Bayes assisted model assessment

点击翻译按钮获取中文摘要

Statistical methods of data analysis rely on probabilistic models which describe the assumed nature of the mechanism which generated the data. The assumptions made are sometimes critical to the validity of the analysis. Statisticians have therefore developed statistical methods to check the assumptions being made; the resulting tools are described as model assessment, model validation or goodness-of-fit techniques. Probabilistic models in current use have become very complex and there is a great need for more such model assessment methods. At the same time the methods already developed are often quite ad hoc -- not based on any overarching principles. Finally the statistical behaviour of model assessment methods can be very hard to compute in modern complex models.