We propose multiple-objective Bayesian optimal designs for the logit model.
As an example, we consider the design problem for estimating several perce
ntiles with possibly unequal interest in each of the percentiles. Character
istics of these designs are studied and illustrated fbr the case when the i
nterest lies in estimating the three quartiles. We compare these optimal de
signs with the sequential designs generated via a generalized Polya urn mod
el and found the latter to be highly efficient. In addition, comparisons ar
e made between locally optimal designs and Bayesian optimal designs. Copyri
ght (C) 2001 John Wiley & Sons, Ltd.