Those who know more forecast very slightly better than those who know less. But those with the most knowledge are often less reliable. The reason is that the person who acquires more knowledge develops an enhanced illusion of her skill and becomes unrealistically overconfident.

—Daniel Kahneman

Daniel Kahneman’s book—Thinking, Fast and Slow—is one long demonstration of the severe limitations of our human brain. He catalogues a variety of cognitive illusions and shows how they lead to persistently irrational behavior—so pervasive that we usually do not even notice it.

One of Kahneman’s best sections is on the limitations of expert judgment. He is devastating on the subject of political pundits—whose predictions Kahneman describes as worse than random—as well as on certain professions such as stock brokers. There are, in fact, many areas of human endeavor in which the final outcome is determined largely by chance, but which our brain insists on seeing as a contest of skill or produced by predictable causes. In reality, the world is far more unpredictable, uncontrollable, and unknowable than we all like to believe.

Now, Kahneman does not discount the possibility of real expertise. This would be an absurd position, as anyone who has played an expert chess player knows. In a great many situations experience and knowledge lead to increased effectiveness. But there are also many situations in which this is not the case. The key difference is whether the environment is regular or not. In a regular environment—one showing predictable patterns, such as when playing chess—then expertise is a major advantage. But in situations that are not regular, and which do not follow any predictable pattern (such as the stock market), the predictions of experts can be worse than random.

This seems like a particularly timely reminder during this coronavirus crisis. Suddenly we find ourselves in a novel situation, without historical precedent, and we naturally and logically turn to experts. But the experts that I have heard seem to sharply disagree on many important things.

A simple example is school closures. While some consider it wise, since children can serve as vectors for diseases, others consider it unwise, since the danger posed to children by the coronavirus is small and many healthcare workers have children (and so might be able to work less if schools were closed). Something else to consider is whether it might benefit the community as a whole if children developed immunity to the virus, thus negating their ability to serve as vectors. If they are at extremely low risk, then this could save many lives. Michael Osterholm seems to think that it is a bad idea, while most governments are coming to the opposite conclusion.

To give you two highly divergent cases of expert disagreement, consider Neil M. Ferguson and John P. A. Ioannidis. Ferguson is one of the world’s foremost experts on epidemics, whose titles are so extensive that I will not even repeat them here. His speciality is in mathematical models of infectious diseases; and his models predict quite bleak outcomes. He predicts one to over two million deaths in the United States, depending on the policies adopted. He recommends a policy of suppression—basically, a maximum of social distancing, locking down the population—in order to prevent the worst case-scenario. If we want to save as many lives as possible, then we will have to seriously disrupt society until a vaccine is developed, which may take quite a long time. (Ferguson himself recently seems to have come down with the disease.)

Ioannidis—the director of the Stanford Prevention Research Center—has a very different message. He emphasizes how much we still do not know about the virus, and how potentially foolish our methods of dealing with it may turn out to be. To take a basic datum, we are still quite in the dark regarding the lethality rate of the virus. It is tempting to take the total number of cases and then divide it by the total number of deaths, and get an answer. But there is reason to suspect that this would severely distort the results. For one, many countries have limitations on tests, and so the total number of infected will seem artificially low. And even if tests are widely available, we still cannot know how many people may be asymptomatic or have such mild symptoms that they never register.

To determine this we would need to conduct large-scale, randomized tests of the population, to see whether subjects have either the active virus or antibodies against it. To my knowledge, no such test has been done. Without such data, we really have no hopes of establishing the lethality of this novel coronavirus, because we cannot know how many people get mild or asymptomatic cases. In Italy and Spain, the virus appears to be extremely deadly. However, this is almost undoubtedly a result of several factors: for example, both countries have elderly populations; and testing is limited to those with serious cases. (Another interesting factor is that many younger people live with their parents in these two countries.)

If you look at South Korea or Germany—where testing is more widely available—the evidence seems much more reassuring. The Spanish newspaper El País calls the low mortality rate in Germany a “medical enigma,” partly because Germany’s population is roughly as old as Spain’s. But it is entirely possible that it is a more accurate picture of the virus than other countries. In Spain there are 625 tests per million people; in Germany they have 4,000 and in South Korea over 5,000 per million. Logically, the greater the testing capacity, the more accurate the mortality rate. Furthermore, it is worth noting that even the best testing capacity might lead to misleadingly high mortality rates, since again tests are usually given to people with symptoms; thus the number of asymptomatic cases remains largely a guess. 

The only situation in which an entire population was tested, to my knowledge, was the Diamond Princess cruise ship. Of the roughly 700 infections aboard, there were 7 deaths, giving a lethality rate of about 1%, and about 20% of the roughly 3,700 people aboard tested positive. (About 50% of those who tested positive were asymptomatic.) If 20% of, say, the United States got ill, and 1% of that 20% died, this would translate to 700,000 deaths. This is obviously bad. However, the case of the cruise ship has several factors that would make it seem a worst case-scenario. For example, cruise ships are more densely populated than even the biggest cities, more time is spent in common areas, and air is commonly recycled; thus, the total infection rate of the virus would be unusually high. Further, the population of cruise ships is significantly skewed towards the elderly, which would make the death rate unusually high as well.

Ioannidis argues that if we adjust our numbers to account for these significant differences, then the total number of deaths in the United States will be 10,000. In his words: “This sounds like a huge number, but it is buried within the noise of the estimate of deaths from ‘influenza-like illness.’” And it is worth remembering that in a bad flu season 70-80,000 people die in a year. 

One can point to the dire situation in Italy as a counterargument, of course. But it is worth noting that a virus does not have to be particularly lethal to overwhelm the medical system; it just has to be quite infectious. This is what is known as the R0 number: the average number of viral transmissions per person. We believe the seasonal flu to have an R0 number of around 1, meaning that each sick person gets an average of one other person sick. But if the seasonal flu had a much higher R0 number, it could be enough to flood emergency rooms like we are seeing. This is because more total people would be infected in a much smaller window of time. Thus, the evidence still appears quite inconclusive as to the real lethality rate. 

A great many other things are currently uncertain. Can we get infected more than once? How will changing weather affect the virus? And how infectious is the novel coronavirus, exactly? Without essential data such as these, we are all essentially flying blind.

Another problem is that the intense focus on the virus may only exacerbate our ignorance, not remedy it. Naturally, headlines focus on the growing numbers of cases and the growing number of the dead. Yet if the news suddenly began to track deaths from heart attacks or traffic accidents, the results would also seem catastrophic. Even on a normal day in America, the news can make it seem as if we are living in a war zone. The ugly reality—which we normally prefer not to think of—is that tens of thousands of people die every day, for all sorts of reasons. The question cannot be resolved by simply measuring the cases that come to our attention. We need to measure vulnerability and lethality against the relevant total, and not as simply a number that keeps rising.

We are thus faced with a difficult choice. If we underestimate the virus, and Ferguson is correct, we will condemn many people to die. But if the numbers used in Ferguson’s models are wrong—and we have no way of knowing this yet—then the measures intended to counteract the virus may inflict more harm than benefit, maybe much more. The virus is an unknown quantity, but so is the damage that could result from government lockdowns. Are we locking unhappy wives in with their abusive husbands? Are we inflicting severe psychological harm on vulnerable people? And if the economy cannot bounce back from this disruption, what will happen to those whose situation was already precarious? And these are just the immediate effects—not the economic or social fallout. The scale of such potential negative results is currently unforeseeable.

The major refrain of our reaction has been to “flatten the curve.” The basic idea is simple. The healthcare system can only attend to a very limited number of severe cases at any one time. So if we get an onslaught of cases in one sharp peak, then there will be no hope of using our limited resources to save what lives we can. This seems simple enough, but I think it leaves out some important considerations. First, the graphic that is normally represented is completely out of scale. The potential peak is not just twice as high as the dotted line, but many times as high (we do not know exactly how high yet). To spread out the curve to below the dotted line, we will require not just eight weeks of intervention, but many months. Can we impose a lockdown for half a year or more?

Another consideration is one brought up by Ioannidis. If our health care systems are, indeed, overwhelmed regardless of our interventions, then our interventions may only succeed in extending the time that our healthcare systems are overwhelmed. To speak in terms of the curves, “flattening the curve” is only a good idea if we can get it below the dotted line. If we cannot get the curve under the dotted line—as is already the case in Italy and in some parts of Spain—then we will only protract the period during which hospitals are over capacity. This means that there will be more time that victims of trauma, heart attacks, strokes, and so on will be unable to get treatment, which can result in more total lives lost. After all, we must deal with all of our usual health problems on top of this. People will still need to give birth.

Here is a piece of pure speculation—from somebody who is in no way an expert. In the 1918 flu pandemic, one reason the disease became so deadly is because of a natural selection process. Normally, mild strains of viruses are preferentially spread, since those with milder symptoms are out and about, spreading the virus, and those who are infectious stay put. But if a lockdown creates a similar situation as the First World War—wherein mild cases stay put (since we are in lockdown and they were in the trenches), while severe cases are the ones which spread via hospitals—then we may be preferentially selecting for more severe forms of the virus. Admittedly, I have not heard anyone respectable express this worry.

I have, however, heard experts express the worry that, by locking people in their homes, we are potentially shooting ourselves in the foot. This would be because it prevents the least vulnerable from developing immunity, which would go a long way in making the entire population less susceptible. This is called “herd immunity,” and it is the strategy urged by David L. Katz (another expert, the founding director of the Yale-Griffiths Prevention Center). His main point is that it is not sensible to shut down all of society if only certain members of society are seriously vulnerable, or in his words that a more “surgical” approach is needed. (Also, certain interventions, such as sending college kids to live with their older parents, do not seem sensible from any perspective.)

My worry is that governments are incentivized to badly underreact and then badly overcorrect. They underreact because each government wants calm, happy, and prosperous citizens, and disrupting life for a seemingly small threat is not politically advantageous. They will overreact because, in the face of a threat that can no longer be ignored, governments must be seen as maximally responsible. What is more, with heavy government intervention, any successful diminution in cases can be claimed as government success. Thus, governments are incentivized to take the most extreme measures available. Anything short of that will appear cavalier in retrospect if the disease is as bad as it may indeed be; and even if it is not as bad, any success will go to the credit of the government.

But everything in life is a tradeoff. Morally speaking, the government must weigh the damage inflicted by the virus against the damage inflicted on the suppression measures. And if both of these are totally unknown quantities—which seems to be the case—what then? We are past the point where we can say “better safe than sorry.” We risk being both sorry and unsafe. 

My own feeling at the moment is one of intense frustration. The governments of Europe and the United States have given conflicting messages to its citizens and have been content to react rather than prepare. Given the sharp shift from blasé indifference to emergency measures, I can only conclude that we are not in a situation like that of an expert chess player, but more like that of a stock broker—trying our best to predict the unpredictable. As Bill Bryson goes at lengths to show, we are still quite astonishingly ignorant about a great many things in our own bodies, including disease. And as Ioannidis explains, we are still quite in the dark even when it comes to something as humble as the seasonal flu or common cold. We only have rough estimates of flu related deaths, because we cannot filter out other common diseases such as colds; and it is also possible to have multiple infections at once.

In any case, to me it seems quite clear that the government’s wild pivot from indifference to emergency cannot constitute a rational response. To act as if the virus is unimportant one moment and the only important thing on earth the next is not evidence of clear thinking. We are in a hurry to embrace policies whose effectiveness, sustainability, and collateral damage are unknown to combat a virus whose danger is undetermined. While we are all collectively obsessing over the coronavirus (since lately it is impossible to think of much else), perhaps it is wise to remember another of Kahneman’s findings: “Nothing in life is as important as you think it is when you are thinking about it.”

One thought on “Quotes & Commentary #71: Kahneman

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s