Student Seminar Series - January 27, 2005
University of Minnesota
School of Statistics
College of Liberal Arts
Computations
for Bayesian Procedures under Linear Constraints in Finite Population
Sampling
Radu C. Lazar
Thursday, January 27, 2005
3:30 PM, 115
Ford Hall
Minneapolis, East Bank Campus
Refreshments at 3:00 PM
300 Ford Hall
Abstract
In survey sampling, one is interested in estimating population
quantities such as the mean or median. Prior information is often
available through the presence of auxiliary variables. For example, the
population mean or median of an auxiliary variable may either be known
exactly or known to lie in some interval. We will present a Bayesian
estimation method which takes into account such prior information which
in some cases cannot be used by the standard estimation procedures.
Various situations are considered wherein the prior information induces
linear constraints on the underlying space of the posterior
distribution, once the sample has been observed. It is shown how one
can estimate the population mean by sampling from such posterior
distributions via a Markov Chain Monte Carlo sampler over such
restricted spaces. The Bayesian estimation method makes use of the
prior information effectively and yields estimates with good
frequentist properties.