Posterior Mapping and Posterior Predictive Distributions

Abstract

" If we view statistics as a discipline in the service of science, and science as being an attempt to understand (i.e., model) the world around us, then the ability to reveal sensitivity of conclusions from fixed data to various model specifications, all of which are scientifically acceptable, is equivalent to the ability to reveal boundaries of scientific uncertainty. When sharp conclusions are not possible without obtaining more information, whether it be more data, new theory, or deeper understanding of existing data and theory, then it must be scientifically valuable and appropriate to expose this sensitivity and thereby direct efforts to seek the particular information needed to sharpen conclusions. " Rubin [38]