Thread by @gro_tsen

I stumbled upon this paper which performs multivariate linear regression analysis to try to infer the effectiveness of various policy measures on covid-19 mortality.
ajtmh.org/content/journa…

One of their conclusions is that mask mandates are very effective. Another is that lockdown duration has no statistically significant effect (the model doesn't even conclude whether the effect is positive or negative).

I'm very skeptical as to whether we can draw any useful conclusion from such multivariate linear regressions between countries. There will be simply too many unknown variables and confounders. But this isn't really my point here.

Now here's how the results of this paper gets reported in a different paper pnas.org/content/118/4/… — an “evidence review of face masks”. I have no idea what “transmission was 7.5 times higher” might mean, but the 7.5 figure is nowhere to be found in the cited paper (ref 19).

And of course journalists took this up (and probably didn't question what “transmission 7.5 times higher” could possibly mean): nbclosangeles.com/news/coronavir…

Now I don't want to question the soundness of mandating mask use (though I think the argument should be “the cost is very low” rather than “the benefit is very high”), but there's a serious problem in how this kind of thing gets cherry-picked in a paper's conclusions.

In particular, if you accept this paper as evidence that mask use is very effective, you should also accept it as evidence that lockdowns are ineffective, and I'm not sure people who will put one conclusion forward will always agree with the other.