Table 2.
Performance of Linear and MARS Models in Predicting eRGC From Time and Time–Frequency Features Derived From ERGs
| Linear | MARS | ||||
|---|---|---|---|---|---|
| R 2 | P a | R 2 | P a | P, Linear vs. MARS | |
| Model 1: Markers | 0.21 (0.07) | — | 0.34 (0.06) | — | 0.004 |
| Model 2: DWT + markers | 0.41 (0.08)b | 0.001 | 0.53 (0.07)b | 0.011 | 0.055 |
| Model 3: MP + markers | 0.43 (0.08)b | 0.001 | 0.63 (0.08)b | <0.001 | 0.002 |
| Model 4: DWT + MP + markers | 0.50 (0.09)b | 0.001 | 0.63 (0.07)b | 0.001 | <0.001 |
Model 1 was informed by the amplitudes of the a-wave, b-wave, i-wave, PhNR1, and PhNR2. Numbers in parentheses indicate standard deviation from the bootstrap resampling.
a
When compared to model 1.
b
Not significantly different at P < 0.05 for pairwise comparisons among models 2, 3, and 4 for the linear and MARS models separately.