An actual randomized control study on masks and Covid:
With dozens of researchers at Yale, Stanford, Berkeley and IPA and several other organizations, we ran a cluster randomized trial involving almost 350,000 people and 600 villages in Bangladesh to assess the impact of community masking on COVID.
We conducted an intervention that increased mask-wearing by 29 percentage points using the techniques described here:
We ran a massive randomized controlled trial investigating many different strategies to get communities of people to wear masks -- we're now scaling up the strategies that worked in many regions throughout South Asia...A randomized controlled trial (N=341,830) conducted by IPA, doctors and mask engineers from StanfordMed with Bangladesh health researchers and officials increased mask wearing by a lot, and the behavior maintained after the intervention ended...[more details in link]
With this 29 percentage point increase in mask-wearing, we saw a 9% drop in serologically confirmed COVID. The reduction was larger in villages where we (randomly) used surgical masks than those where we used cloth masks; in surgical mask villages, we saw a 12% reduction in COVID overall and a 35% reduction among those aged 60+.
Since severe morbidity and mortality are concentrated among the elderly, this suggests that community-wide masking can be an extremely effective tool to combat COVID.
If going from 13/100 to 42/100 people wearing masks leads to reductions of the magnitudes above, near universal mask-wearing (as is possible with enforced mandates in some areas) might lead to substantially larger reductions. As noted, we find especially convincing evidence that surgical masks are effective. Cloth masks reduce COVID symptoms, but the effect we find on symptomatic infections (confirmed via blood tests) is driven by surgical masks. Cloth masks are likely better than nothing, but surgical masks or masks with higher filtration efficiency should be preferred to cloth masks where available. A longer discussion of our intervention is available here, along with the underlying working paper:
Link