Factorial simulations Sample Clauses

Factorial simulations. Correct choice of highest-ARI partition. The percent of best partitions correctly chosen by the global minimum and maximum of the index, the min- imum and maximum difference to the left, and the minimum and maximum second difference are shown in Table 4.1. The maximum second difference of the Xu index and XuSAD index performed the best, choos- ing 62.2% and 62.9% of the best partitions respectively. The minimum second difference of the Calinski- Harabacz index performed next best overall, choosing 61.0% of the best partitions. The maximum second difference of the SSW and SAD chose 56.7% and 59.9% of best partitions, respectively. The two versions of the Xxxxxx-Xxxxxxx index chose correctly for more than 50% of partitions. The maximum second difference of the Xxxxxx-Xxxxxxx (SSW) chose 51.4% correctly, and the global minimum of the Xxxxxx-Xxxxxxx (SAD) chose 56.1% correctly. Several best index/usage pairings agreed with the recommendations of Dimitriadou et al. The maximum second difference chose the best-ARI partition most often for the SSW, Xxxx-Xxxx, and Xu indexes, and the minimum second difference chose the best-ARI partition the most often for the Calinski-Harabasz index. For the Xxxxxx-Xxxxxxx, Xxxxx-Xxxxxx, and Marriott indexes, the usage recommended by Dimitriadou et al. (2002) was not the best index/usage combination in our study, but chose correctly almost as well as the best combination. This probably reflects the relationships between the statistics in “picking up” certain shapes. For example, the maximum difference to the left and the minimum second difference tend to indicate an elbow that is convex upward. The best usages of the Xxxxxxxx, Ratkowsky-Lance, and Xxxxxxxx-Xxxxx indices in our study did not agree with recommendations by Dimitriadou et al. (2002).