Multilevel Modeling

18. Summary

Much, if not all, of the current research linking neighborhoods and health is cross-sectional, and assumes a hierarchical structure of individuals nested within neighborhoods. This simplistic scenario ignores possibilities such as, for instance, the fact that an individual might move several times and as such reflect neighborhood effects drawn from several contexts, or that other competing contexts (e.g., schools, workplaces, hospital settings) may simultaneously contribute to contextual effects.

While a multilevel analytic approach provides a sound rationale for modeling ecologic effects, it obviously does not overcome the limitations intrinsic to any observational study design, single-level or multilevel. Recent discussions on identifying causal ecologic effects (Oakes, 2004), inappropriately conflate issues of study design with the relevance of multilevel models. The critical challenge for identifying neighborhood effects arises from the use of observational (and often cross-sectional) study design, and not from the use of any specific analytic technique. Arguably, multilevel models are appropriate analytical techniques for understanding ecologic effects, regardless of whether the data were generated through observation or were a result of an experiment (Subramanian, 2004).

Identifying true casual associations of ecologic exposures with health:

Careful study design, thorough analysis, and careful interpretation, are essential components of this work. However, there are some other pointe to consider as well.

1. Rely less on interpreting residual associations, and model directly the ecological exposure. The above example of a composite index of socio-economic deprivation is a classic example. It is difficult to interpret a residual association of such an index with individual health for the reasons listed above, and because it is not actually clear what properties of neighborhoods the index is actually capturing. Following the longstanding exhortations of Macintyre and others (Macintyre, Maciver et al., 1993; Macintyre, Ellaway et al., 2002; Macintyre and Ellaway, 2003; Cummins, Macintyre et al.,2005), conceptualizing and directly measuring those characteristics of neighborhoods that are hypothesized to effect health is likely to be more rewarding in the long-run, albeit more difficult.
2. While often impossible to conduct, intervention studies that actually change ecological or neighborhood characteristics should be seized upon by researchers whenever possible.
3. Third, longitudinal studies with repeated measurements of neighborhood characteristics over peoples’ lifecourses should also be sought out.
4. We need to be cognizant of the limits of quantitative multilevel analysis and empiricism more generally.

There is a deep, complex, and dynamic inter-relationship between people and context. Where you live influences who you are (e.g., employment opportunities), and who you are influences your neighborhood. It will not always be possible, nor correct, to decompose health variations to personal and contextual characteristics. Rather, we will also need qualitative and other social science approaches.