Tests for Dependence.
Abstract
This paper is designed to provide a sound introduction for a reasonably well-informed reader who is, however, not a specialist in tests for dependence. The paper contains references to many tests but emphasizes the parametric test of independence based on Pearson's sample correlation coefficient r and certain nonparametric tests based on ranks. The ranks tests are generally preferable to the test based on r in that they have wider applicability, are much less sensitive to outlying observations, are exact under mild assumptions which do not require an underlying bivariate normal population, and have good efficiency (power) properties.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 1979
- Accession Number
- ADA069567
Entities
People
- Myles Hollander
Organizations
- Florida State University