Table 1 Calibration metrics for nominal 95% prediction intervals by season and type of interval. The metrics have been calculated on the test periods of the validation data (2016–2018), i.e. on data not seen either during training or calibration, and aggregated over 23 countries
From: Empirical prediction intervals applied to short term mortality forecasts and excess deaths
Coverage | |||||
|---|---|---|---|---|---|
Annual | Dec–Feb | Mar-May | Jun–Aug | Sep–Nov | |
Negative Bin. PIs | 0.93 (1) | 0.85 (3) | 0.91 (1) | 0.95 (1) | 0.99 (3) |
Empirical SN PIs | 0.91 (2) | 0.89 (1) | 0.89 (2) | 0.91 (2) | 0.94 (1) |
Empirical RQ PIs | 0.86 (3) | 0.86 (2) | 0.82 (3) | 0.86 (3) | 0.91 (2) |
Mean interval score | |||||
Negative Bin. PIs | 0.365 (2) | 0.553 (3) | 0.376 (2) | 0.287 (1) | 0.248 (2) |
Empirical SN PIs | 0.346 (1) | 0.467 (1) | 0.369 (1) | 0.310 (2) | 0.241 (1) |
Empirical RQ PIs | 0.398 (3) | 0.534 (2) | 0.437 (3) | 0.363 (3) | 0.260 (3) |