Decomposition and Bayesian analysis of invariant normal linear models (Q1074970)

In the paper the theory of invariant normal linear models (symmetry models) is supplemented with a Bayesian analysis. The prior distributions are again normal and included via group representation into the underlying model under assumptions, which guarantee that all groups fit well together. It turns then out that for Bayesian inference the original problem can be decomposed into several independent subproblems. Under further assumptions it is shown that the posterior expectation of any parameter is a scalar multiple of its (unique) unbiased estimator. The paper is a very neat contribution concerning the structure of symmetry models. From the given examples with reference to experimental design the reviewer could, however, not detect any help for simplifying the analysis in practical situations.