Small Area and Longitudinal Estimation using Information from Multiple Surveys

点击翻译按钮获取中文摘要

National surveys such as the Current Population Survey (CPS) or the National Crime Victimization Survey give accurate estimates of poverty or criminal victimization at the national level. These surveys do not, however, contain sufficient sample sizes to give reliable estimates by themselves for "small areas" such as counties or minority groups, or to provide detailed information about events such as domestic violence that affect only a small part of the population. Current methods for estimating poverty in small areas incorporate auxiliary information from administrative sources such as tax records through regression. This approach assumes that the administrative data are without errors; it also does not incorporate information from other surveys or make use of longitudinal information.This project focuses on combining information from multiple surveys, with possibly different sampling designs, to improve estimation in small areas. Thus, for estimating poverty in small areas, the CPS can be used in conjunction with the American Community Survey and other sources. Multivariate multi-level models that allow for missing data will be developed that make use of the correlations in the different surveys to increase the precision of the small area estimates, and theoretical properties of the estimates will be derived. Modifications of the multivariate models will allow information from longitudinal surveys to be combined as well, and allow longitudinal analyses from a panel survey to be supplemented by information from related cross-sectional surveys. The investigator will also develop methods for outlier detection and robust estimation using the multiple data sources.Increasing amounts of information are available from surveys and other sources, and there is increasing demand from federal and local governments and from social scientists for estimates in small areas. The models developed in this research will combine information from different surveys and use longitudinal and spatial aspects of the surveys to improve the accuracy of small area estimates, with no additional data collection cost. The theoretical results derived in this research will also be useful for other areas of application such as genetics and quality improvement. This research is supported by the Methodology, Measurement, and Statistics Program and a consortium of federal statistical agencies under the Research on Survey and Statistical Methodology Funding Opportunity.