Estimating neighborhood health effects: the challenges of causal inference in a complex world

Abstract

Oakes raises a series of important questions on the validity of past work on neighborhood health effects and suggests directions the field should take (Oakes, 2003). Indeed the limited nature of the evidence linking neighborhoods to individual-level outcomes, and its many methodological problems, have been noted in health and other fields (Diez Roux, 2001; Duncan & Raudenbush, 1999; Furstenberg & Hughes, 1997; Macintyre, Ellaway, & Cummins, 2002; Sampson et al., 2002; Tienda, 1991). The complex methodological issues inherent in the estimation of causal neighborhood effects are nevertheless worth reiterating and elaborating on as Oakes does. I would also posit that many of the issues Oakes raises are common to epidemiology generally (and to observational studies in other fields) and are not necessarily specific to research on neighborhood effects (or to so-called ‘‘social epidemiology’’ as he sometimes implies). Perhaps one of the problems is that in research on neighborhoods and health (as in epidemiology generally) the word ‘‘effects’’ is often used loosely, leading to the impression that the associations reported are always valid and precise estimates of generalizable causal parameters, when of course they are not. I will begin with some general comments on points raised by Oakes when he describes ‘‘a causal model for neighborhood effects’’. I will then summarize agreements and disagreements with the methodological obstacles Oakes notes, and with the approaches he proposes.