Stafford and Marmot [22], the association between neighbourhood deprivation and health is mediated by perceived neighbourhood and housing problems, or that the associations between specific neighbourhood attributes and health are spurious, and may actually reflect the impact of neighbourhood socio-economic status.

There is some evidence that the place where a person lives may not affect all people in the same way [22–25]. Recent research suggests that the neighbourhood environment is more important for women than for men [26–28]. The social environment appears to be particularly important for women's health status. In contrast, men's health status seems to be more affected by the physical environment and by neighbourhood socio-economic status [28–31].

In this study we examine the importance of different social and physical aspects of the neighbourhood environment for people's self rated health. We specifically focus on the neighbourhood aspects of access to amenities, neighbourhood quality, neighbourhood disorder, and social cohesion; and investigate whether the possible health effects reflect differences in socio-demographic composition and/or neighbourhood deprivation. We additionally examine whether the associations with different aspects of the neighbourhood vary between men and women.

Table 1 presents the characteristics of the CHSNS dataset for men and women separately. It shows that men were slightly more likely to report poor health than women. A gender difference was found for social class: men were more often classified as III M (skilled manual), and women more often as III NM (skilled non-manual). There were also some small differences with regard to employment status: whereas men were more likely to be employed, women were more likely to be a student or to look after the home and/or children. Men and women appeared to have similar perceptions of the neighbourhood environment.

This study examined the associations between different aspects of the neighbourhood environment and people's self rated health, and investigated whether these associations vary between men and women. The results showed that, before and after controlling for socio-demographic differences, the neighbourhood characteristics of poor access to amenities, poor neighbourhood quality, neighbourhood disorder, lack of social cohesion, and deprivation are associated with the reporting of poor health. They further showed that social cohesion is linked with women's health but not men's health, and that unemployment affects men's health but not women's health.

This study provides further empirical evidence for the idea that the neighbourhood environment is associated with people's health. It confirms the findings of a number of previous studies that the presence of neighbourhood problems and crime, and poor access to amenities, are linked with poor health [3–6, 8, 10]. The result that neighbourhood deprivation is detrimental to people's health, even when taking into account individual socio-economic status, is consistent with other studies on the impact of the socio-economic environment on public health [1, 20]. Social cohesion was inversely related to the reporting of poor health, confirming that the social organisation of the local community plays an important role in public health [1, 5, 13, 17, 19, 46]. Overall, the contextual effects of the different aspects of the physical and social environment could be considered modest, in particular when simultaneously controlling for socio-demographic differences and neighbourhood deprivation. In this context it has to be kept in mind that because of its large sample size the study has the statistical power to detect small neighbourhood effects. However, neighbourhood health effects are generally found to be small [5, 6, 17, 28], and therefore large sample sizes are required to detect these effects. It also has to be considered that this study may have underestimated the size of neighbourhood effects through over-adjustment. While individual level variables are often seen as confounders, they could in some cases be conceptualized as mediators. That is, neighbourhoods may influence the life chances of individuals through their effects on achieved income and education, as well as through other residential segregation and selection processes [47].

The results of this study further suggest that the interpretation of the possible health effects of social and physical aspects of the environment may not be straightforward. We found that the health effects of neighbourhood quality, neighbourhood disorder, and social cohesion are substantially attenuated when controlling for individual and collective socio-economic status. This suggests that, instead of being aspects of the neighbourhood environment that independently contribute to people's health, they may mediate or reflect differences in socio-economic status. Unfortunately, the difference between mediation and confounding can not be shown using statistical methods alone, and a decision can only be made based on theory [48]. There is a need for more detailed analyses to determine the specific associations between neighbourhood deprivation and the quality of the social and physical environment, as well as the different possible pathways that link them to health. In either case, the results show that neighbourhood deprivation is associated with worse neighbourhood disorder, poorer neighbourhood quality, and less social cohesion.

The gender-stratified analyses show that a lack of social cohesion significantly increases the odds of women reporting poor health, but not the odds of men reporting poor health. In contrast, unemployment appeared more detrimental to men's than to women's health. These findings are similar to a number of recent studies, and suggest that whilst social aspects of the residential environment are more important for women, individual socio-economic status is more important for men [27–29]. It has been hypothesised that the neighbourhood environment is more important for women because they spend more time at home looking after children, doing domestic work or being primary carers [28]. Although we found that women were indeed more likely to look after the home and/or children, the contextual effects of neighbourhood social cohesion remained significant when adjusting for these differences. Others have suggested that time spent at home is an unlikely explanation, as there are no substantial gender differences in exposure to the residential environment [27]. Another possibility is that there are gender differences in the type and use of social relations [49]. It may be that men are more reliant upon workplace networks, and therefore are more affected by their employment status, but less so by the quality of the local environment. Indeed, men tend to report more support at the workplace than women, which seems to be caused by differences in occupational grade [50]. Although gender differences in the associations between neighbourhood and health have been demonstrated in a number of social epidemiological studies, there is a need to be cautious when interpreting these results. It remains unclear whether social interactions fulfil the same needs and/or mean the same thing to men and women.

It should be acknowledged that the current study has a number of limitations. These limitations are similar to most quantitative cross-sectional studies on neighbourhoods and health. Only associational claims can be made, as reverse causality can not be ruled out. Furthermore, the study focused on perceived aspects of the social and physical environment, which do not necessarily correspond with 'objective' aspects of the neighbourhood. In addition, and as a related issue, the study may have introduced a reporting or same source bias, as the area-specific measures were constructed using variables from the same study population as the health data. It would have been better to collect neighbourhood data independently from the main dataset [27]. However, that was not possible within the scope of this study. We have previously investigated whether same source bias was an important problem in the dataset [19]. In an analysis of associations between individual mental health status, small-area income deprivation and social cohesion, a split-half analysis of the dataset suggested that same source bias is not an important issue in this study.

In this study we used multilevel modelling to construct variables at the neighbourhood level of analysis. That is, we took the level 2 residuals of the neighbourhood perception variables to reflect the area-specific contributions to the overall scores. Although this alternative method of scale construction makes it difficult to compare the results with previous studies, the approach is preferable to the conventional method of aggregating individual level variables. The shrunken residuals derived from multilevel modelling provide more precise estimates of the different neighbourhood characteristics, as they only consider the part of the variance that is specifically attributable to neighbourhood differences.

Another methodological issue associated with many small-area studies is the use of administrative units as proxies for neighbourhoods. Identifying the correct geographical level of analysis is important, as misspecification may have implications for the study outcomes [47]. Although it is not clear whether EDs match people's perceptions of the neighbourhood, EDs are probably the best administrative units to represent neighbourhoods, as these are the smallest and most homogeneous geographical unit available. It may also be difficult to generalise the results to the larger (British) population. The study was conducted in one of the most deprived Welsh boroughs, and may not be representative of the wider population. However, it can also be considered a strength that the study was restricted to a well-defined geographical area. Considering that the residents of this borough share many characteristics and experiences it may be easier to draw conclusions on the basis of the identified neighbourhood effects. An additional strength of the study is the representative nature of the dataset obtained from a high response rate and high sampling fraction of individuals resident in Caerphilly borough. In this study the average response rate was 61.9% across the 325 EDs, and ranged from 32.7% to 87.5%. Although it is possible that the variation in response rate may have introduced a degree of response bias, we do not believe that it has substantially influenced the results of the study. This study's main aim was to examine associations between different area characteristics and health and not to make inferences about the study population itself, for which response rates are more important. Although it is likely that respondents with a lower socio-economic status were somewhat underrepresented in this study, this will only have affected the results if the effects of neighbourhood on health are different for this particular subgroup. Evidence for cross-level interactions between individual and neighbourhood socio-economic status is somewhat conflicting. Whereas some studies found an interaction between individual and area socio-economic status [51, 52], other studies found no such cross-level interactions [53, 54]. In addition, response rates were found to be weakly associated with neighbourhood deprivation, with response rates being slightly lower in neighbourhoods with a lower socio-economic status. This seems to suggest that we may have underestimated neighbourhood effects, as a lower response rate in deprived neighbourhoods may have reduced health variation across the different neighbourhoods.

The current study provides some pointers for health policy and practice. The findings that certain social and physical aspects of the neighbourhood environment are important for public health may help the design of community-based interventions. The results suggest that improvements in the environment, in terms of access to amenities, neighbourhood nuisances and neighbourhood disorder, may help to contribute to people's well being. However, this study also found that these aspects of the neighbourhood environment are strongly interlinked with socio-economic deprivation. As suggested before, this could mean that the aspects of neighbourhood quality and disorder, and social cohesion reflect differences in neighbourhood deprivation. Therefore the underlying socio-economic context should not be ignored when interpreting the results.

The results further contribute to the newly emerging evidence that there are important gender differences in the associations between the neighbourhood environment and health. Hitherto, only a small number of studies have examined gender differences in (contextual) neighbourhood health effects [26–29]. The results with regard to the social environment appear to be in line with those studies, despite differences in the conceptualisation and measurement of the social environment. For example, where Stafford et al. [27] used social trust measures that were collected in an independent sample, the current study used an ecometric approach to transform individual-level variables from the same dataset into contextual variables. Unlike Stafford et al., we found no significant interaction between gender and neighbourhood deprivation – even though we also obtained the deprivation indicator from an independent external source. Despite the absence of a statistically significant interaction between neighbourhood socio-economic status and gender, the results show that individual socio-economic status is more important for men than for women. Overall, the suggestion that men and women interact with their environment in a different way, and thus are likely to benefit from different types of interventions, may have important health policy implications. This could mean that health initiatives should be more community based for women and more workplace based for men. However, further research is needed to gain a more detailed understanding of the social processes underlying these gender health effects.

Our results indicate that various aspects of the social and physical neighbourhood environment are associated with self rated health. However, these associations seem to be confounded by neighbourhood deprivation, suggesting that the neighbourhood aspects of access to amenities, neighbourhood quality, neighbourhood disorder, and social cohesion mediate or reflect differences in socio-economic status. The study further suggests that the social environment is more important for women's health, but that individual socio-economic status is more important for men's health. The suggestion that men and women benefit differently from the neighbourhood environment warrants further investigation of gender differences in neighbourhood health effects.