> unless 100% of the population is administered a test with no risk of false negative or positives.

Actually, the "risk" of false positives or false negatives is dependent on prevalence, 100% is not always required. If you have a test where 5 of 100 cases are false (i.e. 95% times is not wrong), and nobody of these tested is actually infected, you could incorrectly believe that 5% of population is infected even when nobody is, meaning such a conclusion would be completely false.

But if the population is actually already e.g. 50% infected, the same test can "lie" only 5%, giving you 47% or 53% but still being "mostly true" (from the engineering point of view).

So it is important to ignore the test reports as long as they are close to their false positive rate, which they were in a lot of antibody tests done up to now.

Also "false positives" and "false negatives" can lead to wrong handling of the cases, but that's another topic.

Thank you for the sources!

Initially you were implying that the IFR of the flu was 0.1% and comparing it to an IFR of 0.4%, this is what I disagreed with strongly

The bloomberg link you're giving is showing that the vast amount of randomly sampled serological studies are showing IFR of 0.5%-1%. And importantly, looking at how the random sampling has been done for each case and ranking by quality, >0.5% results rank at the top (he studies with the biggest sampling problems show the lowest IFR)

0.2-0.3% is by far in the bottom range of estimates

I think the data shows a pretty clear single order of magnitude difference between the IFR of the flu and novel corona. But I agree that >1% IFR is extremely unlikely

How, exactly are we ranking by quality? The only anti-body study that's been published in a peer reviewed journal has indicated an IFR ranging from 0.12% to 0.2%. The author did make note of these criticisms that the infected rate was over-counted, but even when using an infected rate of 1.5% instead of the paper's estimated 2.5% the IFR would turn out to be 0.33%.

Plenty of the >0.5% results have serious sampling problems of themselves. The Dutch study sampled people who were donating blood. The subset of the population that donates blood is could easily have different behavior than the general population: like being more health-conscious and thus less likely to be infected, which would inflate the infected fatality rate.

And when we compare against the flu, we also have to consider that there are different methods of estimating the infected fatality rate. More conservative estimates on infections of the seasonal flu yield higher IFR values. The epidemiologist being cited here is providing relative estimates of 0.04% and 0.2-0.3%. We could go with more conservative estimates on the infected rate for COVID-19, but we'd have to be similarly conservative when measuring the IFR of the flu.

I was ranking "randomly selected" higher than "people volunteering" (which was one of the main issues with the santa clara study, mixed with the test used being so imprecise the error bars were big enough to include a 100% false positive rate. That is by far one of the worst designed studies I've looked at, even if it has been peer reviewed)

I'm not sure the blood sampling would skew the direction you think it does, healthy individual could just as well mean that they're more likely to have had it since they're less likely to sit at home and stick to a quarantine. But you are right, it definitely counts as a sampling issue and I'll concede that ranking by quality is a lot less clear cut, I should not have been so confident