Why might we need random samples on a regional level repeatedly every 4 weeks in Germany? Some thoughts: There are two broad exit scenarios discussed: 1) controlled infection shielding (somehow) the old while letting the young out doing smart distancing. 2) hammer and dance. 1/n

1) Epidemologists are very sceptical. Besides being almost impossible to separate young and old, it is ethically questionable to start with. Moreover, even if we could do it, and if young have .1% of dying and old 2% we have 60 mio young hence 60000 dead + 120000 hospitalized 2/n

2) So Hammer and Dance Suppression. Great idea but it depends on 3 factors: a) the level of currently infected, the speed of decline (R<1) and the level we want to end up with that we can „handle“ with test and trace. We need to know the number of currently infected for that 3/n

So we would need at least one representative sample once. If we have 2-400000 currently infected (a modest dark number for unaccounted cases) we could do Wuhan style lockdown generating an R=.5 or Italien style lockdown (bit higher R). South Korea controls it with 7000 cases 4/n

and 100-200 new infections. According to some standard basic modeling the expected length of a Wuhan style lockdown (51 days there) is likely substantially higher than that. social and economic costs of imprisoning 80 mio people for so long is -in my view - too much to take. 5/n

So what? Chancelor Merkel suggested a doubling time if cases of 2 weeks then we could open up a bit. 2 weeks doubling still implies that current 10000 hospitalized and ca 2500 in intensive care will increase within 6-8 weeks to a point where we hit health constraint...6/n

I guess in my view that leaves only R close to 1 for a prolonged period of time (until testing gets so much better that we can test most people every morning, but we currently test 350000 a week, or vaccine is found or something) what austria and bavaria essentially argue for 7/n

Given that current measures are already hard to bear much longer we therefore urgently need to figure out which of the many things really are needed and which can be dispensed together with wearing masks in public (the miracle weapon) So far politicians like Söder or Kurz 8/n

Seem to want to use time variation to figure it out i.e. slowly open up some local businesses (at the same time ordering masks to wear) then increase production then they might open up kindergardens or schools each time waiting to see badly measured case numbers to react 9/n

Measuring success with non random case numbers that condition - to get access to the test you must have symptoms and contact with a known infected - or hospitalization rates will be very tricky. Using only time as variation will be a lengthy endeavor with huge costs. 10/n

In Germany we should not let Söder and Laschet do a competition for Merkel succession by playing a game who does opening up better and ending up - in equilibrium - doing precisely the same. to the contrary we should use the German local health system to allow for variation 11/n

Rather than learning over time which measures might be relaxed and which not we should -learn by utilizing space, i e regional variation on a Kreis or better Arbeitsamtbezirk (143 or so of them) How? Well ideally we would obtain random samples on infections by region 12/n

at time t, allowing Kindergarten to open up in Munich but not in Hamburg, wait 4 weeks and compare outcomes (assuming a common trend currently stabilized around R=1 initially or adding a third pre-sample). Similar open up cafe in Cologne but not in Düsseldorf...13/n

Advantage: we could try out more measures at the same time. Moreover it is less risky than using time variation. Assume something goes wrong and we get exponential growth in a region we have half of the countries hospitals (of the control group) to take patients 14/n

Disadvantage: a) we would need to use random testing before and after at a sufficiently high sample size taking away tests resources of PCR tests from medical use. b) the common trend assumption might not hold and we have to deal with it somehow (proposals?) c) people might 15/n

Feel like being used as rats. However, if we try it using variation over time we run the experiment on the large scale...we are all rats then. I think Germany‘s health system being extremely decentralized might help for regional experiments. If random testing can not be used 16/n

I still think we should use regional variation and use hospitalization rates or other observables as proxy- measures. Once we figure it out locally we can roll it out globally still taking it back if system gets under pressure Desperate times might need some desperate designs/end