Comparison of cases reported in epidemiological bulletins to projections produced by our nowcasting models in real-time.

Accuracy is measured in comparison to the ground-truth case counts eventually reported at the end of each outbreak. Accuracy of each model is calculated separately for the second-to-last week with epidemiological data reported in each bulletin, the last week with epidemiological data reported, and the first week without any epidemiological data reported (we do not evaluate longer time horizons since in several outbreaks epidemiological bulletins are produced near-weekly, rendering projections based on previous bulletins obsolete). Accuracy measures are then averaged over epidemiological bulletins for each outbreak (ranging from seven for Zika in Colombia to 17 for Ebola in the DRC), with standard deviations shown in parentheses. The accuracy of the data source producing the most accurate point estimates on average is bolded.