Useable Robust Statistical Modeling Inference

RES570 – NIH “Robust II”

Data arising in medical, biomedical and biotechnical research and analysis often contain outliers, in different forms and from diverse causes. The outliers may for example occur independently or in sub-population groups, and may be due to extraordinary responses of individuals or sub-groups of individuals in a study, or be due to sporadic instrument or recording errors. Whatever the source and form of outliners, they can and often do have very adverse effects on classical statistical modeling methods such as least squares fitting of linear models, analysis of variance, logistic regression, generalized linear models, survival analysis, and covariance matrix estimation.

A primary goal of the research is to provide theoretical justifications and software for a very broad range of robust modeling and analysis methods which provide a good fit to the bulk of the data in the presence of outliers, enable rapid identification of outliers, and provide good statistical inference results. The software will be implemented in the S-PLUS object-oriented environment for data analysis, statistical modeling and graphics, and will emphasize ease of use for both “power” users who are comfortable with command line use of an object-oriented language, and for a very broad range of users who require a well-designed graphical user interface (GUI).