Non-Linear Multidimensional Models, Least Squares and Reduced Residuals,
Abstract
This paper addresses the treatment of such model fitting problems that include an analysis of the residuals, for instance to determine heteroscedasticity. In cases with multidimensional observations the residuals are vectors having model induced correlations between components. It is shown that one can eliminate such correlations and also reduce the number of residual components by introducing a concept of reduced residuals. The application of the new concept is illustrated by an investigation of heteroscedasticity of vapor pressure measurements. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 1984
- Accession Number
- ADP002950
Entities
People
- A. Celmins
Organizations
- Ballistic Research Laboratory