June 2005 Issue

摘要

The Applications and Case Studies section of the June issue of JASA contains four papers that apply hierarchical modeling techniques to a diversity of problems. Scott, James, and Sugar employ hidden Markov models to examine the effectiveness of several medications for schizophrenia. The model developed incorporates information from observations taken over time in forming clusters of "health states." Elliott and Little show how demographic information can be used directly in adjustment of Census counts. Currently, postenumeration surveys are used to adjust the Census under unrealistic assumptions. The model used by Elliott and Little produces more stable estimates for subpopulations and accounts for modeled sources of uncertainty. Dorazio and Royle consider estimating species occurrence from data collected in the North American Breeding Bird Survey. Key difficulties with these data are that detection of species at a given location is imperfect and the actual occurrence of species varies across locations. Thus, one needs to estimate both rates of detection and rates of occurrence—a difficult task. Finally, Iversen, Jr. and Chen consider the problem of controlling for sampling bias in the characterization of rare disease genes. This problem concerns the phenomenon in which higher risk individuals are ascertained in greater proportion than their representation in the population, leading to "size-biased" estimates. In particular, they estimate age at onset in individuals with breast and ovarian cancer genotypes, using an "ascertainment"-corrected likelihood with data from high-risk individuals.