Clustering pattern consistency of corn cultivars

The objective of this research was to evaluate the clustering pattern consistency obtained from the combination of the two dissimilarity measures and four clustering methods, in scenarios consist of combinations number of cultivars and number of variables, with real data in corn cultivars (Zea mays L.) and simulated data. We used real data from five variables measured in 69 trials involving corn cultivars, the number of cultivars ranged between 9 and 40. In order to investigate the results with more cultivars and variables, were simulated under the standard normal distribution, 1,000 experiments for each of the 54 scenarios formed by the combination among the number of cultivars (20, 30, 40, 50, 60, 70, 80, 90 and 100) and the number of variables (5, 6, 7, 8, 9 and 10). Analyses of correlation, diagnoses of multicollinearity ans cluster were carried out. Clustering pattern consistency was evaluated by the cophenetic correlation coefficient. There is a decrease of clustering pattern consistency with the increase in the number of cultivars and variable. The euclidean distance provides greater clustering pattern consistency in relation to Manhattan distance. The clustering pattern consistency among the methods increases as follows: Ward's, complete linkage, single linkage and average linkage between groups.