COMPARISON OF DESIGNS FOR DISCRIMINATING BETWEEN REGRESSION MODELS.
Abstract
Several sequential design procedures for discriminating between two regression models have been suggested in recent years. The essence of the results is that the sequential design is essentially independent of which model is correct. This result is investigated in this study for the fixed sample size design and for the simple case of a linear versus a quadratic model, both through the origin, as this is the only case for which the sequential designs have been tabulated. The results here are substantially in agreement with those for sequential designs as the model dependence of the design allocations are not large by practical considerations. The approach suggested offers the possibility of designing a preliminary experiment for the sequential test procedure. Of course, in physical situations where the sequential design approach is not practical, the fixed sample approach will be necessary. Two criteria are considered in the selection of a design (1) an optimal design using a composite model and a minimal confidence region, and (2) an optimal design using the lack-of-fit test as given in the literature. The first approach is preferred for estimation of unknown parameters whereas the second is preferred for discrimination between the hypothesized models. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 1968
- Accession Number
- AD0666788
Entities
People
- A. C. Nelson Jr.
Organizations
- RTI International