Skip to main content
. 2022 Jul 1;39(3):1366–1383. doi: 10.1016/j.ijforecast.2022.06.005

Fig. 1.

Fig. 1

(a) Predictive medians and 95% prediction intervals for incident deaths in Ohio generated on February 15, 2021 by two example component forecasters. The vertical axis scale is different in each facet, reflecting differences across several orders of magnitude in forecasts from different forecasters; the reference data are the same in each plot. The data that were available as of Monday, February 15, 2021 included a large spike in reported deaths that had been redistributed into the history of the time series in the version of the data available as of Monday, February 22, 2021. In this panel, forecaster names are anonymized to avoid calling undue attention to individual teams; similar behavior has been exhibited by many forecasters. (b) Illustration of the relative weighted interval score (WIS, defined in Section 2.5) of component forecasters over time; lower scores indicate better performance. Each point summarizes the skill of forecasts made on a given date for the one- to four-week-ahead forecasts of incident cases across all state-level locations.