Cluster analysis in a randomized complete block design

The mixture maximum likelihood approach to clustering is used to allocate treatments from a randomized complete block design into relatively homogeneous groups. The implementation of this approach is straightforward for fixed but not random block effects. The density function in each underlying group is assumed to be normal and clustering is performed on the basis of the estimated posterior probabilities of group membership. A test based on the log likelihood under the mixture model can be used to assess the actual number of groups present. The technique is demonstrated by using it to cluster data from a randomized complete block experiment.