|
Abstract:
|
Epidemiology has long had an interest in the effect of neighborhood environments on health outcomes. In recent years, a fervor has erupted as investigators have aimed to disentangle the effects of such environments from the background of residents. Hundreds of papers have been published, but little of this work has sustained methodologic examination. Drawing from a recent paper (Oakes. 2004. Soc Sci Med 58:1929-52), this talk discusses fundamental problems in the identification of meaningful neighborhood effect parameters. I show that except in special cases where sampling mechanisms yield exchangeable subjects, none of the published estimates merits attention. The fervor is a case of "statisticism," a term used to describe a ritualistic appeal to significance testing and sampling error when such issues are a relatively minor concern.
|
