Interactions With Income

Abstract

The authors respond: We thank Jay Kaufman for his remarks,1,2 and we agree with most of them. However, we do not think that the only reason for doing multilevel studies of contextual risk factors is that they provide knowledge of specific relevance for community-level health interventions. There is still a quest for analytic clarity in understanding the influence of context on the individual. Interventions have their own logic in terms of reversibility of effect, feasibility, cost-effectiveness, and value judgments. Do we need concepts like forests, or would trees do the job of informing us about current ecologic problems? We believe that public health research should take interest in the health effects of emergent characteristics of social systems and that the methods of modern epidemiology should be applied in the search for causal relations. Regarding material context of neighborhoods, it is still premature to state that it operates mainly through the provision of resources in the same way as individual income. The social dislocation of inner-city areas, as analyzed by Wilson,3 points at other possible mechanisms. As middle-income blacks left the areas, there were structural changes in jobs and so on and, according to Wilson, a qualitative change of the neighborhood that exceeded the problem of having low individual income. Area median income must in our view also be considered a proxy measure, as discussed by Diez Roux.4 Accordingly, we are not yet pinpointing the factors actually constituting the biologic interaction, but crosslevel interaction type of research questions might lead us there. Furthermore, as the middle-class mobility patterns in housing respond to contextual changes, they represent both the problem and partly also the solution. Hence, it is important to those societal forces influencing segregation patterns in housing whether all social strata are affected by the health effects of certain contexts and, if so, to what extent. The longstanding controversy regarding the scientific understanding of interaction is interesting, knowing that conditional causation is basic to lay persons' understanding of everyday causality.5 We agree with Kaufman that departure from additivity is the benchmark for evaluating biologic interaction. It is also the relevant criterion when assessing public health importance and in individual risk evaluation.6 All conclusions regarding crosslevel interaction in the paper were therefore based on many analyses similar to those presented in Table 5 (which shows a representative selection of the overall results). Figure 3 was included mainly to illustrate the findings. As pointed out by Kaufman, that figure also incorporates the concept of statistical interaction, which makes the presentation less coherent. We regard our findings of crosslevel interactions as suggestive and they are reported cautiously. However, the research question is important from a public health perspective as well as for deeper insights in causal mechanisms. Johan Hallqvist Johan Fritzell Antonio Ponce De Leon Maria K. Stjärne Department of Public Health Science Karolinska Institutet Stockholm, Sweden Center for Health Equity Studies (CHESS) Stockholm University, Sweden [email protected]