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

https://doi.org/10.1016/S0277-9536(03)00414-3Get rights and content

Introduction

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.

Section snippets

The selection or confounding issue

Undoubtedly, the selection issue (the fact that persons may be selected into neighborhoods based on individual attributes which are themselves related to health) is the key problem in observational studies of neighborhood effects. Epidemiologists have attempted to account for this by controlling for often numerous individual-level variables. One of the problems, as Oakes notes, is that when numerous covariates are included it is likely that sparse data will be found in many cross-tabulated

Building the “causal multilevel model for neighborhood effects”

Oakes proposes a series of steps to construct what he refers to as “a causal multilevel model for neighborhood effects”. The steps he proposes are logical, based on existing multilevel analysis literature, and have often been followed (either explicitly or implicitly) by neighborhood effects researchers. As in model building generally, there is no single way to build a “causal multilevel model”, and using a given approach does not guarantee that the estimates obtained will be valid estimates of

Social stratification confounds comparisons

Oakes correctly notes that if “there is (approximately) complete confounding between the background attributes of persons in a given neighborhood and (approximately) complete separation between background attributes of people in other neighborhoods” it will be impossible to identify neighborhood effects. But the extent to which individual and neighborhood characteristics are so strongly associated is an empirical question that can be examined in the data. Although neighborhood characteristics

Where does this leave observational studies of neighborhood effects?

Have contextual and multilevel studies of neighborhood health effects concluded that the observed associations are causal? No. Should the conditional measures of associations they have generated be taken literally as valid and definite measures of causal effects parameters? Of course not, as most rigorous studies of neighborhood effects note. But although still extremely crude, taken together with other data (plus many of our daily living experiences which are after all, an important source of

First page preview

First page preview
Click to open first page preview

References (34)

  • S. Macintyre et al.

    Place effects on healthHow can we conceptualise, operationalise and measure them?

    Social Science & Medicine

    (2002)
  • H. Brown et al.

    Applied mixed models in medicine

    (1999)
  • A.V. Diez Roux

    Investigating neighborhood and area effects on health

    American Journal of Public Health

    (2001)
  • A.V. Diez Roux et al.

    Area characteristics and individual-level socioeconomic position indicators in three population-based epidemiologic studies

    Annals of Epidemiology

    (2001)
  • G. Duncan et al.

    Conceptual and methodological issues in estimating causal effects of neighborhoods and family conditions on individual development

  • G. Duncan et al.

    Assessing the effects of context in studies of child and youth development

    Educational Psychologist

    (1999)
  • S.P. Fortmann et al.

    Community intervenstion trialsReflections on the Stanford five-city project experience

    American Journal of Epidemiology

    (1995)
  • F. Furstenberg et al.

    The influence of neighborhoods on children's development

  • A.T. Geronimus et al.

    Use of census-based aggregate variables to proxy for socioeconomic groupEvidence from national samples

    American Journal of Epidemiology

    (1998)
  • H. Goldstein et al.

    Improved approximations for multilevel models with binary responses

    Journal of the Royal Statistical Society A

    (1996)
  • S. Greenland

    A review of multilevel theory for ecologic analyses

    Statistic in Medicine

    (2002)
  • S. Greenland et al.

    Identifiability, exchangeability, and epidemiological confounding

    International Journal of Epidemiology

    (1986)
  • Harding, D. (2002). Counterfactual models of neighborhood effects: The effect of neighborhood poverty on high school...
  • M.M. Joffe et al.

    Invited commentaryPropensity scores

    American Journal of Epidemiology

    (1999)
  • T.D. Koepsell et al.

    Invited CommentarySymposium on community intervention trials

    American Journal of Epidemiology

    (1995)
  • I. Kreft et al.

    The effect of different forms of centering in hierarchical linear models

    Multivariate Behavioral Research

    (1995)
  • R. Levins

    Ten propositions on science and antiscience

    Social Text

    (1996)
  • Cited by (263)

    • Neighborhood disadvantage and the risk of dementia and mortality among refugees to Denmark: A quasi-experimental study

      2023, SSM - Population Health
      Citation Excerpt :

      First, unmeasured factors such as socioeconomic status and pre-existing health may confound the association between neighborhood characteristics and health (Oakes, 2004). The association may merely reflect the residential sorting that occurs during childhood and early adulthood, such that individuals with poor health may be selected into disadvantaged neighborhoods (Diez Roux, 2004). Second, there are multiple ways to define and operationalize neighborhoods (e.g., where to set the boundary or which scale to use) (Diez Roux, 2001; Diez Roux & Mair, 2010; van Ham & Manley, 2012), requiring attention to historical, social, and contextual factors (e.g., administrative boundaries and people's perceptions) (Diez Roux, 2001).

    View all citing articles on Scopus
    View full text