THE NUMBERS ARE staggering. On May 7, 2,231 Americans died of the disease Covid-19, bringing the total number of deaths in the United States to 75,662, and more than 270,000 worldwide.
The economic effects have been nothing short of American carnage. At the end of April the US Congressional Budget Office hinted that the second quarter of 2020 would see the first drop in the US Gross Domestic Product in six years, and the worst since 2008. Since March, 33.5 million people have filed for unemployment. Companies large and small are going to disappear, along with millions of jobs. Consumer spending, business investment, manufacturing—everything is in freefall, and it’s not likely to get better until 2021, even if the pandemic eases and doesn’t snap back with a second wave. (Pandemics tend to snap back with second waves—especially when social distancing ends too soon.)
Put it that way, and the choice seems stark: Continue strict social distancing and shelter-in-place measures to minimize the spread of Covid-19 and save thousands of lives, or end the lightweight lockdown—open all the shops, restart the factories—and save the economy. Sacrifices must be made for the common good. “We can’t keep our country closed. We have to open our country,” President Trump said while visiting a mask factory in Arizona Tuesday. “Will some people be badly affected? Yes.”
But…really? The point of social distancing was to “flatten the curve,” to slow the spread of the virus so that hospitals wouldn’t be overwhelmed and governments could take public health measures—like widespread testing and tracing the contacts of sick people—to keep people safe. All of those things would have rendered the dichotomy false; the lockdown wouldn’t have to be total and the economic costs could be lessened. None of that happened.
Sacrifices have to be worth it. The good has to be greater. And there’s devilry in those details. New York Governor Andrew Cuomo made the point in stark terms: “How much is a human life worth? That is the real discussion that no one is admitting, openly or freely—that we should,” Cuomo said in a briefing Tuesday. “To me, I say the cost of a human life, a human life is priceless. Period.”
As the Associated Press has reported, the federal government has largely abandoned its own standards for when states should lift their shelter-in-place orders. A researcher at the respected Johns Hopkins Center for Health Security told Congress last week that no state looked epidemiologically ready to go back to normal.
And yet 31 states have decided to just go for it. Texas is letting restaurants and movie theaters reopen at 25 percent capacity, with barber shops to follow—while the governor acknowledges privately that Covid-19 cases will certainly increase as a result. Georgia is lifting its stay-at-home order and allowing places from tattoo parlors to bowling alleys to unlock their doors. Even California, which battened down early, is opening some southern beaches.
Information about the virus is incomplete and sometimes contradictory. So is information about its impact on the national economy. So is information about what people will contribute to the economy even if states end official restrictions. Given that uncertainty, who is going to get on an airplane next week? Or go to a crowded bar? (A minority, according to polls, but the perception of risk has declined in recent weeks, independent of the spread of disease.)
How much is a human life worth? As a society we have historically been willing to incur costs to save lives and improve public welfare. Government forces carmakers to reduce air pollution to help people with asthma, and the price of cars goes up. Laws prevent factories from polluting to save fisheries, and goods cost more. But that kind of tradeoff clearly has limits. Few people suggest deactivating the country’s financial engines to fight opioid addiction deaths or flu or heart disease or traffic accidents. Why do it for this one very bad respiratory virus?
Answer: This virus is not like those other things. In less than five months it has killed more Americans than the wars in Vietnam, Iraq, and Afghanistan—combined. If trend lines continue, it’ll kill more people every day than died on 9/11. More than that, lots of the preventable public health crises that kill Americans also make a lot of money for someone, like opioid-making pharmaceutical companies, or the petrochemical industry. People have incentives to make it hard to fix those problems. But Covid-19 doesn’t have fans.
And so, to save a vast number of lives, we will pay a huge cost. Until that cost seems too high.
This calculation is fundamental to the way Americans make policy decisions in normal times. We have a set of tools to draw on—a winding, fascinating body of knowledge that has, since World War II, helped leaders make decisions like this. In asking whether social distancing, school closures, event cancelations and other “non-pharmaceutical interventions” are in any sense “worth it,” the implicit question is simple and profound: What is a human life worth, in dollars?
THE SCIENCE OF human value began with the preparations for a previous apocalypse. Specifically, military strategists wanted to know how to inflict the most effective nuclear holocaust for the least amount of money.
To be fair, The US Air Force didn’t want to know the value of preserving a life, but of ending it. In essence, this was a ghoulish corollary of determining the value of a life: How much does a death cost? Strategists wanted to know how they could wreak the most damage in a first-strike nuclear attack on the Soviet Union—given their limited budget and a limited number of airplanes to drop the bombs. So in 1949 the Air Force tasked the RAND Corporation with the problem. Freshly independent from its origins as an Air Force-funded aerospace think tank, RAND set about applying a new set of tools to the problem: game theory and binary computers. Dr. Strangelove will see you now.
After crunching hundreds of equations, optimizing 400,000 different combinations of bombs and planes—modeling as well personnel, airbases, procurement, and logistics—the RANDies were ready to show the Air Force how to stop worrying and love mathematical models. The winning strategy, unveiled in 1950, was to field as many cheap planes as possible, to turn the Soviet sky black with antique propeller planes playing hide-the-ball with A-bombs so the Soviets wouldn’t know whom to shoot down. As the Georgia State economist Spencer Banzhaf writes, the Air Force brass weren’t into it. RAND’s game-theoretic approach might’ve beaten the USSR, but it also maximized the number of US pilots killed and minimized the Air Force’s rationale for buying new jet airplanes.
RAND sort of apologized and re-presented their analysis in a way that allowed the Air Force to buy all the deadly new toys it wanted. But the analysts realized they had what they called a “criterion problem.” A bomb or a parachute or a course of training had a dollar value—but what about the person who benefited from all three? They knew how much an airplane was worth, but not its crew. It was really screwing up their theory of the game.
The RANDies weren’t the only ones grappling with the moral and economic problem of the value of a human life. By midcentury, economists and lawyers were trying to rationalize and put statistical frames around this basic problem of the human condition: managing risk and figuring out what outcomes are worth a potential death. Courts of law were doing it to compensate people for wrongful deaths, for example.
The relatives of someone who got killed on the job, let’s say, might receive as compensation the amount of money that person would likely have earned over a lifetime. Of course, that’s not fair at all—why should the family of a coal miner killed in a cave-in be entitled to less compensation than the family of a guy who works in the mine office? By any morally valid reasoning, the size of a paycheck doesn’t make one life worth less than another.
“In some of the early work, it was pointed out that we don’t put a dollar value on an individual life. The example was, if a girl falls down a well, we don’t say, ‘sorry, it’s going to cost $10 million to go down there and get you, and you’re not worth $10 million, so good luck,’” Banzhaf tells me. “We just don’t do that.” As Banzhaf says, economists of the time were trying to distinguish, in terms of benefits and costs, between private consumption choices made by individuals and population-spanning policy choices made by, like, governments.
A former USAF pilot turned PhD candidate named Jack Carlson found the beginnings of a way out. In his dissertation, he tried to put a cost not on a life, but on saving lives—or not saving them. The USAF, Carlson wrote, trained pilots in when to eject from a damaged plane versus trying to land it. Ejecting would save the pilot, and landing might save the (expensive) plane.
Carlson ran the numbers on bail-out versus landing and found that the tipping point implicitly valued the saving of the pilot’s life at $270,000. In another case, Carlson noted that designing, building, and maintaining ejection pods for the crew of the B-58 bomber would cost $80 million and save between one and three lives a year. Making the implicit explicit: Doing the math, the US Air Force pegged the “money valuation of pilots’ lives” at between $1.17 million and $9 million.
Carlson’s thesis adviser, a former RAND economist named Thomas Schelling, built his student’s ideas into the framework that’s still in use today. In 2005 Schelling would win the Nobel Prize for his work on the game theory of conflict, especially nuclear war, but back in 1968, when he was a professor at Harvard, he wrote a chapter in the scintillatingly titled book Problems in Public Expenditure Analysis called “The Life You Save May Be Your Own.”
It’s a weirdly philosophical work, somehow both whimsical and elegiac. “This is a treacherous topic, and I must choose a nondescriptive title to avoid initial misunderstanding,” Schelling begins. “It is not the worth of human life that I shall discuss, but of ‘life-saving,’ of preventing death.” Schelling was trying to get out from under the moral weight of putting a monetary value on life, and after 35 pages of squirming he identifies the lever that’ll shift the mass. You can’t value a life, he says, but you can find out how much money people are willing to accept to risk their own.
Take a program to save lives in a large, well-known population with a risk that’s well-understood but small, and then ask, OK, what’s that worth? You can figure that out through surveys or consumer behavior— “revealed preference,” as economists call it. Take what people will spend individually to avoid a teensy risk, and multiply it by the odds of that risk coming to pass and the total number of people it might affect. That’s it.
Schelling called it the Value of a Statistical Life.
This approach has the advantage of dodging the morally questionable admission that death is part of the cost of doing business. Like insurance, Schelling’s idea spreads out a known risk among a large population, smearing out the question of specific responsibility or guilt so everyone has a share.
A decade later, amid the doldrums of the 1970s, politicians were starting to worry about the financial implications of government regulations. Sure, it was fine to save bald eagles or keep rivers from catching fire, but was it worth making taxpayers or businesses (and therefore consumers) pay their hard-earned dollars for that? President Jimmy Carter ordered agencies in the executive branch to take a new approach, analyzing the costs and benefits of every new rule. When Ronald Reagan took office, his deregulatory mania went further. All executive agencies had to prove, to the Office of Management and Budget, that the economic benefits of any major regulation outweighed the costs of implementing it.
In 1981, an economist named Kip Viscusi suggested using VSL to make these decisions. As he later wrote, the math was pretty simple. The odds said that about 1 in 10,000 Americans died on the job every year—a risk of 1/10,000. And in return, people got paid an extra $300 a year for incurring that risk. So OK: 10,000 workers get $3 million in total to risk one of them dying. The VSL was $3 million, or about $8.9 million adjusted for inflation. Today, estimates for VSL hover between $9 million and $11 million.
“We spend some money to smooth out a curve on a highway and predict it’ll lessen the chance of dying of each person who goes around that curve,” Banzhaf says. “If there are million people driving that curve, and each one has a reduced risk of dying on that curve of one in a million, then by fixing the curve, we saved one life.” If you believe in the VSL, it’s worth spending $10 million to regrade the road.
It was a controversial approach, for some of the same reasons that social distancing is controversial today. Not everyone agreed that risk—or risk-aversion—was the right way to evaluate policies. Maybe outcomes like cleaner rivers and non-dead birds were their own valid metrics, their own reward. Katherine Hood, a sociology doctoral student at UC Berkeley who has written on the history of VSL, points out that the CEO of General Electric gave a speech in 1978 called “The Vain Search for a Risk-Free Society;” industrialists at the time worried (or said they were worried) about risk aversion threatening the American way of life, a position that tech industrialists like Elon Musk still stake out today.
Meanwhile the left side of the political spectrum worried about the same thing but from the opposite direction. In congressional hearings, familiar politicians like Al Gore and Ralph Nader testified that health and safety regulations simply weren’t amenable to cost-benefit analyses, because while the costs were fixed, the benefits were unpredictable. “Requiring factories to not pollute, a lot of the time that regulation ends up spurring innovation and leading to a healthier and more productive workforce,” Hood says. “There’s a real political battle going on here. It’s not just an argument about how to do the math.”
All of which leads to the basic math to calculate whether it’s worth keeping people home and businesses closed to fight the spread of Covid-19 despite the economic consequences—to answer the question all those politicians have been asking on TV. All you need to know is how the GDP will change, and how many lives you save.
So, the math, in broad strokes: First, assume that Gross Domestic Product would decrease by 2 percent this year without social distancing, but that instead social distancing will shrink GDP by 6.2 percent. That’s the cost.
Then also assume that all the mitigation measures reduce the Covid-19 fatality rate from 1.5 percent when hospitals are overwhelmed to just 0.5 percent. That saves 1.24 million lives, with a VSL of $10 million each.
Analysis: Social distancing to fight the spread of Covid-19 saves $5.2 trillion.
That seems good.
I feel OK about that calculation because I asked Kip Viscusi, now an economist at Vanderbilt University. He graciously agreed to metaphorically scribble on the back of an envelope. “Ask an infectious disease expert how many lives will be saved, and the numbers they were coming up with will be at least a million lives. Once you’ve got that number, you can run with it. A million lives at $10 million each is about $10 trillion, which is half the GDP,” Viscusi says. “Unless you have a really catastrophic outcome, the health benefits of social distancing swamp the costs.”
Stop there, and the problem does indeed seem simple. But of course it’s not.
Epidemiologists are reasonably secure in the idea that social distancing instituted sooner rather than later lowers overall deaths. And history bears out that it’s worth it. One analysis—again, an un-peer-reviewed preprint—says that the economies in cities that instituted social distancing measures stricter and earlier in response to the 1918 influenza pandemic bounced back faster and higher. A city that put those non-pharmaceutical interventions into effect 10 days earlier saw manufacturing employment go up 5 percent higher than a city that did it later. Keeping those measures in place for 50 days longer increased that employment by 6.5 percent.
But that said, it’s not obvious whether policymakers and public health experts are thinking in terms of VSL or any other analysis deeper than who will vote, and how. “VSL calculations are rampant among economists and outside analysts who are thinking about this, but I don’t know if anybody in the government is doing these kinds of calculations,” Viscusi says. “They’ll say, ‘the economy has to reopen,’ which is the message targeted to people who favor reopening, and then they’ll say, ‘we have to do it safely,’ which is targeted to people worried about the risk. They’re trying to appeal to both sides.”
Even if they were using VSL, that might be the wrong move. It’s too blunt an instrument. The question of who, exactly, incurs these costs and who, exactly, accrues these benefits acquires all sorts of subtleties. The arithmetic isn’t the problem; it’s the rhetoric.
REMEMBER THE CRITERIA for VSL—small, predictable risk spread out over a population that can say how much it’ll spend to mitigate that risk. “Most of the Value of Statistical Life calculations you have are for one life, or a small number,” says Andrew Atkeson, an economist at UCLA working on VSL and the pandemic. But they’re harder to apply, he says, when the risk is high and the exposed population is huge—potentially everyone, in fact.
And the cost side isn’t some thin slice of a paycheck, or a tiny bit of extra annual salary. “It’s not just, ‘oh, I’ll have to postpone buying a new car for a year,’ or ‘I can’t get a fancy meal out on my anniversary,’” Banzhaf says. “We’re talking about whole ways of life and livelihoods potentially being ruined and not coming back.”
VSL might be one thing to take into account in making globe-spanning, high-stakes decisions, but it can’t be the only thing. “After 9/11, all that response, was it about saving lives, period? Or was it about not letting the terrorists get us, a sort of pride? If it was only about lives, clearly we could have saved more lives by spending that money in other ways,” Banzhaf says. “I have been such a lifelong advocate of benefit-cost and quantitative analysis, but I just don’t know what number you would use right now.” With so much still unknown about Covid-19, nobody even really knows the overall mortality risk, much less the chances that death will happen to any one person.
Also, VSL is different for different demographic groups, though it’s slightly suicidal, career-wise, to admit it. A massive debate over whether to value older people with a smaller number—figuring that they might not pay as much because they had less time left to live, lower the value of their statistical lives overall—got turned into a scandal over the government calculating a “senior death discount.” Richer people are willing to assume less risk than poorer people. Some economists even think that globally, poorer people in the developing world may well value their risk lower because they simply have less to spend, and more to lose. Even if it’s true, acknowledging it puts you on a slip-n-slide to racism and eugenics.
People in the US might be willing to assume more risk for less money during the pandemic because the emergency social safety net doesn’t pay 75 to 90 percent of their income when they stay home, as it does in, say, Denmark. The willingness to assume risk changes with context, and every one of those contexts implies a different cost-benefit analysis.
All that assumes people understand their actual risk—which they can’t, because scientists only just met SARS-CoV-2, the virus that causes Covid-19, less than five months ago. Neither economic nor epidemiological models have enough data to account for known unknowns like how likely it is that someone might get sick after walking behind an asymptomatic jogger who isn’t wearing a mask.
If the risk that VSL tries to account for is unknown, that’s called “Knightian uncertainty,” and it makes it hard to understand how people value that risk and how they’ll act in response. “How do people behave when they don’t know the right model, and they don’t know the right parameters even if they do?” says Martin Eichenbaum, an economist at Northwestern University. “Does that bias them to inaction? Does that bias them to pessimism?”
No one knows.
Just as it’s tough to measure the benefits, it’s also hard to accurately measure the costs. Much of the early work in determining the economic effects of social distancing and business closures uses Gross Domestic Product as a metric, and it’s a bad one. “GDP is a lousy measure of economic welfare,” says Alan Krupnick, an economist at Resources for the Future, a nonprofit think tank in Washington DC. “Economists tend to look at aggregate economic indicators like unemployment rates and GDP, as opposed to getting into the distributional issues—who’s being affected, who’s losing income, where is this GDP growth actually coming from, does it increase the equity in society? Our profession is not as good at doing that.”
GDP might go up if people felt they had no choice but to return to work regardless of the risk of infection. If essential workers are also most likely to be exposed, and they return to work, the economy could improve as social inequality increased. A person who has no income if they don’t go to work is running a very different cost-benefit analysis—the risk of getting sick and perhaps dying versus the “benefit” of being able to afford food and not getting evicted. They incur all the risk to merely not starve, while the more nebulous and conceptual “economy” benefits greatly (and presumably so do private-equity hedge funds and billionaires).
The cost-benefit analysis approach to Covid-19 shutdowns clearly needs some honing. A hodgepodge of closure and reopening policies among populations with wildly different risks of infection and death does not lend itself to balancing a cost in dollars against a cost in blood. What researchers would like to know is which specific interventions are most successful stopping the virus and have the least impact on people’s economic lives. Figuring that out could lead to a new phase in the fight.
THE APPROACH THAT epidemiologists use to map how diseases spread was developed in the 1920s and 1930s, primarily by WO Kermack and AG McKendrick. They divided a given population into three kinds of people, now called “compartments:” Susceptible, Infected, and Recovered (or Removed, which is dead). That’s the basis for an SIR model, but you can add in more categories. (SEIR adds people Exposed but not yet Infectious; SEIRS is for when Recovereds don’t remain immune and circle back around to Susceptible status.)
Those populations grow and shrink according to variables like the rate of infection—how many Susceptibles a given Infected can infect (that’s called the reproductive number), and over how much time. Modelers also hope to know how long it takes for an Infected to start showing symptoms, or what proportion of Infecteds get Removed and how long that takes.
To a certain extent, social distancing measures get wrapped into the Reproductive Number. The strictest kind of quarantine reduces it effectively to zero, because Infecteds can no longer come into contact with Susceptibles. But in even the most sophisticated models, that’s a gross oversimplification because of those same demographic and geographic differences that plague (sorry) the VSL.
The problem gets even worse, though, and the explanation is a clue to why epidemiological models have been so controversial and so all-over-the-place in predicting what’ll happen with Covid-19. They tend to overestimate the number of dead or sick people, because they don’t account for behavioral changes like social distancing or new patterns of consumption like wearing masks, or only getting takeout.
Adding new compartments can help, with different populations showing different levels of adherence to lockdown policies, but you still have to be able to “parameterize” those models—“You need to be able to estimate what the impact of those would be, such as how much would that reduce transmission and then how would that reduction change with differing adherence. And to know those estimates for sure is hard,” says Helen Jenkins, a biostatistician at Boston University. “We are very early on in this pandemic, so we don’t have good estimates. You’re basically using poor data in your model, so it’s questionable how useful that is.”
From a public health and political standpoint, one of the worst things that can happen to a model is that it works. If a model inspires a government to institute social distancing, it becomes a reverse Toynbee Convector, precluding the future it predicts through the act of predicting it. That’s the source of the public phenomenon known as the paradox of prevention—if it works, people assume the thing it was trying to fix must not have been that bad.
“All these SIR models always overestimate the eventual cumulative burden of disease, and usually it’s because they have fixed parameters. They don’t take into account that people are going to change behavior, rationally or not, and disease will slow down more than would be predicted by the model,” Atkeson says. The opposite could also happen—models that build social distancing into the numbers, with an artificially depressed reproductive number, end up lowballing the impacts when social distancing gets lifted before the disease is suppressed.
It’s probably an oversimplification to say epidemiological models can’t take into account change. One subclass, called dynamic transmission models, can reduce contact rates over time, for example, by incorporating mobility data like what you might get from a cell phone. “Though just because it’s possible to include does not mean that models have indeed taken this into account yet,” says Brooke Nichols, a health economist and infectious disease modeler at BU.
A more subtle and useful approach might be to unify the two philosophies here. Nichols says the fields are siloed off from each other, even though an interdisciplinary approach would help with not only Covid-19 but figuring out the true value of any public health intervention that averts deaths.
An economist like Eichenbaum would say that epidemiologists are good at looking at the things people do and coming up with infection rates, but not as good as economists at talking about how infection rates might change behaviors like going to arena concerts and buying at retail. “That’s just not what they do. That’s our job,” Eichenbaum says. (And indeed, he’s co-author of a working paper that came out this April called, simply, “The Macroeconomics of Epidemics.”) “Epidemiological models are basically nonlinear difference equations, and economists are used to that stuff. We know how to solve those. The challenge, mathematically, is to understand that the coefficients in those nonlinear difference equations depend on what people do, and what people do changes those coefficients.”
Economists and epidemiologists might still have some work to do in the quest to integrate the two worlds. “I would venture that the epi models can be slow to adjust, whereas the econometric models are too flexible,” says Jeffrey Shaman, an infectious disease modeler and director of the Climate and Health Program at Columbia University Mailman School of Public Health.
Modelers from any tradition might agree, though, that their work is most helpful in conjunction with experimental data—something sorely missing in the dynamics of Covid-19. The geographically heterogenous lifting of social distancing requirements across the US will put an ugly, tragic end to that lack of data. “There’s all these uncertainties about how people behave and how the disease will react,” says Atkeson (who, to be clear, is not advocating this move). “Since we’ve never done this before, or not in 100 years, it has to be empirical. You impose the measures and see what happens.” Some epidemiologic curves will flatten, others will flex, and more people will die.
That’s…a choice. It’s not one the vast majority of Americans want, and it seems supported mostly by anti-vaxxers and the kind of people who bring guns and tactical vests to nominally nonviolent protests. But it’s one President Trump has been pushing for, even when states haven’t met the most basic conditions of his own policy for “re-opening” the economy. (States were supposed to first report 14 days of falling new cases, not to mention an infrastructure for testing and contact tracing; no state meets both criteria.)
That’s going to be terrible if you don’t want people to die needlessly. But it might open the door for a different, clearer kind of decision making—one that doesn’t depend on necessarily opaque mathematical models and instead drags economics, a dismal science even in the Before Time, into the now. It might provide useful knowledge, maybe for the next pandemic, but it’ll also push the most vulnerable people—the sick, the old, the poor, people of color—toward sickness and death, no matter what their individual appetite for risk is.
The truth is, the question of how to respond to Covid-19 has never really been one of lives versus dollars. At least, it didn’t have to be. The dichotomy was false because of the degree of control a government could always exert on both sides of risk—the risk of infection, flattened by social distance, and the risk of personal financial ruin and societal economic collapse mitigated by aid programs. The federal government is pushing to end the restrictions that flattened the curve, and the aid programs have been grievously inadequate.
And now here we are, forcing (or at least urging) scared people to go back into the world because nobody could be bothered to develop a nationwide program to test people for infection and trace their contacts if they were positive, or to adequately support a pause of economic activity. Consumption behavior has a context. “It’s not, either we can choose to go about life as normal and some people will die, and that’s life, people die, or we can all shut down and give up our productive American way of life,” Hood says. “That’s only a choice people are making because we don’t have a social safety net.”
Absent that kind of response, the invisible hand of the market seems to be giving people the finger. Instead of trading between lives saved and economic stability, we will have neither. We’ll attempt to restart the economy, more people will die, and the economy will auger in. The number of deaths in the United States from Covid-19 is staying at a steady, high rate, with many projections indicating growth to come. Every economic indicator says the losses are continuing. The decisions of leadership reveal a preference: The lives of Americans must now, somehow, be worth less.