A Small Sample Evaluation of a Bayesian Design Method for Quantal Response Model
Abstract
The author proposes a design measure which may be used sequentially to choose the next dose level in a linear logistic quantal response model for bioassay. His design measure averages the posterior distribution of the effective dose over those ED(effect dose) values which are regarded as important. In this evaluation the mode of the design density is used as the next design point, and it is supposed that all ED values between ED 60 and ED 90 are equally important. After ten initial badly designed observations, it is shown that only 20 further, well designed, observations are needed to obtain a design efficiency of about 82%, and an estimated response curve which lies at a maximum of an estimated ED points from the true curve, for all ED values lying between ED 60 and ED 90. If more observations are taken then the design efficiency increases steadily, but it is difficult to increase the accuracy of estimation without either taking many more observations, or by pushing the design points outside the appropriate region. However, within the design region, chosen by any recommended procedure, the method promises excellent robustness, with respect to possible inadequacies in the model, whilst outlying design points would not provide such robustness.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 1983
- Accession Number
- ADA127942
Entities
People
- Tom Leonard
Organizations
- University of Wisconsin–Madison