Comparison of Clustering Methods over 10,000 Simulated Datasets.

Five clustering methods are compared in two simulation studies. Studies are grouped by row, so that the same 10,000 simulated datasets were used to evaluate default MMC and the four remaining methods with . A second set of 10,000 datasets was used to compare the methods with . Columns 2–4 report three related measures of performance. Column 2 considers only pairs of variables that share a cluster according to Figure 2 (e.g. (1, 2) but not (1, 3)) and records the percentage of such pairs that are correctly clustered together. Column 3 considers only pairs of variables that do not share a cluster and records the percentage of such pairs that are correctly placed in separate clusters. Column 4 considers all variable pairs and records the overall percentage correct. These measures are used in Figure 4 as well.