Table 4.
Model performance based on 10-fold cross-validation.
| Simple Differences | Complicated Differences | |||
|---|---|---|---|---|
| Test for Sufficiency | Test for Sufficiency | |||
| Univariate LR | Theoretically Informed Model | Multiple LR | Random Forests | |
| Average | Suicidal Desire & Capability for Suicide |
All Variables | All Variables | |
| Entire Sample (NSSI Only: N = 319; Suicide Attempt with or without NSSI: N = 635) | ||||
| AUC | 0.55 [0.45, 0.65] | 0.58 [0.48, 0.69] | 0.73 [0.63, 0.82] | 0.72 [0.63, 0.82] |
| Precision | 0.72 | 0.74 | 0.84 | 0.84 |
| Recall | 0.68 | 0.58 | 0.75 | 0.73 |
| Brier | 0.45 | 0.42 | 0.27 | 0.28 |
| Subsample (NSSI Only: N = 319; Suicide Attempt without NSSI: N = 52) | ||||
| AUC | 0.57 [0.36, 0.78] | 0.60 [0.38, 0.81] | 0.70 [0.48, 0.91] | 0.74 [0.54, 0.93] |
| Precision | 0.19 | 0.20 | 0.34 | 0.30 |
| Recall | 0.74 | 0.76 | 0.62 | 0.77 |
| Brier | 0.45 | 0.45 | 0.23 | 0.29 |
AUC, area under the receiver operating characteristic curve; AUCs of 0.50, chance-level discriminative accuracy; AUCs of 1.0, perfect discriminative accuracy; precision, positive predictive value; recall, sensitivity; precision and recall both range from 0 to 1; with higher values indicating better model performance; Brier scores of 0, perfect calibration; with scores closer to 0 indicating better calibration; LR, logistic regression.