Jeff in this thread gets at something that I have observed a lot & want to write more on at some point, & that's this: extremely smart pure stats-focused ppl have been getting played by the numbers they're staring at. https://twitter.com/jeffhauser/status/1248973055868755969

I am trying to review a power bank so I can't get into this now, but some random points.

1. A thoroughly sliced & diced set of numbers & regressions & curves gives you confidence, as it should in most cases. But this pandemic calls for humility.

2. Numbers that appear to be the same are often not the same. A line labeled "case count" from Iran arose from a different political/social/medical context from that same line in Florida, & different still from China. If you're just comparing "case counts" & are not being careful

... finding ways to incorporate a lot of context about test availability, positivity rates, & even local political pressures, you're producing really slick-looking GIGO.

A lot of confident nerds right now downloading CSVs & making graphs that are divorced from all this context.

3. The good epi models are tools for developing intuitions about what might happen as you vary the inputs. They aren't predictions. It's not like a physics model, where it generates a testable prediction that you then confirm via observation/experiment. So if your physics...

...intuitions will betray you as you monkey around with these models and datasets that are well outside your normal domain.

When people ask "what has the pandemic proved you wrong about?", I think of this thread. It turns out I probably had it backwards here. https://twitter.com/jonst0kes/status/1226332868756082688