Determining appropriate group number and composition for data sets containing repeated check cultivars

To simplify the investigation of genotypic performance in large plant breeding trials, clustering techniques can be applied to reduce the number of responses that must be considered. However, the determination of an appropriate number of groups to represent such data is often subjective. Previous work suggested a procedure to determine the appropriate number of groups that should be selected to represent the data. The variability in patterns of performance that could be expected through experimental error at each environment was assessed using repeats of a check line (cultivar). The hierarchical classification was truncated when the repeats of the check line were partitioned into different groups. That procedure is assessed further by application to a data set consisting of five cultivars of sugarcane, each repeated six times within each of three blocks in six environments in south-east Queensland, Australia. In such a data set it is reasonable to assume that five groups, each consisting of all of the repeats of a particular cultivar, would be an appropriate summary of the data. As a more general approach, a method in which the information contained in the partitions, obtained from classifying different combinations of repeats within checks across environments, is used to infer the similarity in patterns of performance of the repeated check lines. The appropriate number of groups to summarize the data is determined from this information. Application of this approach identified the assumed underlying group number and group composition within the data set.