EFFICIENT GEOMETRIC AND UNIFORM DESIGN STRATEGIES FOR SIGMOIDAL REGRESSION MODELS

This paper provides practical guidelines for choosing efficient geometric and uniform designs for the logistic class of dose-response bioassay model functions in both the homoskedastic Gaussian and Binomial settings. The efficiencies of the designs provided here are typically above 90%, and since the number of design support points generally exceeds the number of parameters, these designs provide a useful and efficient means to confirm the assumed model. Extensions of our basic strategy include a Bayesian maxi-min design approach to reflect a range of values of the initial parameter estimates, as well as geometric/uniform design analogues when uncertainty exists as to the correct scale or to take account of curvature.