It’s been a while ago since I ran my MCMC-hack to estimate the Swedish Infection Fatality Rate, main reason being that each run takes an awful lot of computing time, around 10h, to get the MCMC to converge… and during those 10h’s my computer is pretty much unresponsive and useless for any other work….
What the FHM study finds is:
- Infection Fatality Rate : 0.6%, 95% CI 0.4 – 1.1%
So that FHM paper motivated me to run my hack.
Let’s check the output of my MCMC-based Bayesian Inference:
The first of the 4 graphs above is the one to focus on – it shows the posterior probability distribution for the estimated Infection Fatality Rate. Turns out the numbers obtained are very much inline with FHM’s findings:
- IFR median : 0.7 %, with a Highest Probability Density Interval (HPDI) 0.4 – 1.0%, and 89% Credible Interval 0.2 – 1.8%.
A remarkably close match. However, the accuracy of these findings is totally dependent on the quality of the data, that is, the official number of deaths and the official number of confirmed. I don’t consider either of these variables as very certain nor accurate: the number of confirmed is heavily dependent on the amount of testing, which has varied over the duration of the virus, the sensitivity and specificity of the anti-body tests which I haven’t seen mentioned anywhere, and the number of deaths, while plausibly a much more certain variable, is still highly uncertain, since it’s still very unclear whether a person registered as dead in the statistic has died FROM Corona or WITH Corona, which makes a world of difference…