• Economic analysis of COVID-19 responses (part 1)

    Julian 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).

    When my wife asks me how she looks, I have learned that “Compared to what?” is not the correct answer. However as an economist, I must admit that that is often the first thought that goes through my head, since otherwise I honestly don’t know how to respond. As Jeremy Bentham wrote: “[G]ood itself is bad in comparison of better.” This discussion will therefore be an economic analysis not in the sense of finances but rather of how an economist approaches the world.

    Is the UK (where I live) doing well in fighting COVID-19? Well, compared to what? We could compare to how China did, or how the US is doing, or to what the optimal response would have been, or to a realistic response under a different government, or to how well the UK did during the Blitz in World War II, or to any of a number of other alternatives. None of these is right or wrong — they are simply different questions with different answers that tell us different things about how to understand the world. The important element is to ask the right question for your purpose, and to be clear — including in your own mind — which question you are asking.

    Setting the stage

    In particular, it is not sufficient to ask whether the current lockdown (aka shelter-in-place) in the UK and many other countries was a good idea, or how long it should be extended. Compared to what? For concreteness, and because I believe it is the most policy-relevant, in what follows I shall primarily concentrate on the comparison between a lockdown and moderate social distancing (MSD). Roughly speaking, by the latter I am referring to:

    • remote teaching at universities and potentially secondary schools
    • working from home if possible but keeping most businesses open, with separate provision for at-risk groups (elderly, those with underlying conditions, frontline health workers) when feasible
    • testing as much as possible (leveraged with data-driven predictions)
    • minimizing large indoor gatherings as well as travel in confined spaces
    • vigilant personal hygiene: hand-washing, covering coughs and sneezes, wearing a face mask if available (even if home-made) mostly to protect others rather than oneself
    • self-quarantine if symptomatic or if recent close contact with an infected individual

    This is roughly consonant with the current approach of Sweden and Holland, and with what was temporarily promulgated in the UK the week of March 15. In distinction, by a lockdown I mean:

    • schools, playgrounds, and most businesses closed
    • no travel other than for necessary supplies such as food and medicine
    • in particular no face-to-face socializing outside one’s immediate household

    The distinction between the two is not only about voluntary actions vs government regulation. Although lockdowns are more likely to involve formal enforcement, MSD will also include some involvement by the state, and even a lockdown relies to some extent on voluntary compliance by the public (as we are seeing now in the news). Rather the distinction is that MSD is the minimum that public health experts are suggesting or that is likely to occur naturally in practice to a large degree, while lockdown is in some sense the maximum that is feasible given the level of severity of this disease. Of course, while the comparison presented here is between two options for simplicity, it is clear that our policy space is not one-dimensional and it is a mistake to frame it as “How extreme should we be?” rather than “Which specific actions make the most sense at which points in time?”

    Another hallmark of economists is that we believe in quantifying decisions, not because we are always confident in the resultant numbers but because it forces one’s assumptions to be transparent and it usually yields at least the correct ordering between alternatives and also the right order of magnitude for the estimates. That will be precisely the goal in the second part of this discussion.

    Background knowledge

    I believe it makes sense to start with an informally Bayesian approach, asking what is a reasonable prior perception, not for whether MSD or a lockdown will work (which is not the relevant question), but for whether a lockdown is likely to be substantially more effective. Unfortunately most previous modeling and empirical work has focused simply on whether social distancing (in various degrees) makes a difference, rather than on the relative effects within the realistic range of options. However there is some evidence we can bring to bear:

    • Markel et al. (2007) study differential non-pharmaceutical interventions (NPIs) across 43 American cities during the 1918-19 influenza pandemic, finding that in general more interventions work better than fewer but that timing (early and sustained) has a larger impact than the number of interventions. For instance New York City rigidly enforced isolation and quarantine from very early on but did almost nothing else and had positive outcomes.
    • Nigmatulina & Larson (2009) model influenza mitigation strategies and conclude (amongst other results) that “…changing the behavior of the highly active individuals has a similar impact as decreasing the behavior of the entire community.”
    • Jackson et al. (2020) empirically analyze a natural experiment in severe social distancing due to extreme snowfall in Seattle that shut most of the city down for a week, finding that the transmission of respiratory viruses falls anywhere from 3% (for strains whose predicted epidemic peak was significantly later) up to 9% (if the shutdown came just before the peak).
    • Ferguson et al. (2020) model the impact of NPIs on COVID-19 in particular. Assuming R0 = 2.4, they show that social-distancing of over-70s (only) for four months reduces deaths by 49% relative to the baseline, whereas social distancing of the entire population but for just three months reduces deaths by 13-19% (depending on when it begins) relative to baseline.

    Overall the conclusion seems to be that optimal timing and targeting of interventions is more determinant than their breadth or severity. In particular, it suggests that it is worth asking whether a lockdown will actually be substantially more effective than MSD, even with respect to the stated goal of slowing the pandemic. Especially if it is only slightly more effective, how does it compare with the impact of other behaviors that trade off gains and losses in similar domains? I shall attempt to answer these questions in part 2.

     
    item.php