"Mit den flächendeckenden Tests wurde am vergangenen Dienstag begonnen. Mittlerweile – Stand Freitagabend – wurden 456 St. Ulricher getestet, und fast die Hälfte weisen Antikörper auf. Über 2 Drittel der Getesteten sind zwischen 20 und 59 Jahre alt. Und wichtiges Detail am Rande: Fast ein Drittel gab an, keinerlei Symptome gehabt zu haben."
Flächendeckend =/= nationwide, it just means wide-area testing, in this case in one town.
The sample was entirely self-selected; the service was advertised, patients presented themselves at a clinic and paid €30 apiece for the 10-min thumb-prick test, presumably most of them because they suspected they'd already had it.
Almost a third = 30% asymptomatic amongst the confirmed infected, which sounds at upper end of ranges reported elsewhere (18..30%).
Yes, we need another study to find out the effect of paid self-selection on a sample. It implies that the participants not only think it is worth the effort, but also that they are actively informing themselves.
In any case, the result would suggest more a severe under-testing, and less an extreme percentage of asymptomatic cases (which aren't tracked either, to distinguish asymptomatic and pre-symptomatic cases).
Severe under-testing is also likely in the US, we have started to expect that here on the forum several weeks ago. I think at that time, more by a factor of 10 or 20, without expecting that factor to be a constant. That was based on looking at the lag times in the graphs in the following way:
Current deaths in the US April 18 are 39,014. For example with a lag of 2 weeks, the number of total cases on April 4 was 313,379.
That's an adjusted CFR of 12%. Nobody expects an IFR that high. So even if you use a relatively high estimate of 1.2%, you would expect there to have been more than 3 million infections 2 weeks ago.
That is a factor above 10x. (With a lag time of 1 week, it would be above 7x.) That's the kind of calculation we were doing a few weeks ago. (Or is it just 2 weeks? It seems so long ago.) At that time, I think I came to the estimate that the factor might be higher than 20x (less than 5%).
So at least for some of us, a factor of 50x would not be as revolutionary or consequential as it may sound. It would almost be more like a confirmation. Of course, all this depends on how much testing has been done in a specific location/area. And on the relative point in time within the curve of the local spread, for example. Could self-selection account for the remaining difference? I don't know, but I can't say it couldn't.