Scholion.ai

A New Family of Multivariate Distributions with Applications to Monte Carlo Studies

Christopher J. Nachtsheim, Mark E. Johnson

Journal of the American Statistical Association · 1988 · Volume 83 · Issue 404

DOI: 10.1080/01621459.1988.10478689

Abstract

We consider a new class of multivariate probability distributions having representation X = RLY ( p ) , where R is distributed as √χ 2 (p) , L is the Choleski factorization of the scaling matrix Σ, and Y ( p ) represents an arbitrary distribution on the p -dimensional unit hypersphere. If Y ( p ) is uniform, then X has a multivariate normal distribution with mean 0 and covariance matrix Σ. The use of classical spherical distributions or other nonuniform distributions for Y ( p ) leads to interesting, controllable departures from normality that are particularly relevant to robustness studies. Their use is illustrated in a Monte Carlo investigation of the robustness of Hotelling's T 2 . Variate generation routines for the cardioid, triangular, offset normal, wrapped normal, wrapped Cauchy, von Mises, power sine, Fisher, and Bingham distributions are developed.

Enable JavaScript for citation tools and additional paper interactions.