Thread by @plaforscience

Sometimes a countersample is better than the explanation

So, let's go Counterfactual

We'll design an outbreak
It'll have the supposed height of Spain's 5th wave with Gompertz develope observed in waves all around the world, as brilliantly showed by @MLevitt_NP2013

Positivity:

We set the 5th wave 16 weeks span for trying 3 testing scenarios

A constant testing floor, increased with tracing, depending on cases found

The ACTUAL testing policy used in Spain's 5th wave, politically decided, for the other 2

We'll check their dot clouds&linear regressions

This is the SUPPOSED policy used: we do more test cos we found more cases

We notice Cases/Tests look UNRELATED

Very different cases results for similar test numbers: it reflects the test policy divergences with the outbreak developing underneath

ACTUAL policy: test policy growing and decreasing with epidemic-like shape, with the incredible ability to EXACTLY FORESEE gradients, length and turning point, with PERFECT timing

Even though, underneath epidemic creates anomalies, and cases can't be perfectly linked to testing

Let's check what happens if you FAIL your timings, and mistake true outbreak ongoing by only 1 week, with the actual epidemic replicating policy

Cases and test become clearly unrelated, as true epidemic peak pushes cases up even with lower testing

And now, let's see what we really find with spanish official data

A perfect linear fit
Cases EXACTLY follow tests

We've seen that with a REAL outbreak occurring, it doesn't matter you follow cases, or you plan how cases will develope. Even doing this perfectly, ANOMALIES HAPPEN

The perfect correspondence cases/test can only be explained ONE way: there was NO outbreak ongoing during so-called 5th wave.

For such a perfect fit you need a ~FLAT prevalence, like that you'll expect at AUGUST basal phase.

Spain's 5th wave was definitely a pure TEST ARTIFACT.