A friend recently introduced me to the newsletters of Tom Woods, a libertarian author with an impressive pedigree (A.B. from Harvard, Ph.D. from Columbia). This is his latest newsletter. You kind of know what the slant of an essay with the title “Pretend this COVID chart doesn’t exist, or get canceled” is going to be, but I thought well, if he’s using valid statistics, it’s worth having a look. The first thing you might notice in his piece is that he shares numbers and graphs without giving sources. I would assume that citing sources would be second nature to someone who had to do the academic research to attain a Ph.D. in history from Columbia, so I’m guessing the absence of sources is intentional. Perhaps requiring source citations is impinging on his rights. At any rate, I’ll be citing my sources as I respond to Woods’s points.
Woods does cite his source for this Vanity Fair article from May that suggests that if 80% of Americans wore masks in public at all times, infection rates would plummet. Not go away as he implies, mind you, but plummet. Reading the article feels like reading history at this point, but one of the takeaways of the study, which uses computer simulation, is that if “80% of a closed population were to don a mask, COVID-19 infection rates would statistically drop to approximately one twelfth the number of infections—compared to a live-virus population in which no one wore masks.” Woods refutes this claim with a statistic of his own, saying that mask compliance is at about 88% in the US “at this time.” I’m not sure where Woods got his number, but I found a CDC report that says in June 2020, 89% of adults reported wearing face masks, while other mitigation strategies were found to be in decline or unchanged. Well, the first problem here is self-reporting. I’m sure we all know someone who says they are self-isolating, but when you look into it, that includes all sorts of mingling behavior that they somehow don’t think counts. And there are bound to be people who are shamed by the questioner into saying they wear masks, or people who wear masks when absolutely forced, like in a supermarket, but not when they’re partying with friends, yet answer “yes” to the question. The biggest problem with using the statistic in a country the size of the US, however, is the “closed population” rule. Different parts of the country have very different rules and norms regarding any of the mitigation measures. Given the free movement of people through high- and low-transmission areas, and the fact that the people least likely to wear masks seem to be the people the most likely to mix in big groups, we’ll never have that ideal closed population. But even if we did, the article says the number of infections would drop to 1/12 of what it would be without mitigation measures. So, for example, New Jersey, which with pretty strong rules and pretty good mask compliance currently has about 6,300 cases, could have 12 times that number, or 72,000 daily cases, less than 1% of its population, without mask wearing. This seems high but not impossible. Our mitigation strategies have been going on for so long that we don’t really know how quickly the virus would burn through the community if everyone were out and about as they normally are.
It seems funny that all of a sudden, Woods is concentrating on case numbers, though, since he and others have been arguing for herd immunity all along, which would mean high case numbers were good. The only valid numbers, this argument goes, are hospital numbers, hospital capacity, and death. There have been studies, like this one by UCSF’s Dr. Monica Gandhi, that suggest that the real benefit of masks is in lessening the viral load of mask wearers, thereby making them less sick. But Woods is concentrating on cases and making plenty of misleading statements even with that limitation, so I will respond to his points.
His first point is that counties in Arizona that had mask mandates had an almost identical case rate as counties that don’t. He uses a nice, unsourced graph to prove his point, which I’m not going to repeat here, but you can see it in his newsletter. The graph shows almost identical curves for the number of cases in counties with mask mandates and those without mask mandates. Let’s assume his numbers are correct. Why might the graphs be largely the same? Well, my discussion of a closed population is even more important when you’re going county by county. When I lived in the United States, I crossed through 2 states and 3 counties to go to work. If my state or county had the strictest controls in the world, but I spent most of my time in a state or county in which no one wore masks, socially distanced, or recognized that there was a virus, my experience at work would be more indicative of mask-wearing efficacy, but if I got sick, I’d be counted in my home county.
“Cop out!” Woods might say. But there’s an even greater reason, statistically, for the similar curves. I couldn’t find the county-by-county data with which he presumably constructed his curves, but I did look into which counties in Arizona had mask restrictions and which didn’t. According to the Phoenix Business Journal, the six Arizona counties (out of fifteen) that have mask mandates are “home to most of the state’s population.” Overall, they made up 83% of the state’s population. So once we allow for movement between homes and work, or movement for other reasons, it doesn’t seem surprising that 17% of the population has a similar COVID experience to the 83%. As a matter of fact, if anything, since many of those non-mask-wearing counties are quite sparsely populated, we’d expect them, even without mask mandates, to have a better COVID experience than places like Maricopa County, which includes Phoenix, given the natural social distancing that occurs when people don’t live on top of one another. Of course, Woods has already discounted my rational response to his argument, thus implying that I am the one manipulating the data to my advantage: “There are too many other factors at work between two given places, so it’s not fair to compare their outcomes with and without masks, we’re told.” Well yes. It’s this turning around of valid arguments that makes me see red when I read essays like Woods’s. He’s obviously preaching to a supportive (and statistically naïve) audience that wants to believe it’s all a con job—I’ve still not figured out what the advantage is, or how the US managed to convince the world to be in on the con, but no matter, somehow, this line of thinking goes, people are being misled if they think that masks (in this case) will make a difference. When someone points out that no one ever said they were 100% effective, and indeed, just about every scientist says social distance is the first line of defense, but masks are a good back-up when you can’t socially distance, Woods has an answer for that, too. He tells us: “The mask religionists are falling back on what they call their ‘Swiss cheese’ model of virus mitigation. I would explain that to you, but really. (It starts with, ‘We never said masks alone would do the trick!’)”
So it appears that some crazed people, or maybe people with shares in a mask-making company, have invented a cockamamie theory called the “Swiss cheese” model, right? Wrong. According to the Cleveland Clinic, the Swiss Cheese Model was introduced in 1990 to explain in simple terms the common-sense idea that several mitigations for a dangerous situation provide more protection than just one. So, if you’re trying to decrease serious injuries to bicyclists in your city, you might add bike lanes, which will make bike riding quite a bit safer, but not perfectly safe. You might require helmets, which will make it less likely that someone who does have a collision despite the bike lanes sustains serious injury. Perhaps you require bike safety classes, or stronger lights for the bicycles, or any number of other measures, each of which will add to the safety of people riding bikes in your city. The Swiss Cheese Model is a perfect tool for describing COVID-19 mitigation strategies. If you look at the Cleveland Clinic’s graphic (below), you’ll see that in a business environment, you can expect mask wearing to lessen the spread of COVID, but not eliminate it. But if you also socially distance, even the COVID that gets through the “hole” in mask wearing will be less likely to get to another person. But, some will still get there, so disinfecting work surfaces makes infection even less likely, etc.:
This concept was not invented during the current pandemic—and it’s a perfectly rational way to explain why procedures are in place for a variety of dangers. I’d suggest that Woods’s biggest reason for not explaining this model is that most people would understand and agree with it.
Now let’s look at the graph Woods thinks will prove that the media are biased and mitigation doesn’t work. I have to reproduce this graph to discuss it, but the only thing I am using it to explain is how not to present data. Once again, it’s sourceless, but it jibes pretty well with Johns Hopkins data, so I’m assuming that’s where he got it:
In this graph, Woods is comparing the New Mexico case numbers to Iowa’s to make a couple of points. First, he thinks it proves that not doing anything is just as good as doing something at stopping the spread of COVID-19. Second, he is trying to persuade the reader that the media have colluded to present a picture that is belied by the data. So if I say this is valid Johns Hopkins data, what’s the problem?
Well, the first thing that confused me was the placement of the two article boxes. If you look at the arrows, it looks like “Iowa Is…” was published around the 8th of November, and “How New Mexico…” was published when the slope was going down, sometime around November 28. But if you look at the publication dates, the Scientific American article about New Mexico was published on September 15, before this graph even starts, and the Atlantic article on Iowa was published on December 3. So New Mexico wasn’t written as case numbers were increasing, although the Iowa article was published as their numbers were decreasing.
That’s the second odd thing about these graphs. Not only was the Scientific American article published months before its placement on the graph would indicate, it was actually published before the start of the graph. Why did the graph start on October 1, at the start of the second wave, if he’s referencing an article printed on September 15? Perhaps because showing the whole graph gives different context, especially to the Scientific American article. Here’s how the whole New Mexico graph looks:
So when Christie Aschwanden wrote her article for Scientific American, New Mexico had a cumulative total of 26,923 cases and 830 cumulative deaths, making it a success compared to most states. Note that the colors of the dotted lines correspond to restrictions being put in place or halted. You can see that when the numbers started to rise in October, New Mexico, which, like most other places, had been able to open up over the summer, put restrictions in place again. If you have time, go to the actual Johns Hopkins site, because you can see a lot more when you look at the dynamic graph, including what kinds of restrictions were put in place when, and how the numbers followed the predictable lag after restrictions were instituted. Restrictions were reintroduced on October 16, statewide measures were toughened on November 13, and case numbers peaked on November 23, then started decreasing. This tells me that restrictions work.
But what about Iowa? Unlike the Scientific American article, the Atlantic article is too anecdotal and emotional for me, so I wouldn’t have equated the two, but then I’m not trying to find reasons to castigate all of the mainstream media (and who knew Scientific American would be included in that? But I guess science is the enemy). If we look at the entirety of the Johns Hopkins graph, it doesn’t look remarkably different from New Mexico, although we can see some worrying blips lagging the decisions to open back up:
Obviously, Iowa, with a total population of 3.18 million, or less than half of New York City alone, is likely to have ICU capacity below that of New York, but it still seems somewhat hyperbolic to say that to “visit Iowa right now is to travel back in time to the early days of the coronavirus pandemic in places such as New York City and Lombardy and Seattle, when the horror was fresh and the sirens never stopped.” New York at its peak was registering 1,000 new deaths per day, while Iowa’s daily deaths reached their peak at 100. Although all the data in the article are correct and sources are properly cited, by the time the article went to press on December 3, the governor’s mask mandate had been in effect for two weeks and case numbers were falling. All this is to say that while Woods might be justified in calling out the Atlantic article as being somewhat slanted, the data certainly doesn’t support his anti-mask mantra.
If you’ve read this far, you’ve noticed that I’ve taken about four times the length of Woods’s original article to refute it. It seems to me that this is one of the problems scientists and statisticians have had all along. They can’t just make statements without backing them up, and they have to acknowledge when the data don’t show what they’d like them to show. They speak the language of confidence intervals and probabilities rather than certainties. So while Woods can tell you, for his own reasons, that masks don’t work, a statistician or scientist will never say it’s 100% certain that they do work. I’ve also heard scientists say repeatedly that it is not their place to make policy decisions; they are there to inform the government and the government makes the decisions. Unfortunately, both the UK and US governments, for different reasons, have made a great many irrational decisions and the scientists have been blamed for statements they never made. But at this point, the overwhelming majority of scientists and statisticians believe that masks are an appropriate mitigation strategy. The Australian comedian Jim Jefferies has a routine on gun ownership, in which he ridicules gun owners’ stated reasons for needing to have guns. There’s one reason, he says, for having a gun, and it’s “f- you, I like guns.” You could replace guns with not wearing a mask. If you don’t want to wear a mask, it’s likely because you don’t like masks and you don’t like being told what to do. Don’t try to dress it up with cherry-picked statistics and false assertions.
I also finally twigged to the reason Woods has written this essay now—Joe Biden has said he wants to enact a 100-day mask mandate upon election. This, to someone with Woods’s libertarian sensibilities, is government overreach. Well, then, stick to that as your argument. Don’t try to make the data prove something they don’t.