I can't recall if I've posted a link to the study at
https://wwwnc.cdc.gov/eid/article/15/2/08-1167_article before but I was reminded of it because I referenced it in a recent debate with a colleague who is in the "masks don't work" crowd.
It was published in 2009 so it pre-dates COVID-19. The condition the investigators were looking at is influenza-like illness (ILI). But I think their experience is instructive. It is an actual randomized, controlled experiment. Here, from the abstract, is the statement that really caught my eye:
We found that adherence to mask use significantly reduced the risk for ILI-associated infection, but <50% of participants wore masks most of the time. We concluded that household use of face masks is associated with low adherence and is ineffective for controlling seasonal respiratory disease.
In other words, mask use worked. But the treatment did not because the treatment assigned was household use of face masks and when they assigned households to use masks they didn't get a high rate of mask use. The takeaway is that, if ONLY they could get people to actually wear the masks most o the time, it would work. But the reality is they couldn't get enough people to wear masks most of the time.
They looked at P2, which are N95, and surgical face masks. If you go to Table 5 at
https://wwwnc.cdc.gov/eid/article/15/2/08-1167-t5, you can see that they combined the two for overall mask use and they got 99% confidence that mask use in general reduced risk WHEN people actually used them.
When it comes to looking at each type of mask individually the confidence levels they report for the N95 is 94% and the one they report for the surgical mask is 91%. But you can tell from the confidence interval statements that they are using two tailed tests. I would argue that using a one tailed test would be reasonable and if you do that the confidence level for the N95 is 97% and that for surgical masks is 95.5%.
