Evidence from 150 Countries Suggests the Virus Flattened Us
We did it. We flattened the curve. Or so the rhetoric goes in many places right now. As a result, many jurisdictions are easing their lockdown restrictions and looking to open up the economy again.
But is it really true? Did the myriad constraints on physical movement have the desired effect of limiting contagion? It is one thing to aim for social distancing and quite another to achieve it through school closures, mass gathering bans, workplace restrictions, and home quarantines.
These questions are especially pertinent right now because we are seeing three emerging trends in this pandemic.
First, active cases, hospitalizations, and death rates from Covid-19 are seeing slowing growth rates in some places, i.e., the curves are flattening. This is a sign of hope that we can start to see an end to the pandemic. So, we need to understand our prior policy actions to help guide us through the next stages of this crisis.
Second, many jurisdictions are starting to ease their social restrictions. This is happening across Europe (with Austria, Germany, and Czechia leading the way), the United States (primarily Midwest and Southern states with low infection rates), and a smattering of other places (Israel, New Zealand, and Iran).
The policy easing is slight and very gradual and reflects the unfortunate truth that progress is measured in smaller increases in infections and seldom in outright declines. A greater truth is that the government may ease the restrictions, but people will largely choose to continue to self-isolate, thus effectively extending the impact of current lockdown policies.
Third, and counter to the second trend, there are a number of places where the pandemic has taken on new energy. Singapore and Japan are seeing upswings in new hot spots (primarily centered on migrant workers and returning travelers), while Russia, Mexico and others are still early in their cycles and so are seeing deteriorating conditions. As a result, these places are tightening their social restrictions even further.
The underlying assumption in all of these policy actions is that we can bend nature to our will. Or, more precisely, that we can manipulate entire societies, isolating humans from each other for long enough that nature will essentially wander off and leave us alone. It is a grand experiment in hubris, akin to the decades-long discussion about controlling the planet’s climate by altering CO2 emissions. Humanity’s urge to control fate is now hitting a new all-time high.
This assumption about our powers is trivially true in the sense that creating physical barriers between people will actually reduce contagion — that’s basic science. It is more doubtful when one considers how little we know about the virus and how difficult it is to control human actions, especially without also creating terrible unintended consequences that result (as the most famous person on earth said) in the cure being worse than the problem.
How to Measure Policy
Fortunately, we have data to study this issue. Lots and lots of data. Unfortunately, the quality of the information often leaves much to be desired and so we should treat analytical conclusions with some caution.
Two new databases are starting points to investigate pandemic policy effectiveness. They are also great examples of how fast researchers can react to developing problems, as each harnessed distributed IT and hundreds of researchers to gather and record the real-time information.
The first one comes from the CoronaNet Research Group and the second is based at the University of Oxford.
We will rely more on the latter source because it has a Stringency Index that combines seven features of public health policy actions regarding pandemics: closing schools, closing workplaces, cancelling public events, closing public transport, initiating public information campaigns, restricting internal movement, and controlling international travel. The higher the number on a scale of 0 to 100, the more that the authorities are intervening to control social interactions.
We will compare this policy response over time to the underlying pandemic caseload and resulting deaths for 150 countries on a daily basis since the start of 2020.
Since the objective of the health authorities is to initially control and then eliminate the pathogen by reducing person-to-person contagion, it is not enough to simply track recorded cases and deaths, for the simple fact that they occur at different points in time.
The case numbers come after initial contact occurs, after the virus infects the new victim, after the person is asymptomatic, after the person shows symptoms, after the person is selected for testing, after the test is performed, and then after the results are reported as positive. This line of conditions shows why we do not have any clear idea of the extent of the disease across our societies, since testing, even when widespread, usually only covers 1 or 2 per cent of the overall population and is almost never a random sample.
Similarly, death numbers follow this caseload time-line into acute care facilities, a period of treatment and worsening symptoms, transfer to intensive care, and then, for the unfortunate few, termination of life. Deaths can also occur outside the hospital setting (for example, in nursing homes or at home) and, like cases, are subject to uncertainties owing to “cause of death” coding practices and other measurement biases.
What we really want are data that reflect the degree of contagion that policy is meant to inhibit, while what we get are two items, cases and deaths, that are summed up to the national level (the virus acts locally) and measured (poorly) well after the contagion. Better analysis awaits the detail hidden in the world’s current 2.5 million cases, but for now such information is unavailable.
Let’s Get to It
The next chart shows total cases and deaths across the 150 countries. They are presented on a logarithmic scale because there is such a huge span of results.
The takeaway from the chart is that there is a close relationship between cases and deaths, though the more cases there are, the fewer the number of related deaths (shown by the data cloud bending down to the right). This probably reflects the fact that countries in a worse and worsening pandemic situation are testing more and therefore finding more positive cases, even though the deaths will mostly be determined by the disease itself.
We have chosen deaths as our preferred proxy for the original contact contagion because it is probably a cleaner statistic, more related to the pandemic than to data practices, and also because the amount of case testing is a policy choice (whereas deaths can be affected by medical practices and population characteristics but are mostly a function of the disease).
The last important data issue is how to adjust for the fact that death occurs at the end of the pathogen cycle, while contagion occurs at the start. In the analysis that follows, we have shifted the death statistics back in time by four weeks to take account of this discrepancy. This 28 day gap is based on current estimates of the progression of the disease, including an incubation period and reporting delays. We will show below that our results do not depend on whether this lag period is one week shorter or longer than four weeks.
Italy and Croatia
So, we have a measure of policy (the Stringency Index) and a measure of contagion (deaths which occur on average four weeks later). Let’s have a look at what these two measures tell us about what authorities have done over the past few months.
The first example comes from Italy, one of the cautionary tales from this global pandemic.
As with all of the other country charts, death data (our proxy for contagion) in blue are scaled on the left-hand side and the policy index in red is scaled on the right-hand side. The data are daily since January 1st, total deaths are those added up since then, daily deaths are the new ones added each day, and both death measures are shifted back by four weeks.
What do the charts tell us?
First, Italy has had a lot of deaths in a very short period of time, a classic pandemic pattern. The latest numbers are over 24,000.
Second, policy has lagged the epidemic. The red policy line is almost always to the right of the blue contagion proxy line, showing that the epidemic continually swept through and then decisions were made with a delay. This is partly understandable since the true contagion numbers were not observed at the time. However, knowing a pandemic was underway means that policy makers should have anticipated the continuing stages of the outbreak.
Third, policy moved in fits and starts. For example, there was an initial policy move in late January and then three weeks went by with no further moves. Was there a plan or were changing circumstances determining policy?
Fourth, the biggest policy moves happened when the rate of deterioration in the pandemic was at its greatest. The largest jump in policy occurred right when the daily lagged death numbers peaked. This speaks to something like policy panicking, suggesting that the authorities were reacting to a situation that was perceived to be out-of-control.
Fifth, there has been a very modest response to the policy. The policy stringency eventually went to 95 and we are seeing a flattening of the curve, but progress has been very moderate and gradual. Perhaps it is simply too soon to see the full policy effects.
So, policy in Italy was not proactive at all but was rather delayed in response to the virus. Too little was done in the earliest stages and then the full weight of government edicts came down when the situation was at its most extreme, ironically at the precise point when the daily curve reached its peak. Too little too late and probably too much too late (when we consider the enormous economic costs of the lockdown).
We really cannot say whether the policy flattened the curve from the contagion peak or not. It may be that the virus had a contagion cycle that devoured its more vulnerable victims earlier and then slowed of its own accord, coincident with the policy timing. What seems clear, though, is that the set of policies did not have an overwhelming impact on the timing or magnitude of the pandemic in Italy.
Croatia is the next example, as it was and is one of the countries most vulnerable to the earlier Italian outbreak.
Even though Croatia’s policy generally changed after Italy’s, the moves there were more aggressive and compressed in time.
Also, unlike Italy, the red policy line in the chart always precedes the blue contagion proxy line. Croatia took a very strong proactive approach to the pandemic, probably after seeing the sobering Italian example next door. That approach is even more impressive when you consider that total deaths in Croatia are still below 50. And in fact, Croatia is the most proactive country in our set of 150 nations, as measured by the time between seeing over 25 contagion proxy deaths and the subsequent policy response of raising the Stringency Index over 50.
Overall, Croatia is an example of a country learning from and responding to the experience of its nearest (and potentially most dangerous) neighbor.
Other Country Pairs
Maybe unsurprisingly, given the Italian and Croatian example, we can see other pairs of countries where the first one was hit harder and earlier by the pandemic and then the second one (always a close neighbor) responded more forcefully in policy terms to avoid the same dire outcome.
The second country pair is shown below: Spain and Portugal.
The Spanish situation looks a lot like Italy, with a galloping pace of contagion, a late and spotty policy response, and a jump in the harshness of policy right after the peak in the daily deaths contagion proxy. There is even less evidence here than in Italy that the policy measures have had a significant impact on flattening the curve. Again, perhaps we simply have to wait to see more progress (but even then there would still be little proof that the national lockdown led to a more rapid viral retreat, since we do not know whether the epidemic would have petered out of its own accord).
Portugal meanwhile acted just slightly faster and to a greater extent than Spain in the policy arena. Their total deaths are one tenth of the Spanish level (and one sixth on a per capita basis) and yet policy is a bit tighter. Like Croatia to Italy, Portugal acted faster than Spain (by two weeks) in raising its policy level past 50 after having more than 25 contagion proxy deaths.
Unlike those other three countries, where there might be some effect of policy on contagion trends, there is little discernable effect of the stringent policy setting on the Portuguese pandemic, since lagged daily deaths were flat before and after the great tightening of policy.
The third country pair is Iran and Iraq:
Iran’s situation, like China’s, is probably not entirely accurate, given accusations of political manipulation of cases and deaths. However, the data taken at face value paint a picture of a country whose policy lagged well behind the contagion curve and where policy tightened several weeks after the contagion proxy peaked. The extent to which senior government officials were affected by the virus is perhaps an indirect indicator of the adverse politicization of their public health reporting.
Iraq, by comparison, looks very much like Croatia: early, rapid, aggressive policy measures from a mortality level that was tiny in comparison to most countries and certainly to Iran.
The overall lesson from these country pairs is that countries have not necessarily been coordinating but did learn from each other throughout the pandemic, especially those that were at the greatest risk from close neighbors. There has been transmission of policy ideas and actions but only to the extent that suited each country’s national interests. Globalization seems to be dying from disuse and its nationalist replacement (which was always there) is thriving in the vacuum left behind.
These country pairs have been examined to give some sense of how to look at the entire 150 country sample. It is not practical to show all of the data here (though I am willing to send anyone more charts and detail for any country upon request), so we will next turn to summary information for all of the countries together.
What do we find when comparing policy responses to the proxy contagion curves across the world?
1) Most governments did too little at first.
China is the poster child in this regard, since their failure to control the initial outbreak led to a global catastrophe.
The median date in our sample for a modest policy response (raising the Stringency Index over 25) was March 12, 2020, fully 11 weeks after China informed the WHO about the coronavirus.
Only 24% of the countries had this modest response before March. Notable early responders are on the geopolitical periphery of China: Mongolia, Macau, Singapore, Hong Kong, Malaysia, the Philippines, and Vietnam.
2) Most governments were slow to respond.
The median time lag from having more than 25 contagion proxy deaths (typically equivalent to about 1000 cases) to raising the policy level past a level of 50 was 11 days.
Fully 88% of all countries with more than 25 deaths had such a lag. Even 67% of all countries would still have had a lag if the average delay between contagion and death was three weeks instead of four weeks.
Notable laggards are Sweden, the US, the UK, Iran, France, and Spain, all intense pandemic countries.
3) Most governments took too long to act.
The median time lag between a modest policy response (index over 25) and the maximum policy response in each country was 17 days.
4) Most governments got really serious only after the peak in contagion.
The median time delay between the peak in our contagion proxy and raising the policy level past 50 was 7 days.
Over 86% of all countries with a policy level over 50 had such a lag. Even 62% of all countries would still have had a lag if the average delay between contagion and death was three weeks instead of four weeks.
Notable exceptions are the Philippines, Japan, Ukraine, and South Korea, all of which acted before the peak in contagion.
5) Delay cost deaths
This next chart is literally a killer: the policy lag discussed in 2) above by country and its correlation to today’s death rate per million. The red line shows that there is a weak positive relationship between the two, suggesting that the longer the country delayed its policy action, the greater the ultimate death rate.
6) The policy effect on flattening the curve is still unknown.
The maximum policy response for the world was reached on March 30, 2020 at a median level of 86 and 91% of the countries reached their maximum policy values by the end of the first week of April.
It is therefore too soon to measure the effect of this maximum policy pressure on the pandemic, given the four week lag in our contagion measure.
However, it is worth noting that the median world policy index rose sharply in March from 14 to 86 and so we should be seeing some of that effect, if it exists, by mid- to late-April.
Let’s all cross our fingers that May will be a better month for the pandemic, whether that decline is caused by humanity or nature.