A Test of Goodness of Fit for Multivariate Normality.
Abstract
This paper presents a procedure for testing the composite hypothesis of p-variate normality, 1 < or = P < or = 10, when the parameters are estimated from the data. The test is based on the weighted and integrated modulus squared of discrepancies between sample and population characteristic functions. This measure is shown to be equivalent to the integral of the squared difference between a population density and an empirical estimate. Tables of percentage points are provided. The performances of our density-based or characteristic function-based test statistic, the multivariate skewness, the multivariate kurtosis, a multivariate Shapiro-Wilk, a multivariate Cramer-von Mises, and the multivariate-Anderson-Darling test statistic are compared under several alternatives. The density-based omnibus test is shown to be generally better that tne competitors examined in this study. Keywords: Density estimation; Parametric density estimation; Empirical characteristic function; Composite hypothesis; Multivariate skewness; Multivariate kurtosis. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1986
- Accession Number
- ADA178543
Entities
People
- A. S. Paulson
- H. L. Hwang
- Mark E. Fuller
- P. J. Roohan
Organizations
- Rensselaer Polytechnic Institute