Table 2.
Characteristics of different cases of hypothetical participants showing ground truth values for the total clusters K and the exploration-exploitation ratio, as well as the degree of agreement (Adjusted Rand Index) and the estimated ratio from the clustering analysis
| Sampling Rules | K | Agreement | EER |
|||
|---|---|---|---|---|---|---|
| Ground Truth | Estimated | Difference | ||||
| 1 | Data from fk were uniformly sampled to give a high level of exploration | 4 | 97.52% | 3.06 (2.27, 4.17) | 3.05 (2.21, 4.23) | 0.01 (−0.23, 0.24) |
| 2 | One cluster had a 70% chance of being sampled, the remaining had 10% chance; moderate exploration | 4 | 97.28% | 0.94 (0.64, 1.28) | 0.97 (0.65, 1.37) | −0.04 (−0.16, 0.02) |
| 3 | Two of the four clusters had 80% and 20% chance to be sampled; low exploration | 2 | 96.39% | 0.48 (0.32, 0.67) | 0.51 (0.33, 0.74) | −0.03 (−0.13, 0.01) |
| 4 | Two of the four clusters had equal chance (50%) to be sampled; balance between explore vs. exploit | 2 | 97.69% | 0.99 (0.76, 1.31) | 1.02 (0.76, 1.38) | −0.03 (−0.12, 0.05) |
The index assesses the proportion of time the true clusters and the emerging clusters from the algorithm agree. All values are the average statistics obtained from bootstrapping with 500 repetitions. The range in the brackets is the 95% confidence interval. EER, exploration-exploitation ratio.