A Bayesian approach to parallel line bioassay

This thesis considers parallel line bioassay from a Bayesian point of view along the lines laid about Lindley (1972) and de Finetti (1975). The mathematical model used for the analysis is a non-linear one in which the log potency appears explicitly as a parameter. This enables prior knowledge about the log potency ratio to be incorporated straightforwardly in the analysis. The method of analysis follows closely the ideas of Lindley and Smith (1972) for the linear model. Extended models in which experimental design features such as randomized blocks and Latin squares are accounted for are also considered, and a method for the use of prior information to design an assay is given. In addition to the analysis of a single essay the problem of combining information from several assays is considered and two different models which combine such information are discussed.