A very, very impressive cluster randomized trial. Very hard to do! See thread for findings.https://twitter.com/Jabaluck/status/1433036923610742789 …
-
Show this thread
-
It's usually hard to show any effect with this type of trial due to statistical power limitations. But they find some anyway, despite also having other significant factors pulling against finding an effect besides just the thorny, inherent difficulty in cluster RCTs!
1 reply 1 retweet 55 likesShow this thread -
They find effects despite neither reaching universal mask coverage (they can't force, only nudge and they reach 42%) nor having a control with no masking (since would be unethical) *and* being conducted in a much lower-risk setting—rural Bangladesh, few crowded sealed buildings.
2 replies 4 retweets 53 likesShow this thread -
They find that, even with a design necessarily biased against finding effects due to statistical power limits and ethical barriers to having a no-mask control group, protective effect of masks is stronger for surgical masks than cloth masks, and significant for the elderly.
2 replies 5 retweets 82 likesShow this thread -
Honestly, when the authors told me, I was skeptical it could be done and the design limits would overpower— unavoidable: cluster randomization has limited power especially for overdispersed phenomena and you can't confine people to one village or order controls not to wear masks.
1 reply 0 retweets 49 likesShow this thread -
Do note that when you have a study design that really works against being able to detect an effect, the key interpretation is relative—within study. Don't read anyone who doesn't understand statistical power, cluster designs, overdispersion and the ethical limits of controls.
3 replies 2 retweets 58 likesShow this thread -
Parts that surprised me: surgical masks retain filtration through many, many washing. Part that did not surprise me, almost nothing the (pre-polled) policy makers thought would work to nudge mask use higher worked. Policy folk and actual human behavior don't talk much, sadly.
5 replies 13 retweets 107 likesShow this thread -
Part that doesn't surprise me: people who will conflate what a randomized cluster trial measuring an ecological effect in a real life setting with a randomized *clinical* trial (where we control everything and don't deal with overdispersion and only measure individual benefits).
3 replies 2 retweets 49 likesShow this thread -
I'll try to write longer eventually, but just wanted to put it here. Any discussion you trust should mention measuring ecological effects in real life, cluster trials, overdispersion, limits to controls, statistical power and difference with clinical trials. Glib tweets won't do.
5 replies 6 retweets 61 likesShow this thread
Also congratulations to the authors. I honestly did not think they could pull this off, but they did an incredible job despite all the hurdles: unavoidable ones for statistical power, funding, ethical limits (which they respected, of course), and, wow, implementation! Amazing.
-
-
Replying to @zeynep
Leonhardt (Times Sep 7 2021) does use 3 month period, but not for 1 in 5,000 risk per day (2% risk 3 month) offers 1 percent risk. Says "the risks of getting badly sick remain minuscule" WTF? CDC studies say breakthrus yield 1/2 to 1/3 rate of hospitalized as unvax'd.
0 replies 0 retweets 0 likesThanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.