To a vocal group of people around the world, masks are the civil rights issue of our era. According to this line of thinking, mask mandates are a slippery slope that will embolden the government to encroach on other freedoms we currently take for granted. 

It would be a mistake to put this opposition to masks down to lack of education. Some of the loudest skeptics believe they are following the science, although the scientists they trust tend to have a great deal of professional baggage. As you might expect, our old friend, the bestselling COVID contrarian and former New York Times journalist Alex Berenson is one of the most vocal opponents to masks. Although he no longer claims that COVID-19 is largely harmless, he does take issue with public health measures designed to prevent the virus’ spread. In part three of his pamphlet, Unreported Truths about COVID-19 and Lockdowns, Berenson lays down a series of arguments against wearing masks, with an impressive array of references to scientific studies.

At first glance, Berenson’s arguments against masks may appear persuasive (especially to people already inclined to agree). He questions their filtration efficiency, arguing that they let through too many virus particles to be of any use. He also points to a study from Denmark that failed to come to a firm conclusion on whether surgical masks protect the wearer in everyday life. As before, this can be quite convincing, as few of us have the time or the knowledge to go through the literature and fact check every claim.

So what do we actually know about masks? And do Berenson’s claims hold up?

The Danish Study

There are generally two types of trial in medical science. The first is the observational trial, which compares existing cohorts of patients undergoing different types of treatment. Although these trials are informative, they’re susceptible to bias because the researchers are looking at existing patients without any control over who gets what treatment. The second type of trial—regarded by scientists as the gold standard—is the randomized controlled trial, which compares a medical intervention to a placebo. Because individuals are randomly assigned to one of the two groups, each is representative of the population as a whole, which makes the comparison fairer.

Berenson’s primary evidence that masks don’t work is a randomized controlled trial conducted in Denmark last year that aimed to find out if surgical masks offer protection to the general public. To do so, they recruited 6,000 volunteers who worked at least three hours per day outside their home, and randomly chose half of them to wear masks for a month. At the end of the period, the volunteers were tested for COVID-19. The relatively large sample size, and the fact that the trial was of the randomized controlled variety, might seem to give Berenson’s argument a credibility boost.

However, the study did not look at masks’ protection to others, or the aggregate effect of widespread mask usage.  Despite these limitations, Berenson believes that the study puts an end to the mask “debate”:

Mask wearing “did not reduce, at conventional levels of statistical significance, the incidence of Sars-Cov-2 infection,” the authors wrote in their discussion. Unless future large randomized controlled trials find different results, the Danish mask study essentially should end the debate if surgical masks protect people who wear them outside hospitals.

Designing and carrying out a medical trial is difficult in the best of times, so let’s pick this one apart. When I’m evaluating a trial, the first thing I look at is if it has the statistical power needed to measure what is being studied. There’s always a degree of randomness in the outcome of a trial because whether or not a treatment works depends, to some extent, on chance. If there are too few participants, there’s a good chance that you won’t find any difference between the treatment group and the control group even if the intervention is effective. To have a high statistical power, a trial needs enough participants to ensure that the difference between the treatment group and the control group is much larger than the fluctuations you would expect from random chance—that is, to achieve statistical significance. 

Conventionally, medical studies are designed with 80 percent power, which means that the study has an 80 percent chance of a statistically significant result. To calculate the power, you first need to estimate how effective the treatment is. The authors of the Danish study designed it under the assumption that surgical masks are 50 percent effective, and although they claim 80 percent power, they give no indication of how they calculated that figure. This is worrying, because although I see a lot of medical doctors among the authors, I don’t see anyone who’s an experienced statistician. According to my own calculations, the study only achieves this power if you make the assumption that we have a totally accurate test for COVID. This is unrealistic, as even the most reliable PCR tests—which detect viral RNA—give a true positive result in only eight out of 10 COVID-19 cases, and give a false positive in around one in 100 people who don’t have the disease.

This brings me to the next thing I look for in a study: how well is it measuring what it claims to be measuring? At the start of the Danish trial, individuals with COVID were screened out with a self-administered blood test that detects IgM and IgG antibodies—biochemical markers produced by the body to tag the coronavirus for destruction. But these antibodies don’t appear until an average of 18 and 20 days post infection, and then remain for over a month. This means if you caught COVID two to three weeks ago, the antibody tests are unlikely to detect it. According to PCR tests—which detect coronavirus RNA directly and don’t suffer from a time lag — the average COVID infection lasts 31 days, so the antibody tests will have missed around half of the subjects with an active infection. With 2 percent prevalence of COVID-19 reported in Denmark at the time, around 1 percent of the trial subjects will have been COVID positive at the start of the trial due to false negative test results.

The second COVID test at the end of the trial is another source of error: the test manufacturer reports that the false positive rate is 0.8 percent. Combining the two sources of error, you would expect 1 to 2 percent of participants to test positive even if they were locked in a hermetically sealed room for the duration of the trial. In other words, the experimental error is almost as large as the 2 percent prevalence of COVID in the general population.

Taking false negatives and false positives into account, I recalculated the statistical power of the trial to be 36 percent.  This is likely to be an overestimate, as I haven’t attempted to account for low compliance in the mask-wearing group. Because the trial was conducted by correspondence, there was no reliable way to check that participants were wearing their masks as instructed—and when asked, only 46 percent reported wearing the masks as recommended.

It’s also worth pointing out that the authors of the Danish study didn’t conclude that masks are useless (emphasis mine):

The recommendation to wear surgical masks to supplement other public health measures did not reduce the SARS-CoV-2 infection rate among wearers by more than 50% in a community with modest infection rates, some degree of social distancing, and uncommon general mask use. The data were compatible with lesser degrees of self-protection.

All in all, the Danish study was very unlikely to produce a statistically significant result, and it’s not alone—a systematic review from the New England Complex Systems Institute found that other mask studies that failed to find an effect were similarly underpowered. And we’re unlikely ever to have a randomized controlled trial that definitively shows that masks protect the user and others from COVID, because to conduct such a study under the controlled conditions needed would mean putting people in unnecessary danger.

Is Incomplete Protection Useless?

Elsewhere in Unreported Truths, Berenson acknowledges that surgical masks provide some protection:

The scientists found the surgical masks barely worked. Masks with ties filtered about 70 percent of small particles. Those with ear loops filtered less than 40 percent and often had “visible gaps between the face mask and the wearer.”

But he then compares wearing a mask to being inside a leaky submarine:

Based on the pictures [study leader Dr. Osterholm] saw of people wearing masks, about one person in four was wearing them wrongly, the equivalent “of fixing three of the five screen doors on your submarine.”

It’s true that surgical masks can’t offer complete protection, but is it really fair to compare the risk of catching COVID-19 to drowning in a poorly constructed submersible? It’s important to distinguish protective equipment for use by the general public from that used by health workers. In a COVID ward, there’s almost certainly aerosol coronavirus floating about, which is why nurses and doctors need properly sealed N95 respirators and face shields. N95 masks are so-called because they have a 95 percent filtration efficiency at the most penetrating particle size—anything less and the medics would soon become sick in even larger numbers than we’ve seen.

But in everyday life, the risks are far lower. As long as we socially distance and avoid crowding into poorly ventilated areas, we’re generally limited to infection by a random passerby. Although Berenson doesn’t provide a citation for his figure of 40 to 70 percent filtration efficiency of surgical masks, this is the range generally reported in the literature, even taking into account the incomplete seal that such masks provide. This is actually a decent amount of protection for the general public. 

It appears that that severity of infection depends on the dose of virus received, and although we don’t know the exact threshold, we can be pretty sure that not every encounter with a sick person results in infection: to give a ballpark figure, 7 people in 1,000 become clinically ill after encountering someone with COVID in daily life. This suggests that the viral load transmitted at such an encounter is marginally infectious, so even a mask that filters just 40 percent of viral particles can greatly reduce the spread. Moreover, focusing on filtration efficiency alone misses the fact that masks slow down exhaled air, which makes water droplets containing the virus less likely to reach people around us.

But surgical masks aren’t universally available, and they produce a lot of waste. As long as they’re made right (more on that in a second), washable cotton masks are a more accessible and sustainable option.

Do Homemade Masks Work?

Wearing a cotton mask to stop the coronavirus is like putting up a chain-link fence to stop mosquitos, according to one popular meme. The image below shows the material in an N95 respirator (left) and a standard lightweight cotton (right) taken with a microscope. Each subdivision on the scale bar is 200 microns (the width of a fine grain of sand).

Now, my eyesight isn’t very good, but I struggle to see the similarity to a chain-link fence. It’s true that there are some rather large holes in the cotton mask, but our mosquito would have to search around to find one. Although the coronavirus measures around 0.1 microns, when it’s expelled from the mouth, it’s usually contained in water droplets that are between 1 and 300 microns. Some of these will make it through the mask, which is why the best homemade masks have at least two layers of cotton.

But let’s go a bit deeper and really try to understand what it’s like for a virus to be launched out of your mouth at a semi-permeable barrier. The microscopic world is random and bouncy, and hard to empathize with for those of us who have outgrown inflatable funfair attractions. It’s not as hard to visualize as you might think, though—now, let’s put ourselves in the virus’ shoes.

Imagine you’re a virus, freshly liberated by a sudden cough. Most of your buddies are lumbering along in huge droplets of water, but you’re a free soul! OK, you’ve got a few freeloading water molecules stuck to you, but apart from that the world’s your oyster. You’re heading directly for that pesky barrier humans like to put over their face, and you like your chances of making it through one of the pores. But before long, you find you’re being buffeted every which way by little air molecules. You’re zigging, zagging, and spinning so much you don’t even know which way is up or down.

You’ve somehow made it to the cotton barrier, bouncing your way through a gap, but with each collision you’re losing speed. As you near the second layer of cotton, things start to go badly wrong. Through no choice of your own, the little electric charges that keep you together disperse away from the cotton, leaving you electrically polarized. As if drawn by an unseen force, you find yourself pulled toward the giant cable looming above you. Thunk! Before you know it, it’s all over. The little charges on your surface keep you there, and no amount of positive thinking can get you unstuck.

(Imagining being a mosquito is left as an exercise for the reader.)

So multi-layered cotton masks are a reasonably effective barrier, but they lack one thing that makes N95 respirators so effective. As I alluded above, electrostatic forces are very powerful when you get to the size of a virus (far stronger than gravity, in fact). The virus sticks to cotton because of the uneven charge distribution on both the molecular strand of cotton and the virus’ surface, but you can get an even stronger effect with a material that carries a net electric charge.

Most people have experienced a mild electric shock after their clothes have built up a static charge, and you’ve probably noticed that this tends to happen with synthetic fabrics rather than cotton. N95 masks are designed with this effect in mind: they’re made with extremely fine fibers of polypropylene plastic that readily holds a charge.

In a recent study, researchers at the State University of New York set out to find household materials that would hold a charge just like the polypropylene in an N95 mask. They collected air that was filtered through a variety of fabrics with a particle counter, and measured the filtration efficiency of each fabric. I’ve reproduced their results below (the particle size is in microns).

Although none of the home-made masks worked as well as an N95 respirator, just including a dried out baby wipe in a cotton mask filters out 80 to 90 percent of even the finest particles. And even a plain cotton mask is highly efficient at filtering out larger droplets.

So the mechanistic evidence we have points to homemade masks being a reasonably effective barrier for COVID-19 transmission. They’re not good enough for frontline healthcare workers, but they’re a sensible precaution for members of the general public who face transient encounters with the virus. And even if masks aren’t as effective as we think, we have no reason to believe that they cause any harm.

Do Masks Make Classrooms Safe?

Because of widespread school closures and a lack of government aid, parents are struggling to balance childcare with work responsibilities. What’s more, without free or subsidized school meals, the number of households with children going hungry has increased from 14 percent to 28 percent, and as children are stuck at home we’re seeing more and more reports of domestic violence.

The solution? Alex Berenson suggests that teachers need to step up and face the risks:

1: The evidence is pretty strongly that [kids] aren’t “silent spreaders.” 2: Unless they’re hanging out at nursing homes, who cares? “Silent spreader” is just another way to try to keep schools closed EVEN THOUGH KIDS ARE AT ~ZERO RISK. We run schools for kids, not teachers…

Any teacher too dumb to understand that she’s not going to get virus from her students, and/or too scared of doing the job for which she’s paid… should find another line of work.

Is Berenson correct in claiming that teachers face minimal risk? 

Since children have largely been at home, they haven’t been significant spreaders of the virus. There’s very little information on kids infecting teachers with COVID-19, and there’s a temptation to confuse a lack of evidence with evidence that children aren’t spreaders. However, last October, the COVID-19 National Emergency Response Center in South Korea reported that 10 to 19 year olds who have COVID infect 19 percent of their household contacts, the highest rate out of any age group. For zero to 9 year olds, the rate was 5 percent, although this number isn’t certain due to the low number of children in this age range included in the study. And just like adults, children can be asymptomatic spreaders. These numbers may increase with the new strain that emerged in England last year, which is up to 70 percent more infectious according to European health authorities.

In any case, it is clear that classrooms (like any place where humans gather) aren’t magical sanctuaries from the coronavirus. When you cough, you produce thousands of droplets that travel over six feet. Sneezing is more powerful, producing tens of thousands of droplets that may travel 20 feet or more. We know from an experiment where subjects were forced to sneeze into a laser beam (by stimulating their noses with an irritating unguent) that these droplets measure between 50 microns and 600 microns when first expelled. The larger droplets fall to the ground in a few seconds, but the smaller ones evaporate to form droplet nuclei measuring around 1 micron. 

And you don’t even need to cough to produce such droplets. As a team from UC Davis found:

These small particles are believed to be generated during breathing and talking from the mucosal layers coating the respiratory tract via a combination of a “fluid-film burst” mechanism within the bronchioles and from vocal folds adduction and vibration within the larynx. The particles emitted during breathing and typical speech predominantly average only 1 μm in diameter and are thus too small to see without specialized equipment… Despite their small size, however, these micron-scale particles are sufficiently large to carry a variety of respiratory pathogens… Indeed, recent work by Yan et al. has confirmed that significant amounts of influenza viral RNA are present in small particles (<5 μm) emitted by influenza-infected individuals during natural breathing, without coughing or sneezing.

In other words, you get micron-sized particles generated in the throat and lungs—two of the coronavirus’ favorite locations—that are known to carry pathogens.

Because gravity hardly affects such small droplets, any coronavirus held inside is free to float through the air, where it remains active for hours. We know that the virus can propagate through the air this way because two separate studies have found coronavirus RNA in hospital air conditioning ducts, potentially making this a leading source of infection. As Berenson argues in his book:

The fact dormitories, prisons, ships, and other “congregate” settings can see very high rates of coronavirus infection — as high as 90 percent — also provides strong real-world evidence that SARS-Cov-2 spreads through tiny aerosol particles that stay in the air for long periods.

The closest analogy I can give is cigarette smoke: it’s quick to dissipate outside, but permeates indoors and is hard to get rid of. And just like smoke, these micron-sized droplet nuclei can reach deep into the lungs, taking the coronavirus directly to where it causes the most damage.

So can we really make schools safe for teachers? We know that children can be asymptomatic carriers, and we can be pretty certain that COVID-19 can spread through speaking and breathing. Giving the kids masks might help, but even with 100 percent compliance (a doubtful proposition to anyone familiar with children) you’d still get some airborne coronavirus in the classroom. To be reasonably safe, the teacher would need a properly fit-tested N95 mask—but if you’ve ever worn one you’ll know that they’re rather stuffy and restrict your breathing. It’s hard to imagine that the teaching would be any better than the much-reviled Zoom lessons.

I can’t pretend that there’s an easy solution to child wellbeing during lockdowns, but given that 40 percent of teachers and adults living with school-age children have definite risk factors for severe COVID-19, reopening schools isn’t the simple solution that Berenson would like it to be. But now that we have increasing supplies of vaccine, why not give teachers priority? After considerable feet-dragging, teachers in the U.S. are now eligible to be vaccinated, although other countries have been less responsive.

Following the Evidence

There’s no question that N95 respirators protect frontline healthcare workers from COVID (even Alex Berenson does not dispute this). Yet this will never be proven in a randomized controlled trial, because to do so would put medical workers at risk. We also know that you can make a mask with household materials that’s almost as good a filter as an N95 respirator, except it won’t provide a complete seal. And again, we will never know exactly how protective homemade masks are, because there’s no ethical way to conduct a trial with the necessary controlled conditions.

But clinical trials aren’t the only form of evidence. As I discussed previously, numerous studies have looked at virus cultures from patients who were asked to cough into a petri dish, with and without a mask. Other studies have rigged up breathing mannequins and looked at how well masks protect them from synthetic microscopic particles. When a team from Arizona State University surveyed these mechanistic studies last year, they came to the following conclusion:

We therefore estimate that inward mask efficiency could range widely, anywhere from 20 to 80 percent for cloth masks, with 50 percent possibly more typical (and higher values are possible for well-made, tightly fitting masks made of optimal materials), 70–90 percent typical for surgical masks, and 95 percent typical for properly worn N95 masks. Outward mask efficiency could range from practically zero to over 80 percent for homemade masks, with 50 percent perhaps typical, while surgical masks and N95 masks are likely 50–90 percent and 70–100 percent outwardly protective, respectively.

Even if we assume only 25 percent inward and outward efficiency, the combined effect of both spreaders and susceptible bystanders wearing a mask translates to a 44 percent reduction in the spread of the virus. This is not good enough for frontline healthcare workers, and may not be enough to protect teachers spending hours in crowded classrooms, but together with social distancing masks offer incredibly effective protection for the general public.

None of this information is particularly new, so why did it take so long for masks to be taken seriously as a public health policy? There’s a worrying trend in medicine—paradoxically called evidence-based medicine—to disbelieve any result that doesn’t come from a randomized controlled trial. This makes total sense when evaluating pharmaceuticals because their effects on the human body are wholly unpredictable. But it is an unrealistically high barrier of evidence for interventions that are known to work based on mechanistic evidence and clinical experience. Had we followed the evidence and adopted masks sooner, countless lives could have been saved.