Using Prediction-Oriented Software for Survey Estimation - Part III: Full-Scale Study of Variance and Bias

by James R. Knaub, Jr .

Abstract: Applications of this method include small area estimation and imputation, accompanied by estimates of standard errors. This method can be used for imputation with any kind of sample or census survey. It has been tested and developed, and now enters full-scale testing and implementation. Advantages include ease in revising models, flexible organization, storage and usage of data, and the ability to maximize the effectiveness of collected data. For purposes of estimation, collected data may be grouped such that each data set contains as many members as may be well defined under a single model per group. That is, each group should be as large as it can be and remain basically homogeneous. After the models are exercised, there will either be an observed response or an 'imputed' value for each member of the population, for each data element, which can be rearranged and published, with estimated standard errors, for any subtotals desired. Under full-scale testing, more results have become available for a better study of variance estimation, and bias is also studied with instructive results. Other areas illustrated are the degree to which use of this technique may be appropriate under an extreme condition, and the application of this method across strata.

Key Words: Small Area Estimation, Imputation, Standard Error, Bias, Inference, Model-Based Sampling

Author:
James R. Knaub, Jr. , James.Knaub@eia.doe.gov

Editor: Richard Graf , rgraf@sunstroke.sdsu.edu

READING THE ARTICLE: You can read the article in portable document (.pdf) format (130055 bytes.)

NOTE: The content of this article is the intellectual property of the authors, who retains all rights to future publication.

This page has been accessed 3027 times since July 24, 2006.


Return to the InterStat Home Page.