Journal of Official Statistics, Vol.22, No.3, 2006. pp. 541–563
Alternative Designs for Regression Estimation
Abstract:Restricted random samples partially balanced on sample moments are compared with a two-per-stratum design as designs for use with regression estimation. The shape of the average weight for the best linear unbiased predictor as a function of the auxiliary variable is similar to the shape of the inclusion probability for a sample partially balanced on the auxiliary variable. In the simulation study the MSE of the regression estimator with the stratified random sample is comparable to the regression estimator with the balanced sample when the assumed linear model holds. When estimating the population cumulative distribution function at selected points, two-per-stratum stratified random sampling shows better performance than partially balanced samples.
Keywords:Best linear unbiased predictor, regression estimator, regression superpopulation model, balanced sampling
Copyright © Statistics Sweden 1996-2018. Open AccessISSN 0282-423XCreated and Maintained by OKS Group