No Fluke

[l33tminion]

This is from last week, but I really wanted to post about it: Scott Alexander did a fantastic deep dive into meta-analysis of studies of the COVID-miracle-cure du jour, ivermectin. There's a family of very professional looking websites publishing meta-analyses of alternative COVID treatments that make it seem like ivermectin is a wildly effective COVID-19 treatment. But also everything else. Prompting the obvious suspicion that they've forgotten that publication bias is a thing, or are using statistical methods that make everything seem like a success. The whole thing is fascinating, but the surprise conclusion is that the big confounding variable seems to be the should-have-been-obvious one, given the straight-forward question of "why would you expect an anti-parasitic to help with COVID-19?"

(He also did a follow-up post that takes a serious look at the question of "why not just take everything that might work?")