An analysis of some inference procedures derived via relative surprise

This thesis is concerned with the concept of relative surprise as a technique for deriving inferences. Further refinements in the development of relative surprise are made beyond those presented in Evans (1997). Relative surprise is applied to the normal location-scale model and the properties of the inferences derived are studied. Similarly relative surprise is applied to the one factor random effects model and inferences are derived for the intraclass correlation coefficient. The thesis develops a technique for using the distribution of the cross-validational observed relative surprise under all possible splits for model checking, hyperparameter estimation and model selection inferences.