Then for the general population below whatever that age is (maybe 65?), you'd get roughly `(51-30)/4449 ≈ 0.47% FR` (non-PCI deaths/total positive). It's a bit unclear as it seems the numbers are changing as test results arrive. For the PCI cohort: 30/1258 ≈ 2.38% (deaths/recovered).
The 0.47% FR would seem much more plausible given the spread of the virus and the number of asymptomatic cases that appear to exist from serological testing.
You're hugely overestimating still. You seem to assume that all of the over-65s in the system are at PCI. In fact, only a small fraction of older people, even in late decades, need to be in long term medical care.
True, it's more of an upper-bounds of sort. There's several preprints from European researchers giving IFR's of 0.08% and 0.37%. So it works pretty well from a Fermi estimation method (https://what-if.xkcd.com/84/) (e.g. Bayesian inference really). Also, the age distribution in prison isn't necessarily the same as that of the general population. There's lots of limitations for a comparison to the general population but it gives some bounds.
I'd think the statistic of "average years of life lost" based on expected average of lifetime. Otherwise not sure of a better statistical way to measure age-adjusted IFR, which would be helpful.
One thing I’m curious about is the influence of people having symptoms being more likely to be tested. As others have mentioned this is only 15% of the total prison population.
I’m also curious how you could account for that if you could. Besides random sampling + tracking individuals afterwards even if they left the prison.
I’ve recently started digging more into statistics and probability theory and looking forward to learning how these biases might be factored in.
The 0.47% FR would seem much more plausible given the spread of the virus and the number of asymptomatic cases that appear to exist from serological testing.