Small Area Estimation by Combining Information from Multiple Data Sources on Correlated Variables at Different Levels of Aggregation


Small Area Estimation with Data from Multiple Sources

Author: Trivellore Raghunathan (Survey Research Center, Institute for Social Research; School of Public Health, University of Michigan; Joint Program in Survey Methodology, University of Maryland)

Demands for small area estimates are ever increasing and are useful for the local policy evaluation and implementation. Increasing concerns about privacy and confidentiality is preventing agencies from providing data at the desired level of geography. This paper develops procedures for combining information from multiple data sources that provide data at different levels of aggregation on correlated variables. The levels of aggregation may be nested (such as counties within States and States) or non-nested (Standard Error Computing Units, Census Tracts, Counties, Zip code etc.). The procedures are motivated from a Bayesian perspective and using the missing data framework. Several examples are used as illustration and the sampling properties of the small area estimates are evaluated using simulated data sets. The bias and loss of efficiency are compared to the estimates derived from the ideal and the same level of aggregation from every data source.