Inferring COVID-19 spreading rates and potential change points for case number forecasts

Jonas Dehning; Johannes Zierenberg; Frank Paul Spitzner; Michael Wibral; Joao Pinheiro Neto; Michael Wilczek; Viola Priesemann

(2) People that are infectious are observed with a delay that is now lognormal distributed. In the prior SIR model we assumed a fixed delay between infection and observation. The delay has a scale parameter σ of 0.3 and as median a LogNormal(5, 0.2) (days) prior, to match approximately the total delay between infection and observation of the previous model. We changed the prior for the recovery rate µ to a median of 1/3, which is similar to other SEIR simulation studies [17] . The priors for λ0 to λ3 were increased to 2,1, 0.5 and 0.25 respectively and a scale parameter of 1. A: Time-dependent model estimate of the effective growth rate λ * (t). Note that λ * (t) in the SEIR-like model is not directly comparable to λ * (t) in SIR models, because of the lognormal-distributed incubation period, decreasing the effective growth rate. B: Comparison of daily recorded new cases and the model, linear scale (inset: log scale). C: Comparison of total recorded cases and the model, linear scale (inset: log-scale). D-F: Prior and posterior distributions of all free parameters.