Title: The method of tail functions for confidence estimation
Authors: Borek Puza - ANU (Australia) [presenting]
Abstract: The aim is to describe how tail functions can be used to construct confidence intervals with attractive properties. The theory is applied to several scenarios, including inference on the binomial proportion and normal mean. In each case, the standard interval is generalised using a class of tail functions and the optimal interval is then engineered, taking into account one or more criteria, such as the length of the interval, its maximum proportional length, or its expected length under a prior. The relationship between this approach, the Bayesian, and some others is also discussed.