Linear Dependence Statistics for High-Dimensional data


This package consists of functions with statistical methods related to the estimation and testing of multiple correlation, partial correlation and regression coefficient matrices when data is high-dimensional (n<p).

Joint estimation of two partial correlation matrices (see wfgl) and joint estimation of two regression coefficient matrices (see wfrl) are currently implemented in the package. These use a weighted-fused lasso penalized maximum likelihood estimator such that they encourage both sparsity and similarity between estimated matrices.

ldstatsHD also contains approaches to select the sparsity tuning parameter of graphical lasso estimators such that several risk functions based on characteristics of the estimated networks are available (see lambdaSelection). Among others, statistics that measure clustering structure or network connectivity can be used to find the desired networks.

It finally includes statistical methods that test global dependence characteristics: (i) a test for equality of two correlation matrices as well as a test for Identity correlation matrix (see eqCorrMatTest); (ii) a test for equality of two correlation matrix rows as well as a test to check if a variable is linearly independent of the rest of the variables in a dataset (see eqCorTestByRows).

A particularity of the implemented methods in ldstatsHD is that it permits cases where datasets are dependent (e.g. paired data).

ldstatsHD provides two partial correlation matrix simulators such that all methods can be tested using using simulated data: see pcorSimulator to generate a single partial correlation / dataset and pcorSimulatorJoint to generate a joint partial correlation matrix and two (dependent) datasets.


Package: ldstatsHD
Type: Package
Version: 1.0.1
Date: 2016-07-08
License: GPL-2
LazyLoad: yes


Caballe, Adria <>, Natalia Bochkina and Claus Mayer.


To come

Want to suggest features or report bugs for Use the GitHub issue tracker.