Sofia Normark: Efficiency of locally optimal designs for binary response models

Optimal designs for binary response models often depend on unknown parameters. Estimated values of the parameters may be used to obtain a locally optimal design, but if the guessed values are inaccurate, the resulting design may be inefficient. On the other hand, if the guesses are close to correct, the design may be “close to optimal”. This paper examines the efficiency loss of locally optimal designs when the guesses of the parameters are incorrect by approximating the determinant of the information matrix by a Taylor function, which in turn is used to approximate the efficiency. In addition, a characterization of locally optimal two-point designs is given.

(Work with Hans Nyquist.)