Parameters used in generating simulated data
| Model number | Data distribution | Case | β | Number of clusters | Average cluster size | Range of cluster sizes | Number of observations |
|---|---|---|---|---|---|---|---|
| 1 | Normal | 0 | 20 | 10 | 5–16 | 200 | |
| 2 | Normal | 1 | 1 | 20 | 10 | 5–16 | 200 |
| 2 | 0.2 | ||||||
| 3 | Skewed | 0 | 20 | 10 | 5–16 | 200 | |
| 4 | Skewed | 1 | 1 | 20 | 10 | 5–16 | 200 |
| 2 | 0.3 | ||||||
| 5 | Normal | 0 | 8 | 6 | 3–9 | 48 | |
| 6 | Normal | 1 | 2 | 8 | 6 | 3–9 | 48 |
| 2 | 0.5 | ||||||
| 7 | Skewed | 0 | 8 | 6 | 3–9 | 48 | |
| 8 | Skewed | 1 | 2.5 | 8 | 6 | 3–9 | 48 |
| 2 | 1 |
The simulations with β = 0 (no difference between group 1 and group 2 data) investigate how liberal/conservative the tests are (performance under the null), whereas the simulations with β ≠ 0 (real difference between group 1 and group 2 data) investigate the power of the test (performance under the specified alternative). Case 1 refers to data where only a single group is represented in each cluster, and case 2 refers to data with approximately equal numbers from both groups in each cluster. For each of the 8 models, 10,000 datasets were generated.