Simple models for COVID-19 death and fatal infection profiles
share
Simple smooth additive models for the observed death-with-COVID-19 series adequately capture the underlying death rate and strong weekly pattern in the data. Clear inference about peak timing is then possible. Further, inference about the earlier infection rate dynamics driving the death rate dynamics can be treated as a simple Bayesian inverse problem, which can be readily solved by imposing a smoothness assumption on the infection rate. This straightforward semi-parametric approach is substantially better founded than the running mean smoothers which generally form the basis for public debate. In the absence of direct statistically based measurement of infection rates, it also offers a usefully assumption-light approach to data analysis, for comparison with the results of the more assumption-rich process simulation models used to inform policy. An interesting result of the analysis is that it suggests that the number of new daily infections in the UK peaked some days before lock down was implemented, although it does not completely rule out a slightly later peak.
READ FULL TEXT
