News

Alex Z. Fu, PhD discusses results of a study finding that Statins Are Underused in Type 2 Diabetics in Renal and Urology News.

Xiaofeng Wang, Ph.D. et al proposed testing methods to compare nonparametric surfaces. They considered a test statistic by means of an L2-distance under a completely heteroscedastic multivariate nonparametric model.  They also extended the test statistic for use in the case of spatial correlated errors. Two bootstrap procedures were described in order to approximate the critical values of the test depending on the nature of random errors. Both theoretical properties and numerical properties of the proposed methods were investigated. The resulting algorithms and analyses were illustrated with a real medical image study.  Full details of the study can be found here.

Classification and regression trees are non-parametric statistical models that are well suited to handle non-linear effects and interactions between variables.  Random forests (RF) is a powerful extension of this method that uses a two-stage randomization procedure to derive a forest of trees.  RF exhibits low variance and bias, is not prone to over-fitting, does not assume linear effects, and is able to model all interactions between variables.  However, despite these favorable attributes of RF, selecting variables is not straightforward.  To address this, Hemant Ishwaran, Ph.D. et al. developed an order statistic for trees that measures the predictiveness of a variable.  This statistic is called the minimal depth of a maximal subtree and measures the depth of a variable from the root node of the largest subtree that splits the variable.  The distribution theory for this new statistic as well as the behavior of key parameters has been worked out in detail.  The method can be applied in general settings and can be used for variable selection in genomics when the number of variables can be substantially larger than the sample size. As an example, the methodology has been used to predict metastasis for early stage breast cancer patients.   Future work will include constructing gene regulatory networks.  An R-software package is available here . The full study is available here .