Yves here. This study tries to come up with some clever ways to identify counties where they can look at the effect of various government interventions and see what impact they have. The authors look at fatality growth, which I am not confident should be used without adjusting for the demographics of the counties being compared (yes, the authors are using growth, but some of the counties would have had low baselines, making early increases large in percentage terms).
I’m also bothered by the exclusion of the largest counties in each state. Here in Alabama, you can drive an hour from Birmingham, where there’s reasonably decent compliance with the state-wide masking order, and be in a very affluent lake-front community where people socialize and masking is not done. Any mandates regarding bars and restaurants have to be on the state level to have any teeth, since it is the state that typically controls liquor licensing and the threat to revoke a license is the way to bring these venues to heel. In other words, even though the authors say they focus on county-level policies, it’s not clear how you parse them out from state level.
So the implicit assumption that you have compliance on all fronts, including beach and park closures, which are hard to monitor and local police might choose to ignore, seems questionable. I also wonder about their view that school openings are not risky; this seems to be contradicted in quite a few communities between school and/or college reopenings and infection levels over the next few weeks in that community.
Nevertheless, this is an important line of inquiry and hopefully other scholars will build on this study.
By Matthew Spiegel, Professor of Finance, Yale School of Management and Heather E. Tookes, Professor of Finance, Yale School of Management. Originally published at VoxEU
As the COVID-19 pandemic continues worldwide, policymakers are still grappling with the question of which non-pharmaceutical policy interventions are effective. In the US, state and county policies varied widely, as did the growth in fatalities due to COVID-19. This column examines US business policies to help shed light on which policies save more lives. Stay-at-home orders, mandatory mask requirements, beach and park closures, restaurant closures, and high-risk (Level 2) business closures most consistently predict lower fatality growth four to six weeks ahead. Closures of low- and medium-risk businesses do not appear effective and, despite their costs, may even be counterproductive.
Policymakers face the unenviable task of trading off the benefits of restrictions that can slow the spread of COVID-19 against their economic costs. A growing body of research aims to aid the decision making by relating a variety of policies to COVID-19 cases and fatalities. For example, Dehning et al. (2020) report evidence that social distancing measures, school closures, and retail closures slowed the growth of cases in Germany. Hsaing et al. (2020) examine large-scale policy interventions such as travel bans, lockdowns, and gathering limits in an international setting. They report substantial declines in COVID-19 case growth when these policies are in effect.
There is also mounting evidence that masks are beneficial (see Howard et. al 2020 for a review). Abaluck et al. (2020) exploit differences in social norms and attitudes towards mask-wearing and estimate that the value of each additional mask worn by the public is between $3,000 and $6,000. Given the relatively low cost of wearing masks, this benefit is substantial. Other policies, such as widespread business closures and lockdowns, are more costly. It is therefore important to understand their relative effectiveness.
In Spiegel and Tookes (2020), we run what is essentially a ‘horserace’, in which we consider the relationship between future fatalities due to COVID-19 and a wide range of business restrictions. We then compare the magnitudes of the relationships, which can be interpreted as policy effectiveness.
Several studies rely on cross-country evidence (e.g. Askitas et al. 2020), where social norms, healthcare infrastructure, and demographics are likely to vary widely. Others focus on policies introduced at the US state level (e.g. Abouck and Heydari 2020, Friedson et al. 2020, Dave et al. 2020), missing potentially useful within-state variation that can help with the overall interpretation of any findings.
In the US, counties represent the finest level of detail for which one can observe fatalities. Both state and county governments have been active in implementing and reversing a range of policies including: general business closures; targeted closures of such venues as restaurants, bars, and gyms; gathering limitations; beach and park closures; mask mandates; nursing-home visitation policies; and restrictions on elective medical procedures.
With 3,141 counties making different choices about how to regulate, county-level data can be informative – even in the absence of randomised controlled trials, which would be ideal to identify the most helpful policies. Courtemanche et al. (2020) exploit county-level data to examine the relationship between a range of policies and case growth; however, differences in the availability of testing across both counties and time can complicate the overall interpretation. In Spiegel and Tookes (2020), we instead focus on county-level fatality growth.
Reactive Policies Can Complicate the Interpretation
Any study that tries to link policy interventions and outcomes must somehow distinguish between correlation and causation. Policies that are put in place near the natural peak of the outbreak will be followed by mechanical declines in death rates and can lead to findings that attribute the cause of declining fatality growth to policies (‘false positives’).
Atkeson et al. (2020) challenge the interpretation of existing influential work, pointing to rapid increases and then declines in fatality growth that might cause researchers to overstate the importance of policy. Policies that only partially mitigate death rates can make it difficult to detect their effectiveness (‘false negatives’). In our paper, we try to deal with the false positive and negative issue in two ways.
In a first test, we use the fact that many county regulations are imposed at the state level through governors’ executive orders. Following the intuition that smaller counties often inherit state-level regulations that are intended to reduce transmission and deaths in more populous regions, we remove the five most populous counties in each state from the sample.
In a second test, we match counties that lie near (but not on) state borders to counties in different states that are also near (but not on) state borders and are within 100 miles of that county. Unlike typical designs that exploit discontinuities by focusing on differences in policies and outcomes between counties that lie on state boundaries, we focus on the subset of counties that lie near, but not on, a state border. We examine near-border counties (as opposed to on-the-border counties) to reduce spillover effects.
These spillovers come in two forms. First, if a neighbour’s policy reduces disease transmission in its jurisdiction, it will also lower the transmission level across the border and thereby reduce fatalities in the county of interest. Second, a restrictive policy in one county (such as bar closures) and a less restrictive one in a neighbouring county may induce residents of the county with the tighter restriction to travel to and engage in otherwise prohibited activities in the less restrictive one. These spillovers can cause direct comparisons between the counties that share a border to generate false negatives and false positives.
We try to mitigate this problem by putting at least a one-county buffer between any near-border county in our sample and its neighbouring state. Still, the counties are sufficiently close that, absent policy differences, they should see similar trends in virus transmission. Once a nearby-county match is selected, the near-border neighbouring county’s policies are added as control variables in the predictive regressions. The underlying assumption is that differences in policies across state borders are due to differences in opinion (which introduces random error, along the lines of what one would obtain in a randomised trial).
In addition, all of our regressions control for the current and lagged levels of deaths per capita, lagged fatality growth rates, and several demographic and weather-related variables. Thus, our regressions predict differences in the future growth in fatalities in two counties that today have the same current level of deaths per capita, the same recent trajectory in deaths, and similar demographics and climate. They differ in that their governments have introduced different policy interventions.
Across specifications, we find that stay-at-home orders, mandatory mask requirements, beach and park closures, restaurant closures, and high-risk (Level 2) business closures are the policies that most consistently predict lower fatality growth four to six weeks ahead.
For example, baseline estimates imply that a county with a mandatory mask policy in place today will experience fatality growth rates four weeks and six weeks ahead that are each 1% lower (respectively) than a county without such an order. This relationship is significant, both statistically and in magnitude. It represents 12% of the sample mean of weekly fatality growth.
The baseline estimates for costlier measures – such as stay-at-home orders and closures of restaurants and high-risk businesses – are similar in magnitude to what we find for mandatory mask policies. We fail to find consistent evidence in support of the hypothesis that some of the other business restrictions (such as spa closures, school closures, and the closing of the low- to medium-risk businesses that are typically allowed in Phase I reopenings) predict reduced fatality growth at four-to-six-week horizons. Some policies, such as low- to medium-risk business closures may even be counterproductive.
We leave policymakers with the unenviable task of balancing public health concerns with the costs and benefits associated with the various restrictions that have been considered. Our primary goal is to provide evidence that can help inform the calculation. Mask mandates appear to be effective in every specification and are accompanied by relatively low economic and social costs. Other policies come at a higher cost. In those cases, we hope our results can help policymakers better assess the necessary trade-offs.
See original post for references