I wanted to present some facts that would be useful to anyone interested in making a very rough estimate (l of the infection fatality rate of COVID-19. The main reason I’m putting this out there, to be honest, is because the figures pretty clearly contradict some of the more extreme claims about COVID-19 being much less deadly than is often imagined. I have no idea what the real infection fatality rate is, but I think we know enough from New York to say that in that city it’s unlikely to be under 0.5%, or much over 2% (unless there’s as yet unknown long-term health consequences).

- Between March 11 and April 27, 21000 more residents of New York City died than would normally die during this period. This is according to New York Times analysis using government agency figures. New York City has 8.4 million residents. This gives a death toll of 0.25%, or 1 in 400 New Yorkers in a month and a half.
- Excess deaths could be an overestimate of the effects of COVID-19 because people might not be going to hospital due to fear of COVID-19, and thus there are more deaths from other causes. On the other hand, deaths usually fall during recessions (weird, I know), and there are likely to be fewer accidents and violent crimes, so excess deaths might be an underestimate of how many COVID-19 has killed.
- A recent antibody study found 25% of New Yorkers had COVID-19 antibodies. The exact time needed to develop antibodies isn’t known, but 3-4 weeks has been suggested as an average. The population tested was out shopping and thus possibly more likely to have the disease. The false-positive & negative rate of the test isn’t known with precision.
- According to this paper: https://www.thelancet.com/journals/laninf/article/PIIS1473-3099(20)30257-7/fulltext it takes an average of 18 days to die of COVID-19, from the onset of symptoms. Naturally, the time to die after the infection is even longer. Mean time to onset of symptoms is often given as 5 days. Adding these together suggest a mean time to death of 23 days, or about three and a half weeks.

Having presented the facts, let me venture my own estimate. This is extremely tentative, and mostly is just meant to illustrate how we could use these facts to estimate fatality. Let us gamble that excess death is an approximately accurate measure of the number killed by COVID-19. It seems to me that 21,000/(25% of 8.4 million)= 1% is a good estimate of the rough ballpark of fatality in New York because:

A) On the one hand, the 25% infected rate is probably an underestimate, because it takes 3 1/2 weeks to develop antibodies.

B) On the other hand, the death rate is an underestimate of the proportion currently infected who will die, because it takes 3 1/2 weeks to die.

We can regard those figures as cancelling out- people infected 2 weeks ago probably won’t show up on the antibody test, but even if they are going to die, they probably won’t have died yet. Thus people affected less than 3 1/2 weeks ago are unlikely to affect either the numerator or denominator.

**EDIT**: Note that antibody testing started on the 20th, a bit earlier than the 25th, when death statistics go up to, so slightly reduce the denominator