Distance based methods for exploratory data analysis

Abstract

Summary form only given as follows: Research and the corresponding aspects of data analysis can be broken into two parts, one exploratory and one confirmatory. In exploratory data analysis one tries to narrow down potential hypotheses for subsequent studies. Examples of these can include screening drugs for potential use in cancer treatment using in-vitro tests or screening monoclonal antibodies for use in disease identification. In each of these cases, there are way too many "treatments" for traditional hypothesis testing. Exploratory tools provide a means of reducing the number of treatments for subsequent evaluation. Here, the authors examine the use of distance based methods for exploratory data analysis. Such techniques include cluster analysis and multidimensional scaling. These techniques can be used to group observations and detect outliers. The authors also describe methods for comparing the results of 2 or more cluster analyses or 2 or more ordinations using multidimensional scaling analyses. Examples from a variety of medical and biological applications are included.<>