Fig. 5: The relative information in case and death data for EVD and COVID-19 case studies.
We compute information metrics θ(.) for case (\({C}_{1}^{\tau }\)) and death (\({D}_{1}^{\tau }\)) time series using empirically derived under-reporting and delay noise distributions for COVID-19 (top panel) and EVD (bottom panel). See the main text for the specific distributions used, which account for the uncertainty in noise estimates (in the absence of knowledge of this uncertainty, maximum entropy distributions are applied). We take 104 samples from each distribution and calculate the logarithmic difference \(\log \theta ({C}_{1}^{\tau })-\log \theta ({D}_{1}^{\tau })\), with positive or negative values indicating when case or death data have the higher information content, respectively.
