Distance correlation is a powerful new dependence measure that was recently introduced by Székely, Rizzo and Bakirov (2007) and Székely and Rizzo (2009). Different from classical dependence measures such as Pearson or Spearman correlation, the distance correlation coefficient characterizes independence and can hence be applied for detecting arbitrary associations between multivariate datasets.
In this talk, we present the new R package dcortools, which provides easy-to-use distance correlation functions for statistical applications. Special emphasis was put on the computational efficiency of the package, enabling fast independence testing and variable screening in high-dimensional data. Moreover, its flexibility allows to run numerous extensions and generalizations of distance covariance and distance correlation, using only a single function. Visual interactive tools make it possible to explore the dependence structure of large datasets.
Various simulation and real data examples demonstrate the broad applicability of the new R package and show how it can be helpful in obtaining a better understanding for complex dependence structures.
|