For the binary response model, we determine optimal designs which are robust wit respect to the misspecifications of the unknown parameters. We propose a maximin approach and provide a numerical method to identify the best two point designs for the commonly applied link functions. This method is broadly applicable and can be extended to designs with a given number (>= 2) of support points and further link functions. The results are illustrated for the logistic and probit model, for which the maximin optimal designs are found explicitly.
Binary response model robust optimal design maximin D-optimality Bayesian D-optimality prior distribution