Compound optimal designs for percentile estimation in dose-response models with restricted design intervals

In dose-response studies, the dose range is often restricted due to ethics concerns over drug toxicity and/or efficacy, particularly when human subjects are involved. We present locally optimal designs for the estimation of several percentiles simultaneously on restricted as well as unrestricted design intervals. Our results are applicable to most of the commonly applied link functions with respect to the model under consideration. This work is a generalization of Dai [2000. Compound optimal designs. Ph. D. Dissertation, State University of New York, Stony Brook] where he showed that the same results hold for the logit model using Elfving's [1952. Optimum allocation in linear regression theory. Ann. Math. Statist. 23, 255-262] approach on trace optimal design.