Figure 7. Predictive performance of self-reported "experts" and "non-experts" across forecast horizons.

Forecasts from "experts" and "non-experts" were combined to two separate median ensembles, including both direct and R _t forecasts. A– D: WIS stratified by forecast horizon for cases and deaths on the natural and log scale. E, F: Empirical coverage of the 50% and 90% prediction intervals stratified by forecast horizon and target type. Grey dashed lines denote the nominal coverage that a model should ideally achieve.