Statistical Methods for Incomplete Data with Measurement Errors

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Missing data and measurement errors are common problems in statistical data analysis. We are interested in experimental and observational studies where there exist missing data and measurement errors problems. Examples include health surveys containing non-responders or missing items, surrogate marker data with measurement errors, etc. The applications could be longitudinal clinical trials, multilevel community studies and health surveys. The incomplete data could be the non-ignorable missing response used in a model or as predictors, i.e. missing response, missing covariate, and covariate measurement errors. The most complicated scenario is the combination of such difficulties, i.e. the missing response with covariate measurement errors. The results from this project include innovative statistical methods, case studies, tools, solutions, and publications. These resources will be incorporated in our Longit Informatics Center for sharing and illustration. The Longit Informatics Center is an online data analysis environment. Subscribers can access many statistical packages and dynamic graphics for data analysis. In this project, the ultimate results will be two statistical packages added to Longit: 1) MiMe: statistical methods for missing data and measurement errors, and 2) Laso: joint modeling methods for longitudinal and survival outcomes in the study of surrogate marker for clinical event time. These packages include innovative statistical methods, sensitivity analysis and graphical methods. There is no commercial software to deal with complicated case as Laso. PUBLIC HEALTH RELEVANCE: This project aims to develop statistical methods and tools for analyzing incomplete data with missing data and measurement errors.