Brilliant thread here. We need a big collider bias Klaxon we can all hit. If you have never done hand to hand combat with trying to make causal inference from observational data (aka "what on earth is actually going on in the real world") for COVID19 - read this. https://twitter.com/bristimtom/status/1259458854020165632

Collider bias is one of these things that is actually really obvious that it is exists, and once you realise it is exists it makes you question *everything* from observational studies. Really ... irritating. True, but irritating.