CONFIDENCE REGIONS FOR TIlE PARAMETERS OF A

Three methods in common use for finding confidence regions for the parameters of a nonlinear regression model are studied. These are the lack of fit method, the linearization method, and the maximum likelihood method. The objective is to determine which of the three methods is the preferred choice in applications. The conclusion is that the linearization method has the best structural characteristics but its nominal confidence level is apt to be inaccurate. A choice of the likelihood ratio method is a compromise between structural characteristics and accuracy. The lack of fit method would only be chosen if accuracy was of extreme concern.