Table 1.
Knot selection accuracy: comparing our parametric bootstrap MLE 2 df selection procedure to AIC and BIC by how frequently they correctly select two knots in 100 datasets simulated and evaluated at each combination of sample size, prevalence, and selection criterion settings.
| Sample size (n) | Prevalence (po) | Complex sample weights | Selection criterion for our test (α) | Frequency of correctly selecting 2 knots |
Test 1: Ours vs. AIC, p* | Test 2: Ours vs. BIC, p* | ||
|---|---|---|---|---|---|---|---|---|
| Our test | AIC | BIC | ||||||
| 500 | 0.10 | No | 0.10 | 13 | 22 | 2 | 0.136 | 0.006 |
| 500 | 0.33 | No | 0.10 | 20 | 38 | 1 | 0.008 | <0.001 |
| 500 | 0.10 | No | 0.25 | 19 | 20 | 3 | >0.999 | <0.001 |
| 500 | 0.33 | No | 0.25 | 27 | 37 | 1 | 0.172 | <0.001 |
| 5,000 | 0.10 | No | 0.10 | 78 | 76 | 21 | 0.867 | <0.001 |
| 5,000 | 0.33 | No | 0.10 | 92 | 76 | 94 | 0.003 | 0.783 |
| 5,000 | 0.10 | No | 0.25 | 63 | 69 | 18 | 0.456 | <0.001 |
| 5,000 | 0.33 | No | 0.25 | 83 | 77 | 93 | 0.377 | 0.048 |
| 5,000 | 0.33 | Yes | 0.10 | 73 | 0 | 0 | <0.001 | <0.001 |
*p-Value from two-sided Fisher’s exact test.