Originally Posted by

**RubeRad**
Thanks, that does look up my alley, I will enjoy reading it tonight and report back.

Thanks for the link, I read it, and although it is very old by now (Feb!) it did teach me about one important factor that I had overlooked, and that is

**LAG**. With a growing pandemic, it is not accurate to divide today's deaths by today's cases. Today's deaths need to be divided by the number of cases that led to the deaths, which was one incubation period ago, a smaller number of known active cases. Dividing by a smaller number makes the fatality rate bigger.

Here is my new spreadsheet, anybody can view it and make their own copy to play around with.

Here is a screenshot.

So new assessment: as I've been discussing all along, the numbers in the OP were inflated due to unknown undercounting of cases, but also deflated by not accounting for lag.

This graph has the lag error somewhat fixed*, but the undercounting of cases I can't do anything about besides making a guess for exactly what the right count should be, which I have no reasonable way to do.

So the new graph is also an overestimate of fatality rates, but a larger upper bound than before, so the truth, wherever it lies below these numbers, is higher than I might have thought before.

The 80+ jumped from 23% to 29%. In line with everything I said before, I think the 80+ demo will have the smallest amount of case undercounting, meaning that ridiculously large 29%, although it's an overestimate, it's probably not by much. I would speculate the true fatality rate for 80+ in CA is north of 25%. (Note I originally speculated that with an overestimate of 23% the truth might be north of 20% -- no matter how you slice it, it's huge)

* to truly 'fix' the lag, you need to use a lag equal to the time from infection to death, or at least the incubation time (assuming it takes a certain number of days to die). The number that is always thrown around for corona is 14 days. But the LA Times graphs didn't have enough detail for me to read off the total number of cases on Jun 9. My original spreadsheet was made on June 6, 17 days ago, so I used exactly the same numbers for case totals and demographics as in the original spreadsheet. So I'm dividing by numbers of cases which are too small by about 3 days, which means the lag fix is a little bit of an overcorrection. But not much. As a sensitivity test, I tried putting 130000 in for total deaths instead of 126000 (as a guesstimate of what 3 more days might have added), and the 80+ number only dropped to 28%.