Parameter neutral optimum design for non-linear models

Some Bayesian approaches to D-optimum design of experiments are considered from the viewpoint of invariance under reparameterization of the underlying statistical model. An invariant criterion is proposed which does not require the detailed specification of a prior, and which is shown to be equivalent to G-optimality under a Jeffreys prior. The methods are applied and discussed in the contexts of exponential decay and quantal response models.