I think we're mostly talking past each other since we agree on the IFR. I'll just say "it's complicated."
It is tricky, yet I don't see any force in your points, and one is flatly incorrect:
- There is nothing wrong with including early deaths. It would be wrong not to include them, since the number is compared to the total number of known cases.
Your stated method was to use the case count from
two weeks ago and to use the death count from
today. That is too crude a method because many deaths occur between 3-14 days after diagnosis. I would use a 1-week lag, not two weeks. Could be 6 days, or 8 days, but this is just an estimate.
See these Korea cases. They happen to have an inflection point in the ramp so you can kind of line that up in deaths and cases. Looks like about a one-week separation. Which makes a lot of intuitive sense - we have a lot of info showing that people often die in about a week. But it can be shorter and it can be longer.
Let's apply that to Germany, specifically: Using your method you get 10% but it's obviously wrong - a lot of the deaths occur sooner than two weeks after identification. But not all. See these graphs:
My crude (but more accurate than yours) method results in a death rate of 123/7272 = 1.7%. This method is crude because it depends on the testing levels, etc. And it's obviously not a rigorous model - it's just a way to quickly see the approximate mortality during a ramp. If the cases aren't ramping anymore it will also potentially be inaccurate.
Conclusion: It doesn't take much at all to get from 9.6% to 0.9% when the case count explodes by a factor of 20 over two weeks (due to testing increases as well as inherent case growth).
You appear to be repeating the same mistake in this post by claiming the current mortality rate to be 1.3%, as it is apparently derived by comparing today's numbers.
Sure, it isn't exactly right, but since the case count has been fairly steady for quite a while (this only applies in South Korea and China), it does not introduce a lot of error. Maximum error you can see above is likely about the number of critical cases which will convert to some more deaths.
Incidentally, if you do this method with yesterday's numbers for the US, you get 553 deaths / 4663 cases = 12%, which implies a case undercount
at that point in time 7 days ago of about a
factor of 10 or more - but this is also dependent on the early deaths - US numbers are skewed a bit by the nursing home factors early in the outbreak.
Anyway, seems reasonable that the undercount may have been that severe 7 days ago - it was widely accepted at that point by epidemiologists that there were about 80k active cases (big error bars) in the US at that point in time. And you can see from NYC that a huge number of these cases were in NYC. But they are elsewhere too.
We're probably at about 200k+ cases in the US right now. Really need to get that testing going! Good to see NY got 27k people tested yesterday (~5700 of which were positive!!!).