Partial and multiple Bayes factors are introduced to obtain pairwise c
omparisons of hypotheses in a statistical experiment with a partition
on the parameter space. Robust Bayesian analyses are performed by intr
oducing suitable classes of priors and by calculating lower and upper
bounds of Bayes factors and posterior probabilities. Classes of intuit
ively meaningful priors are introduced, including unimodal densities w
ithout the constraint of symmetry for the case of precise hypotheses.
Procedures for the corresponding optimizations are specified, and exam
ples are given.