Perceived neighbourhood quality and adult health status: new statistical advices useful to answer old questions?

Interest in the quantitative effects of neighbourhood characteristics on adult health has recently increased in social epidemiology. Particularly, investigations concern the statistical influence on health of several individual demographic and socioeconomic characteristics and of neighbourhood characteristics as perceived by respondents. We analyze these issues within an original conceptual framework and employing statistical models unusual in this context. We use data collected in the Los Angeles Family and Neighbourhood Survey (L.A.FANS) to model the number of hospital admissions occurred to each individual as a function of some individual and neighbourhood characteristics, the latter being related to the individual perceptions about the neighbourhood he lives in. We employ generalized additive models with different distributional assumptions: Poisson, Negative Biomial and Zero Inflated Poisson (ZIP). Such models allow us to estimate (through spline functions) potential non linear effects of the covariates on the response. Moreover, non standard representations are used to overcome difficulties in interpreting the results for ZIP models. It turns out that perceived neighbourhood characteristics, and in particular the perception of social cohesion, have a significant effect after controlling for individual characteristics relevant to hospital admissions frequency. From a modeling point of view ZIP and Negative binomial models prove to be superior to standard Poisson model. We have confirmed the role of the neighbourhood where an individual lives in determining his health status. A strength of this analysis is that, due to the choice of the neighbourhood characteristics to be included in the model, the results do not depend of a particular definition of neighbourhood (which is traditionally based on administrative boundaries), since each individual refers his perceptions to his personal definition of it.