EDF Tests for Normality in Linear Models After A Box-Cox Transformation
Abstract
The Box-Cox transformation procedure has been used extensively in data analysis, for example in regression, where the response variable is subjected to a suitable power transformation so that the standard normal-theory linear regression models can be fitted to the transformed values. In this paper, distribution theory is developed for a family of EDF statistics, including the Anderson-Darling statistic A2 and the Cramer-von Mises statistic W2, so that these statistics can be used to test for normality in the linear model after applying the Box-Cox transformation. A table of asymptotic critical points is given for A2 and W2 and numerical examples are given to illustrate the use of the table.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 02, 1993
- Accession Number
- ADA269068
Entities
People
- Gemai Chen
- Michael A. Stephens
- Richard Lockhart
Organizations
- Stanford University