Medical data often require complex models. For example, clinical monitoring produces time induced correlation, and relationships among variables change due to medical intervention. Many popularly used biostatistical procedures depend on approximations made for mathematical tractability. Resampling methods are computationally intensive techniques which approximate the distribution of a statistic using only the observed data. The two-fold advantages of resampling methods are that (1) they are conceptually simple and (2) they often apply in complex problems inaccessible through other techniques. Phase I research will focus on designing software in the S-PLUS environment in a way that facilitates use of resampling methods. We plan to achieve this through judicious use of presentation tools, and through a flexible software design. This design will support the different needs of (1) data analysts and (2) biostatistical researchers who want to modify and extend resampling capabilities. In Phase I we will prototype the design, focusing on the linear model. Resampling methods offer the opportunity to revolutionize statistical practice: making it easier to use and understand, yet applicable to complex problems.
Implementation of resampling methods in a modern interactive statistical language and graphics system as S-PLUS will find a ready market in all areas of statistical application.
