Hierarchical Modeling : Into Statistical Practice

Hierarchical Modeling: Into Statistical Practice Alan M. Zaslavsky Harvard Medical School, Boston, Massachusetts, USA Pardoe, Weidner, and Friese (henceforth PWF) present a nice application of hierarchical modeling, representative of current practice of this methodology. Hierarchical regression modeling now occupies a methodological middle ground. The main principles have been established, as have a fairly general set of computational algorithms for a reasonably general set of models, implemented (in a more or less general manner) in software in some of the major statistical packages. The use of these models is unevenly diffusing across fields of application, scientific journals, and individual researchers and statistical practitioners. Thus an approach that appears novel and challenging in one context might appear routine in another. I am a firm believer in hierarchical modeling as a uniquely effective framework for coherently describing complex social phenomena and for building complex models from simple pieces; on the other hand, as a working applied statistician I have learned to be cautious about treating them as a panacea for dealing with the complexities of data. In this spirit I offer a miscellany of remarks suggested by PWF’s article.