One view of the development of the EB approach is that it is an attractive compromise between the classical non-Bayes and the full Bayes approaches to statistical inference. These represent extremes in that the former uses no prior information whereas the latter requires complete specification of a prior distribution. The EB approach uses previous data to get an estimate of the prior distribution. The previous data and current data are linked in the form of a two-stage sampling scheme by a common prior distribution G of the unknown parameters; see section 1.8.