Table 5.
Power (percent) for detecting a treatment effect based on 4000 simulations in a two-arm, two-treatment crossover study, by μ and the standard deviation of the values after treatment (). Results are for at least 99% of the simulations, except as indicated.
| Model | μ (16 per arm, σTrt = 1) | μ (30 per arm, σTrt = 1.4) | ||||||
|---|---|---|---|---|---|---|---|---|
| 2.0 | 2.5 | 3.0 | 4.0 | 2.0 | 2.5 | 3.0 | 4.0 | |
| Nonparametric analyses | ||||||||
| Friedman test (NQ only) | 56.8 | 64.9 | 68.4 | 69.3 | 28.4 | 50.6 | 64.2 | 75.0 |
| Friedman test (both NQ and ND) | 61.5 | 66.6 | 70.0 | 68.9 | 41.0 | 57.0 | 66.3 | 75.6 |
| Schouten/Kester test | 19.5 | 28.6 | 37.4 | 40.7 | 14.4 | 29.4 | 47.3 | 64.7 |
| Mixed models using follow-up data only | ||||||||
| Equal variances | 68.9 | 78.4 | 84.6 | 87.1 | 30.5 | 58.4 | 76.8 | 89.9 |
| Unequal variances | 42.4a | 70.5a | 83.3 | 86.7 | 56.1a | 77.2a | 86.0 | 89.6a |
| Mixed models using both baseline and follow-up data | ||||||||
| Equal variances | 44.3 | 51.6 | 56.9 | 58.6 | 18.5 | 36.1 | 53.5 | 71.4 |
| Unequal variances | 36.8 | 49.4 | 56.3 | 58.6 | 31.5 | 49.8 | 63.5 | 73.2 |
| Mixture models: likelihood L2, follow-up data only | ||||||||
| Equal variances | 14.1 | 17.0 | 17.0a | NA | NS | NS | NS | NS |
| Generalized estimating equation logistic regression models for a significant decrease | ||||||||
| NQ and ND combined | 21.0 | 31.1 | 38.9 | 42.5 | 14.3 | 29.4 | 47.3 | 64.7 |
| ND separate from NQ | 30.4 | 36.4 | 42.0 | 43.0 | 24.3 | 40.0 | 53.6 | 65.8 |
Assumptions : treatment reduces the underlying mean μ by 0.5, the standard deviation of the untreated mean is 1, Type I error is 0.05, the quantification limit is log10(80) = 1.90. NQ: non-quantifiable; ND: not detectable.
NQ only : NQ and ND results were combined.
a
Results for 90% to 98.9% of all simulations
NA: Results for less than 20% of all simulations.
NS: not shown, excessive Type I error.