Anticipating epidemic transitions with imperfect data
Three time series for the same simulated outbreak are shown: daily snapshots of the number of infected (black line), weekly aggregated case reports with no reporting error (perfect reporting, blue line) and weekly aggregated case reports with reporting error (imperfect reporting, red line). Reporting error is modeled using a negative binomial distribution with ρ = 0.25 and ϕ = 0.1. The high overdispersion (small ϕ) means that there is little visual correspondence between the number of case reports with and without reporting error. Additionally, the number of case reports can exceed the true number of cases (overreporting), in spite of the low reporting rate. SIR simulations performed using the same parameters as Fig 1.