Constrained noninformative priors in risk assessment

Abstract A constrained noninformative prior distribution, a generalization of the Jeffreys noninformative prior, is defined for a single unknown parameter as the distribution corresponding to the maximum entropy distribution, subject to the assumed constraint(s), in the transformed model where the unknown parameter is approximately a location parameter. This note illustrates this idea with binomial and Poisson data models, and gives an example from risk assessment showing the practical usefulness of the constrained noninformative prior.