Abstract
Population-based surveys are of limited utility to estimate rare or low-incidence groups, particularly for those defined by religion or ethnicity not included in the U.S. Census. Methods of cross-survey analysis and small area estimation, however, can be used to provide reliable estimates of such low-incidence groups. To illustrate these methods, data from 50 national surveys are combined to examine the Jewish population in the United States. Hierarchical models are used to examine clustering of respondents within surveys and geographic regions. Bayesian analyses with Monte Carlo simulations are used to obtain pooled, state-level estimates poststratified by sex, race, education, and age to obtain certainty intervals about the estimates. This cross-survey approach provides a useful and practical analytic framework that can be generalized both to more extensive study of religion in the United States and to other social science problems in which single data sources are insufficient for reliable statistical inference.