because of fear; in this model, a two-standard-deviation elevation in collective efficacy was associated with a 26 percent reduction in the expected homicide rate (Sampson et al., 1997:922). Measures of social cohesion and trust have also been found to predict mortality rates at the state level. The level of distrust (the proportion of residents in each state agreeing that most people cannot be trusted) was strongly correlated in one study (Kawachi et al., 1997) with the age-adjusted mortality rate (r = .79, p < .001). Lower levels of trust were associated with higher rates of most major causes of death, including coronary heart disease, unintentional injury, and cerebrovascular disease. A one-standard-deviation increase in trust was associated with a 9 percent lower level of mortality.
Despite promising leads from existing research, numerous limitations must be addressed if scientific knowledge is to progress. Indeed, methodological issues such as the differential selection of individuals into communities (compositional and selection effects), indirect environmental effects that work through family and peer mechanisms, measurement error, spatial interdependence (e.g., diffusion processes), and simultaneity bias (e.g., poor health causing poverty) represent serious challenges to our ability to draw definitive conclusions on the role of neighborhood and community social contexts. Equally important, there is a need to further develop multilevel methodologies for contextually based research. Health data collected at nested levels of aggregation (e.g., neighborhood, city, state) pose important challenges to the standard analytic procedures used by health researchers. A methodological program of research is thus needed to develop tools for the proper evaluation and analysis of community-level data.
A central challenge in this regard is to build strategies for direct measurement of the social mechanisms and collective properties hypothesized to predict health. As interest in the social sciences turns increasingly to an integrated scientific approach that emphasizes individual factors in social context, a mismatch has arisen in the quality of measures (Raudenbush and Sampson, 1999). Standing behind individual measurements are decades of psychometric and biological research, producing measures that often have excellent statistical properties. In contrast, much less is known about measures of ecological settings. Neighborhood-level research in particular is dominated by the study of poverty and other demographic characteristics drawn from census data or other government statistics that do not provide information on the collective properties of administrative units. We thus recommend a concerted methodological effort to enhance the science of ecological assessment (“ecometrics ”) of social environments relevant to