ABSTRACT
In 1982, when Mason and Entwisle introduced hierarchical linear modeling to demographers and members of other disciplines at the Population Association of America meetings, they titled their paper A Better Way to do Contextual Analysis (Mason & Entwisle, 1982). The Duncan and Raudenbush chapter could equally be titled Better Ways to do Contextual Analysis, perhaps more appropriately so than the name they gave it. In the first half of their chapter, the authors very clearly articulate the “methodological challenges of getting context right.” Many of these challenges are not new to researchers. Clearly, the omitted variables problem is an issue that we all need to address, whether we are modeling contextual effects or not. Nor is the endogenous membership problem new to contextual effects researchers. It is something that we all have been pondering for some time. For example, in a 1993 article by Brooks-Gunn and her colleagues that examined neighborhood influences on child and adolescent outcomes, the authors commented about the selection into context issue as follows: “We are not very sanguine about the likelihood that standard adjustments for selection bias would liberate us from these problems… As do other authors of empirical work on neighborhood effects …we leave the task of modeling selection bias on the agenda of important future research” (Brooks-Gunn, Duncan, Klebanov, & Sealand, 1993, p. 358). What the Duncan and Raudenbush chapter (chap. 8, this volume) is telling us is that in order to obtain, in the authors’ words, “precise, robust, and unbiased estimates of neighborhood effects,” the time has come for us to give greater consideration to these problems and make every attempt to deal with them. What this chapter offers us is an agenda, guidelines, and tools for doing contextual analysis better.
