Testing the Correlation Coefficient with Incomplete Observations.
Abstract
Correlation is often investigated (and tested for significance) in situations where some of the observations on one of the variables are missing. Throwing away these unpaired observations may seem to be a waste of information; a test based on all the data at hand would seemingly be better than a test based on only some of the data available. An exact test using all the data, which is similar in form and distribution to the usual t test based on the sample correlation coefficient, is derived and examined. However, this exact test proves to be a relatively inefficient way to incorporate the extra information. This counterintuitive result provides an interesting lesson concerning the realtionship between power and degrees of freedom. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 1981
- Accession Number
- ADA109548
Entities
People
- James Beckett
- N. J. Bosmia
- William R. Schucany
Organizations
- Southern Methodist University