Table 15.
Operational linkage of statistical Metrics.
| Statistical Metric | Computation Source | Decision Threshold | Operational Impact |
|---|---|---|---|
| Hybrid t-test (p-value) | 10,000 Monte Carlo (Section "Statistical method performance.Statistical method performance") | p < 0.001 | Model retention in ensemble; reject if violated |
| Cohen’s d effect size | Heatmap (Fig. 7) | d ≥ 0.8 (large); decline > 0.2 | Prioritize weighting; trigger adaptive retraining |
| Cross-Dataset Robustness Index (CRI) | Equation 18 (Table 14) | CRI ≥ 0.98 | Full ensemble contribution; else partial fine-tuning |
| Variance ratio (heteroscedasticity) | Bootstrap distributions | Ratio > 2:1 | Temporary confidence down-weighting in ensemble |
| Longitudinal drift | Monthly effect-size tracking | d shift > 0.2 or p rise > 0.05 | Activate adaptive learning rule (Eq. 4) |