Abstract:
|
In the study of optimal designs for logistic regression models, it is known that the solutions actually depend on the unknown true values of the parameters, as is typical of nonlinear models. It is generally assumed that close approximations to these parameters are known, either from previous experimentation or from pilot studies. The problem with this approach is that in practice, the guess values are likely to differ from the true parameter values. So essentially, the design that we implement in practice is merely a pseudo optimal design. In this work, we assess whether we can circumvent this problem in optimality from ill-guessed parameter values by implementing a 2-stage adaptive optimal design. Furthermore, for such 2-stage designs, we examine the nature of optimal allocation of resources at each stage. We also study how the 2-stage adaptive optimal design compares to the single-stage pseudo optimal design depending on the departure of the guess values from the true values of the parameters. We restrict our study to the well-known A- and D-optimality criteria.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.