With more detailed data on the varying introduction of policies across US counties and over time as well as the resulting spread of the disease we could provide evidence on which measures we should keep and which measures we should lift in order to reactivate the economy (see e.g. this post by Tomas Pueyo on “Coronavirus: The Hammer and the Dance” and the need to quantify the contributions of different measures to this trade-off).
There are great projects gathering necessary data and contributions are very welcome:
- collaborative collection of data on the interventions by Keystone Strategy partnered with Susan Athey, Stanford Professor of Economics, and Marco Iansiti, Director of Harvard Business School’s Digital Initiative; follow this link to contribute
- data on cases and deaths by The New York Times based on reports from state and local health agencies that the NYTimes use for their U.S. tracking page
Which NPIs did US counties implement and when?
As shown in the figure above, there has been a major wave of implementation of NPIs between March 9 and March 21 in many US counties. One of the “early” NPIs was social distancing for vulnerable groups from early March on. Most counties adopted more restrictive NPIs like gathering size limitations and the closure of public venues, schools and universities as well as non-essential services within a very short time period of just a few days mid-March. By late March, social distancing for all has become common in most counties.
What can we learn from the data?
We can use the staggered adoption of NPIs in US counties to learn about the contribution rates of the different measures by comparing the change in growth rates between US counties that implement NPIs vs. those who do not as well as different types of NPIs at varying points in time. As an example, when vulnerable groups stay at home and gatherings of more than 100 people are not allowed, how much does the closure of schools and universities contribute to decreasing the growth rate of confirmed cases?
In which order and at which point of the spread of COVID-19 did counties take on the NPIs?
Let’s compare Santa Clara County, San Francisco and New York City to illustrate the different developments in the figures below.
Santa Clara (figure above) already introduced social distancing of vulnerable groups on March 5 when they had their first 20 cases. Following an increase in cases, gatherings were limited in size. Schools, universities, public venues and finally non-essential services were closed until March 17 when 155 confirmed cases and five deaths were reported. Since then, social distancing policies are in place as well.
San Francisco introduced social distancing for vulnerable groups on March 8 with 11 confirmed cases. Within five days, the city then introduced similar measures as Santa Clara county from March 11 to 16 with a total of 43 confirmed cases and no deaths on March 16.
New York City introduced their first NPI on March 12 when they reported 96 confirmed cases. Additional measures followed between March 16 and March 22. When NYC introduced social distancing on March 22, they already had 9,045 cases.
These different timelines potentially contributed to the high variation in the number of cases present today — only one month after Santa Clara introduced their first NPI. With a potential delay in the actual effect, these NPIs have the potential to change the growth rate in the number of COVID-19 cases and hence “flatten the curve”.
Comparing these timelines provides suggestive evidence on which measures are effective. Given the nature of exponential growth of COVID-19 cases, any flattening of the curve is likely to be caused by the interventions implemented.
Where are the pitfalls when analyzing the data?
However, it is difficult to provide exact causal estimates of the impact of each measure. Keep in mind that we only have data on 29 US counties and their stories differ in many dimensions that we would like to account for:
- potential differential effects of combinations of NPIs and delayed effects of previous interventions
- differences in test rates causing differences in reported number of cases by county or focus on deaths (see suggestion by Tomas Pueyo here)
- other characteristics affecting the baseline reproductive number of COVID-19 like higher population density making social distancing more difficult, better healthcare infrastructure implying a higher capacity of intensive care units, differential composition of the workforce and their ability to work from home, cultural perception and importance of gatherings
- additional problems arise due to spillovers between regions — if people are allowed to travel between counties, stricter policies in one county might be less effective
More data on other US counties and additional insights gathered collaboratively will help us understand which policies should be optimally implemented.