Use of population data bases with double sampling for choice-based sample surveys and time-to-event data

点击翻译按钮获取中文摘要

In many research applications based on administrative data bases, certain covariates and proxy measurements are available for all persons in a sampling frame. After sample selection, usually only the precise covariate information collected in the sample is used for data analysis. Ignoring the information in the sampling frame results in an efficiency loss in parameter estimation and has been acknowledged in epidemiology, where two-phase or double sampling methods have been developed to make better use of the combined data. As a consequence, the cost of covariate measurements in surveys can be reduced by double sampling. This project will apply the ideas of double sampling developed in epidemiology for the design and analysis of sample surveys where the sampling frame is based on a population data base. Likelihood oriented approaches and methods based on weighting schemes have been developed to combine the data sources and enhance efficiency. The actual performance of the diverse two-phase methods will be compared in choice-based samples and time-to-event analyses of authentic sociological and epidemiological data. New methods for stratification that build on available phase-one information will be investigated. The goal of the project is the development of empirically based recommendations how phase-one information from large administrative data bases can be efficiently used in the design of two-phase studies for choice-based samples and time-to-event data.