The parameter-effects curvature measure proposed by Bates and Watts (1980) is examined for a growth model and the Fieller-Creasy problem. Exact confidence regions are constructed and compared to linear approximation regions. For the growth model the agreement between the regions is good despite high curvature. In the Fieller—Creasy problem it is shown that the agreement can be quite poor despite low curvature.
Bibliographical noteFunding Information:
* R. Dennis Cook is Professor and Chair, Department of Applied Statistics, University of Minnesota, St. Paul, MN 55108. Jeffrey A. Witmer is Assistant Professor, Department of Statistics, University of Florida, Gainesville. FL 3261 1. Research for this article was supported in part by U. S. Army Contract DAAG29-80-C00041 and the Monsanto Company. The authors thank Ken Portier for computational assistance and the referees for many useful sugges- tions.
- Exact confidence regions
- Fieller–Creasy problem
- Nonlinear regression
- Parameter transformations