You use a combination of methods to determine an approximate number of actual (versus diagnosed) cases. Serology testing for a statistically representative sample of the population is one method. Another is doing blanket testing of an entire town and comparing that with diagnosis data. Other isolated populations can also be verified en masse, like people on cruise ships or islands.
One by itself is insufficient to get a good estimate of how many cases are being missed, but when you do a lot of these, the numbers begin to coalesce. In the US, IFR estimates have varied from study to study between 0.63 - 0.67%, putting the CDC’s number of 0.65% right in the middle. We arrive at that by seeing that actual cases are around 2.7 - 2.75 times the number being diagnosed.
Lots of different studies have been done with this in the US, so there’s a lot of data on different parts of the country. The results are consistent with each other.
So, what is the estimated number of missed cases in the US vs the number of diagnosed cases? Exclude those cases where covid was listed as a cause of death since there’s a lot of reason to suspect that those numbers are greatly inflated - a close friend is a mortician and has confirmed that this is the case, at least where he is. In other words, number of diagnosed and recovered cases vs estimated missed cases.