A0341
Title: New software developments for the analysis of ranking data
Authors: Michael Georg Schimek - Medical University of Graz (Austria) [presenting]
Abstract: In various fields of application, lists of distinct objects are presented in rank order because one can always rank objects according to their position on a scale. An observed ordering might be due to a measure of the strength of evidence, an assessment based on expert knowledge, or a technical device. Also, variable values can be replaced by corresponding ranks, but the resulting loss of accuracy is compensated by a gain in generality. The fact that rankings are invariant under the stretching of the scale is a major advantage of this kind of data representation. Various statistical tasks can be performed: (i) measuring the association between ranked lists, (ii) measuring the distance between ranked lists, (iii) identification of significantly overlapping sublists (estimation of the point of degeneration of paired rankings into noise), (iv) aggregation (consolidation) of ranked lists or sublist, and (v) reconstruction of the signals (i.e. estimation of latent parameters) that inform observed ranked lists. The use of the recent R packages on CRAN, TopKLists and TopKSignal, developed by the author and co-workers, is exemplified for selected tasks.
