04/01/2020
guest contributorJulian C. Jamison, PhD, is a Professor of Economics at the University of Exeter Business School and is affiliated with the Jameel Poverty Action Lab (MIT) and the Global Priorities Institute (Oxford).
In part 1 I emphasized the importance of clearly specifying counterfactuals when making comparative judgments; consequently defined dichotomous policies of lockdown vs moderate social distancing (MSD) in the context of the present COVID-19 pandemic; and summarized previous literature suggesting that in fact most of the epidemiological gains come from MSD and in particular changing the behaviors of the most socially active individuals, the most at-risk individuals, and symptomatic individuals. These changes should occur early and be sustained throughout the crisis, but the marginal benefits of applying them to everybody else are relatively low.
Individual behavior
At an individual level, what does the difference between MSD and lockdown look like? We can estimate the probabilistic mortality cost of an asymptomatic individual spending the next 24 hours in one scenario versus the other. The excess mortality from MSD will be roughly the probability of currently being infectious but not realizing it (call that A) times the additional marginal probability of infecting someone else under MSD (call that B) times the probability that that person dies and would not have been similarly infected by anyone else (call that C).
Using the most recent data (all details and calculations are available here), I estimate that A = 0.3%, B = 6%, and C = 0.2%. The last number may seem low, but recall that the relevant parameter here is the infection fatality rate rather than the case fatality rate. Overall this implies a 1-in-2.8 million probability of an additional death. Of course these estimates are uncertain, dynamic and variable across locations, and readers may want to conduct their own sensitivity analysis. However the qualitative conclusions below are robust to all but the most extreme variations.
How do we interpret this cost? One benchmark is that there are about 1.25 motor vehicle fatalities per 100 million miles driven in the US, so the probability above is equivalent to driving an additional 28 miles. To put it differently, if society generally considers it acceptable to drive 14 miles each way to a friend’s house for dinner of an evening (with the concomitant small risk of causing an additional death by doing so), then consistency would require that we consider it acceptable to engage in ‘only’ MSD rather than a lockdown. We still need to take precautions, but for example going (on foot!) to a café or store in the middle of the outbreak involves the same order of magnitude of risk as driving to that friend’s house for dinner in normal times. Whether we acknowledge it or not, these are the types of trade-offs that we are (implicitly) making every day.
We can also interpret this number in monetary terms. Recent estimates suggest a value per statistical life (VSL), which is the appropriate metric when evaluating small changes in mortality risk, of about $10 million in the US. Hence an unadorned benefit-cost analysis implies that if the social benefits (both financial and psychological, summed across everyone involved) are $4 higher per person per day under MSD than under lockdown, then MSD is the optimal strategy.
A third approach is to compare the lockdown to the best interventions already in existence. The most cost-effective global charities (e.g. see here or here) can save the life of a child for about $1800 in expected terms. So a $1 donation to such a charity has more than 1500 times the impact of sheltering in place for one day in terms of lives saved.
Other considerations
The analysis above is from the perspective of an individual representative decision maker trying to maximize social welfare (rather than their own utility), so in that sense it is as applicable to aggregate policies as to personal behavior. That being said, the basic parameters do not fully take into account externalities due to health system constraints, which may temporarily increase the value of C (although recent estimates from Wuhan suggest not by as much as had been thought), nor the road safety and pollution benefits from depressing the economy.
On the other hand they also do not take into account the negative externalities due to a sustained lockdown. For instance, unemployment carries not only a financial and emotional toll, but also a health toll. It kills people as surely as COVID-19 does, especially in the long run, and it harms the next generation as well. Furthermore, physical and social isolation also kills people, even after controlling for other factors such as loneliness. The magnitudes of all these effects are quite large, although it is impossible to perfectly predict how they will transpire in the present context. It would be a mistake to think that a lockdown saves lives at the expense of money: it saves a very modest number of lives according to the calculations above) while costing money, unhappiness, probably more intimate partner violence, and importantly other lives. The goal is not simply to ‘flatten the curve’; rather the goal is to maximize total welfare in the face of this crisis.
There are also difficult equity issues involved. The costs of a lockdown are likely to be disproportionately felt by the poor and less virtually connected, while the costs of the disease are felt by the elderly and co-morbid. This recent bioethics article provides reasonable and balanced guidelines regarding scarce medical resources: prioritize frontline health care workers; prioritize the elderly for vaccines but the young for life-saving interventions; use random lotteries if necessary. The link to economics is that we need to acknowledge and be transparent about trade-offs; there is no single correct moral calculus, and it’s not as easy as saying “every life is equal”.
Unfortunately even this guidance doesn’t tell us what to do in all situations, nor does it address the inequity around who carries most of the burden of interventions. Furthermore we have yet to see how COVID-19 will impact developing countries, with their generally younger populations but crowded conditions and subpar health systems.
Conclusion and predictions
The implications of this analysis is not that we should be complacent or do less, but that we should be smarter in our response. We need to focus our limited resources, attention, and suasion where they are most effective, both to maximize the immediate impact on the spread of the disease and also to minimize the health as well as financial costs of the interventions themselves. We should provide more support for at-risk groups (e.g. giving them priority access to food, both at stores and for online delivery). We should invest heavily in testing and in protective equipment. We should practice sustained yet moderate social distancing — but not a lockdown or shelter in place.
In Italy a small lockdown (affecting about 50,000 people) began between Feb 21-27, which made national and international headlines. It is not unreasonable to suspect that individuals across the country voluntarily initiated elements of MSD around this time. The national lockdown came on March 9-11, two weeks later. Given that the typical delay between initial infection and death (for those who die) is 15-22 days, and that the number of daily deaths in Italy stabilized around March 21, only 11 days after the lockdown, this suggests that MSD was sufficient to halt the exponential spread of the virus.
For the UK, modest measures began around March 14-15 and various stages of lockdown (e.g. school and business closures) came into effect March 20-23. Thus my prediction is that daily deaths will stabilize starting sometime in the first week of April, before the lockdown would have had time to have an impact on deaths. Similar analyses can be done across US cities and states, as well as other countries, each with their own timing and approach.
