In this issue, Stjärne et al1 describe socioeconomic gradients in the incidence of myocardial infarction (MI) using population-based data from Sweden. The authors estimate contextual effects, meaning effects attributable to the characteristics of communities rather than to the characteristics of people who live in these communities. This partitioning between individual-level and community-level effects is a long-standing strategy in social epidemiology and is pursued by adjusting for measured individual-level characteristics. Partial regression coefficients at the second level of the multilevel regression model are then viewed as representing purely contextual phenomena because their interpretation is conditional on the modeled individual factors.2 This strategy has been promoted on the grounds that substantial outcome variability from one place to another, conditional on measured individual factors, motivates public health intervention at the community level.3
Various critiques of this inferential strategy have been raised, including the hubris of thinking that the important individual level factors have been identified, adequately measured, and appropriately modeled.4 Even a highly elaborate predictive model of MI would likely leave much outcome variability unexplained, and so it would seem that the assumption of having accounted for all compositional effects (that is, all effects due to individual characteristics) is unreasonable. If obesity, diabetes, hypertension, or any other significant but unmeasured determinant of MI were differentially distributed by neighborhood income—but not wholly determined by neighborhood income—then the appearance of a contextual effect could be entirely spurious.5
Thus, the notion of a contextual effect must be viewed much more loosely in these types of studies and understood in relation to the specific individual factors that were measured and modeled. Imbalances in unmeasured individual level factors between neighborhoods are just as much a part of the estimated contextual effect as are “true” neighborhood level factors. Indeed, it may not always be obvious (even conceptually) what is truly a contextual factor because compositional factors often are expressed directly at the contextual level. For example, religious affiliations of individual community residents are compositional factors, but their manifestations in the form of Halal butchers or Kosher delicatessens are contextual. The distinction is even more ambiguous for directly transmissible outcomes.6
The point here is that the compositional versus contextual dialectic is largely an abstraction of our conceptualization and analytic choices, not an objective quality of the real world. Indeed, the argument that contextual effects warrant community-level interventions is equally artificial. If a community has high levels of individual poor health, an effective intervention could be made at either level, for example by improved individual medical care (compositional) or by modifying the built environment to facilitate healthy lifestyle (contextual). These intervention choices are necessarily determined much more by our values and economic priorities than by our statistical models.
ABOUT THE AUTHOR
JAY S. KAUFMAN is Associate Professor in the Department of Epidemiology, University of North Carolina. His research interests include social epidemiology, minority health, statistical methodology, and health care. Dr. Kaufman serves on the editorial board of Epidemiology and is Editor of Epidemiologic Perspectives and Innovations. He teaches courses in categorical data analysis, epidemiologic methods, and social epidemiology.
1. Stjärne MK, Fritzell J, Ponce de Leon A, et al for the SHEEP Study Group. Neighborhood socioeconomic context, individual income and myocardial infarction: analyses of within- and between-level interactions. Epidemiology
2. Pickett KE, Pearl M. Multilevel analyses of neighbourhood socioeconomic context and health outcomes: a critical review. J Epidemiol Community Health.
3. Merlo J. Multilevel analytical approaches in social epidemiology: measures of health variation compared with traditional measures of association. J Epidemiol Community Health.
4. Oakes JM. The (mis)estimation of neighborhood effects: causal inference for a practicable social epidemiology. Soc Sci Med.
5. Greenland S. Ecologic versus individual-level sources of bias in ecologic estimates of contextual health effects. Int J Epidemiol.
6. Koopman JS, Longini IM Jr. The ecological effects of individual exposures and nonlinear disease dynamics in populations. Am J Public Health.