Bayesian clustering of data sets

Binder's (1978) Bayesian clustering technique is extended to cover clustering of data sets, i.e., all observations in the samedata set are known a priori to fall into the same cluster. Further-more, two methods are suggested to examine the importance of various subsets of the variables used for clustering. The technique is applied to a situation where usually one-way multivariate analysis of variance is used and gives additional information about those treatments between which there are no significant differences. A simple example is analyzed extensively.