|
Abstract:
|
We investigate adaptive designs with a trinomial response and a single quantitative factor. Depending on the level of the factor the response may be classified as substandard, satisfactory or defective. The probabilities of these responses depend on logistic models with unknown parameters. One constraint is on the probability of defective product. This can define the experimental region. Given that this constraint is met, the optimum level of the factor should maximizes a function of the other two probabilities. We compare design approaches, including penalized D-optimality, with those developed in Phase I clinical trials for estimating a specific objective function, such as the 'biologically optimum dose.' Depending on the purpose of the experiment and the physical conditions, different strategies should be recommended. We discuss the pros and cons of the approaches.
|
