Prior Elicitation: A Compromise between Idealism and Pragmatism

摘要

The problem of specifying subjective prior distributions for model parameters is not easily resolved. We review strategies for selecting families of priors and propose methods to evaluate the corresponding hyperparameters. Specifically, we emphasise the natural beauty and philosophical benefits of predictive elicitation using quantiles and cumulative probabilities. However, we acknowledge the inherent practical difficulties relating to constraints and computation, concluding that some compromise is necessary. These ideas also apply when we need to elicit statistical information from subject matter experts, including suitable distri-butions for inputs into a complex computer model.