9312931 Gelfand The subject matter of this conference is Bayesian Methods in Finite Population Sampling: Theory and Applications. Bayesian techniques have found applications in virtually every branch of statistics, and have proved to be useful in the analysis and interpretation of a wide range of data. Bayesian analysis which utilizes prior information in addition to the information from the collected data, is particularly well-suited in finite population sampling. In most surveys, there is some prior information available about the units of the population to be sampled. For example, for estimating the yield of a certain crop based on samples from a large number of agricultural units, information is usually available regarding the area devoted to that particular crop in all these units. Bayesian methods enable one to incorporate this prior information formally in the form of a prior distribution, and produce estimates in a fairly straight forward manner. In addition, there is a growing need to produce reliable small-area estimates, that is estimates for counties, municipalities, and other local areas. Classical sample survey methods are usually designed to provide effective estimates at a much higher level of aggregation, for example at the state level. But standard estimation based only upon data for a given local area typically results in unacceptably large standard errors due to small sample size within a local area. Bayesian methods enable borrowing strength, or use information from similar areas or sources to improve accuracy of individual local area estimates. This is used for a wide range of survey data being collected by both the public and the private sectors. The Conference will provide a mechanism for synthesizing the substantial and diverse literature in the general area of Bayesian finite population sampling, and will illustrate the methods with a number of real life applications. ***
