An intuitive way to grasp the effectiveness of masks: extrapolating from our results, every 600 people who wear masks for a year in public areas prevents 1 person from dying of COVID given status quo death rates in the US.
Note that this is *taking into account current vaccination rates in the US*. Despite the availability of vaccines in the US, the weekly death rate is higher than at any point prior to November 2020.
Here is how I arrived at this number. Our study shows that inducing a 30 pp increase in mask-use prevented 35% of COVID cases among the elderly.
First, let's assume that we see a proportionate reduction in mortality. In the US, death rates for the past year and a half (lower than in the past week!) have been 1,200 people per day.
Of these, 80% are among those 60+. Our 35% number would imply about 120,000 preventable deaths.
A 30 pp increase in mask-use among adults is ~75 million people wearing masks regularly. Dividing 75 million by 120,000 gives ~600 people that need to wear masks for a year to prevent each death.
Let me first contextualize this number and then run through some caveats. The average American knows about 600 people by name that they associate with (not celebrities). nytimes.com/2013/02/19/sci…
That means that if you and every person you know by name wear masks in public areas for a year, on average, it will prevent one of you from dying of COVID.
Caveat #1: of course, there is substantial heterogeneity across places and times. If you live in a place where few people are dying of COVID (e.g. due to higher vax rates), the benefits are smaller. If more people are dying, they are larger.
Caveat #2: this doesn't mean you need to wear a mask 24/7. Our study doesn't assess this. Based on other work, I suspect that mask-wearing in crowded, indoor environments with poor ventilation is especially valuable.
My best guess is that mask-wearing outdoors has few benefits outside of very crowded areas, like rock concerts, rallies, sports events, or the kind of crowded markets we see in Bangladesh where people are breathing directly on one another.
Caveat #3: there are numerous external validity issues depending on the operative mechanism in our original study. For example, if part of the effect comes from the fact that people with masks physically distance more, would this be equally true in the US?
This kind of calculation is very far from definitive. But it's also not "fixed" one way or the other. I didn't choose the numbers and assumptions to get a desired result. You can get something that looks even better for masks if you use death rates in the last week.
I plugged in the numbers from the study plus reasonable assumptions in order to produce our best guess estimate of the impact of masks on mortality, and that impact appears to be very large given status quo vaccination rates in the United States.
The value of masks in places where nearly everyone is vaccinated is clearly lower. Caution is necessary since, in many parts of the world, the vaccines being used have lower efficacy than in the US, meaning that masks likely have value on top of vaccines.
Additionally, masks may prevent breakthrough cases which may eventually spread to unvaccinated people. However, I haven't seen a quantitative calculation of the magnitude of this benefit -- it may be small.
The idea that we should patiently educate the aggressively ignorant sounds laudable, but it practically means disengagement. How many who liked the above post consistently attempt to do this? They might try once, but they'll give up because it's too time-consuming.
For those keeping track at home, this is definitely not what a confidence interval is. A 95% CI is a function of the data such that, given the data generating process with an unknown true parameter, the CI constructed in this way will contain the true parameter 95% of the time.
The idea that all values in a 95% CI are equally likely is preposterous. If one were instead constructing Bayesian credibility intervals, you do not need a gaussian prior to rule this out.
In the Bayesian problem, this would represent an absurd corner case where the data was completely uninformative about the underlying parameter within a specified range. I can't imagine how this would be a reasonable model of the situation at hand.
Firstly, our study does not say that masks can only prevent 11% of COVID. Our study says that our intervention -- which raised surgical mask-wearing from 13% to 43% -- prevented 11% of COVID cases, and 35% among age 60+.
To put the point on your own terms -- if you vaccinated 30% of the population, would you prevent 35% of cases age 60+?
I do agree that vaccines are probably even more effective than masks but there are three very important subtleties here that make the 11% way too conservative @ProfEmilyOster.
First, the 11% comes from a 30% increase in mask-use. The IV estimate (naively scaling things linearly) would thus be more like a 37% reduction in COVID from going from zero to universal masking.
Second, we find much larger effects among the elderly (a 35% reduction among 60+ without the above scaling). This suggests that the total reduction in morbidity and mortality from universal masking may be considerably larger than even the 37% number, perhaps more than 50%.
1) We only studied adults, but for both surgical and cloth masks, we find impacts on COVID symptoms at all ages tested
The fact that we do not find an effect blood tests of age<50 is likely due to the fact that we collected blood only from a subset of symptomatic people and so our estimates on blood-test confirmed COVID are less precise.