In December 2019, the COVID-19 outbreak was registered in Wuhan, China. The World Health Organization declared it a ‘Public Health Emergency of International Concern’ on 30 January 2020 and escalated it to a pandemic on 11 March 2020. In an attempt to save human lives and shield health systems from being overwhelmed, a number of lockdown policies were implemented around the world. As important as they may have been for public health, these measures contributed to an unprecedented economic shock with several implications across many dimensions. These include, among others, economic policy (Gourinchas 2020), the future of labour (Baldwin 2020), inequality (Furceri et al. 2020), long-lasting economic damage (Portes 2020), and the possibility of lessons to be learned (Wyplosz 2020).
Against this backdrop, it is critical to understand which policies have proven thus far to be the most effective in containing the virus, and to describe the channels through which these effects operate. In our recently released study (Askitas et al. 2020), we evaluate the effect of lockdown policies on the daily incidence of COVID-19, as well as on various population mobility patterns across 135 countries.
We use data on non-pharmaceutical interventions collected by the Oxford COVID-19 Government Response Tracker (Hale et al. 2020), daily number of infections from the European Centre for Disease Prevention and Control (ECDC), and population mobility data from the Google Community Mobility Reports.
We consider eight interventions, which vary in their intensity and timing. The policy responses in focus are: (i) international travel controls, (ii) public transport closures, (iii) cancelation of public events, (iv) restrictions on private gatherings, (v) school closures, (vi) workplace closures, (vii) stay-at-home requirements, and (viii) internal mobility restrictions (across cities and regions).
For each of these policies, we exploit information on the date of introduction as well as qualitative time-varying information on their intensity. Intensity is measured on a scale from one to six, which reflects whether the intervention is (i) recommended, (ii) mandatory with some flexibility, and (iii) mandatory with no flexibility, and whether it is geographically targeted or applied to the entire country.
We also link policy interventions to mobility patterns across several types of places such as: (i) retail and recreation, (ii) grocery and pharmacy, (iii) parks, (iv) transit stations, (v) the workplace, and (vi) residential areas. Each of these is characterised by different epidemiological features (e.g. numerosity, density, behavioural norms, epidemiological range, back tracing ability etc.) and, therefore, has a different contagion potential. The mobility data can then be viewed as a measure of compliance to the policies introduced as well as a mediator between policies and the spread of the disease.
We develop a multiple-event model to estimate the dynamic effects of each intervention, while taking into account the presence of concurrent interventions. We can estimate the net effect of each policy, in the presence of other policies, by exploiting the variation in their intensity over time, and across policies and countries. Accounting for confounding policies is important because it prevents attributing the effect of other interventions to the policy of interest. In addition, it allows us to establish that these policies mitigate the spread of the virus by affecting mobility patterns.
The main result of the analysis is reported in Figure 1, which shows that cancelling public events, imposing restrictions on private gatherings, and closing schools have the quantitatively most pronounced effects in reducing the incidence of COVID-19. They are followed by workplace closure and stay-at-home requirements, whose effects are not as pronounced. Instead, we find no effects for international travel controls, public transport closures, and restrictions on movements across cities and regions.
Figure 1 Effects of lockdown policies on COVID-19 confirmed new cases with controls for concurrent policies.
Cancelling public events and imposing restrictions on private gatherings start to lower the incidence of COVID-19 about one week after implementation, becoming statistically significant within two weeks. Around the end of the event window (35 days after implementation), a unit increase in the intensity of the policy of interest leads to a 20% decrease in the number of new infections in the case of public events cancelation, and a decrease of about 12% in the case of restrictions on private gatherings. For school closures, we find that new infections start declining a few days after they are closed, with the effect becoming negative and significant about 25 days after implementation. Around the end of the event window, a unit increase in the intensity of school closures leads to about a 15% drop of new infections.
It is worth noting that estimating the effect of each policy while ignoring the contemporaneous influence of multiple interventions would have led to the erroneous conclusion that all policies are effective in reducing new infections. This is shown in Figure 2, which demonstrates the importance of accounting for confounding policies.
Figure 2 Effects of lockdown policies on COVID-19 confirmed new cases without controls for concurrent policies.
In the second part of the analysis, we link lockdown policies to mobility patterns in order to shed light on the mechanisms through which they help flatten the curve. We find that lockdown policies tend to increase time spent at home, and their impact on the incidence of COVID-19 is determined by a number of factors. For instance, cancelling public events, and to a lesser extent restricting private gatherings, lower new infections by reducing exposure to numerous and dense locations, where contact tracing is difficult, and can have a large epidemiological range within and across countries (e.g. football games, concerts etc.). Workplace closures, instead, restrict activities away from home but have a lower impact on lowering new infections possibly because of the differences in numerosity, density, behavioural norms, and ability to trace new infections in these environments.
Quantifying the net contribution of each lockdown measure in containing the pandemic is a highly relevant empirical question with important policy implications because of the public health benefits as well as the economic costs associated with these restrictions. We find that policies preventing close contacts in large groups, such as public events, private gatherings, and schools are the most effective in reducing new infections. Instead, we find no effects for international travel controls, public transport closures, and restrictions on movement across cities or regions. That travel controls had no impact, although imposed relatively early in many countries, is likely explained by their lack of stringency, allowing the virus to cross borders. The framework proposed in this work can also be applied to evaluate the upcoming exit strategies, as well as offer guidance in possible subsequent waves of this or other future epidemics.