Tests of Fit Based on the Correlation Coefficient
Abstract
In testing fit of a sample to a specific family of distributions F, probability-plots use a plot of the order statistics of the sample against a set of suitable constants. These are chosen so that if the parent population is one of the family F, the resulting plot should appear like a straight line. Members of F may contain unknown location and/or scale parameters. Historically, the judgment of a straight line fit was usually made by eye. In this article we propose the correlation coefficient R between the sample values and the constants as a test statistic. Asymptotic properties of R are derived in general, and tables produced to make the test for the case where the proposed distribution is uniform. (kr)
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 04, 1990
- Accession Number
- ADA228178
Entities
People
- M. A. Stephens
- Richard A. Lockhart
Organizations
- Stanford University