Statistical Analysis with Computerized Linked Data

点击翻译按钮获取中文摘要

This research project will develop a new theoretical framework for analyzing data obtained from multiple sources to solve important small area estimation problems. It would be prohibitively expensive to produce statistics on spatial and temporal granularity using traditional methods that rely solely on sample survey data. Large databases such as administrative or census records in combination with sample surveys offer potential solutions to these sparse data problems, but methodological questions arise when these alternative data sources are brought together to produce small area statistics. This project will address these methodological questions by developing a framework that extracts the maximum possible information from multiple sources of data. The project will enable formulation and analysis of complex and challenging questions of scientific interest and societal importance. The research results will impact the definition of new data structures and models and software tools and strategies for analyzing complex computerized data settings, such as social statistics and business. Graduate student research will be supported by this project. To further the dissemination of research results and support the training needs of current and future survey statisticians, the investigator will offer workshops/seminars/webinars in the Washington, D.C. area.Government agencies need timely and finer-grain data to effectively plan and evaluate different government programs for the public good. People generally are more concerned about statistics for their own community, such as the crime rate in their neighborhood last week, than statistics for the entire nation in the last year. The investigator will develop a new general integrated model that combines a permutation linkage model with a small area model to extract maximum possible information from multiple sources of data available at different hierarchical levels. The model will be implemented using a frequentist approach. The investigator will explore the theoretical properties of empirical best predictors and the jackknife estimator of uncertainly under the integrated model. The methodology to be developed will be evaluated by extensive simulations and real data applications. During this project, the investigator will address a number of challenging issues in the presence of certain linkage errors, including modelling, model diagnostics and model selection, measuring uncertainty of the proposed estimators, and evaluation.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.