Statistical Software for Data with Measurement Error

DESCRIPTION (provided by applicant): The ultimate objective of this research application is the development of S+ME: a next generation software for handling covariate measurement error. The developed methodologies are also applicable to data with missing covariates. To achieve this, much of our research effort will be devoted towards combinging algorithms in regression calibration, SIMEX and iterative imputation methods into a software package. These methodologies will be matured by incorporating diagnostic techniques, sensitivity analyses, and graphical methods into the software package. In the Phase I feasibility study, algorithms have been developed for univariate Gaussian and logistic regression. These methods will be extended to covariate measurement error in modeling correlated responses such as in longitudinal or spatial data. The result of this research will make fundamental contributions to the conduct of public health studies by developing mature methodology and user-friendly software for handling missing or mismeasured covariates. Currently, there is little or no commercial software in this area. The aim of the proposed research is to overcome this deficiency and bring the benefits of measurement error models to a wide audience of biomedical analysts and practitioners. The S+ME module will be implemented as an object-oriented software in the S-Plus language. A comprehensive case study guidebook will be developed involving real problems in exposure and risk factors which inherently contain measurement error. PROPOSED COMMERCIAL APPLICATION: NOT AVAILABLE