A Predictive Approach to Subdomain Estimation in Finite Populations
Petter Laake · Journal of the American Statistical Association · 1979
The problem of estimating finite population characteristics within subdomains is considered. Synthetic estimation has proved to be an important method for deriving such statistics. In constructing the statistics, one combines auxiliary information and sample estimators of the original domain of study. In this article we consider the construction of subdomain statistics as a prediction problem. We establish a superpopulation probability model and derive optimal predictors. With a squared error loss function we show that the predictors compare favorably with the synthetic estimators. A comparison of predictors based on different probability models is also carried out.