BSC researchers study mobility in Spain during the COVID-19 pandemic period using data from the Facebook and Google apps

The work demonstrates how the correct use of these data can provide a detailed interpretation of mobility, for epidemiological and socio-economic analysis.

These data are a tool to evaluate the measures taken during the peak of the pandemic, and can contribute to improve the response to future mobility crises.

In retrospect, the confinement measures applied and their duration were adequate for the situation and evolution of the pandemic in Spain.

The objective of this study is to explore the relation between the movement restriction policies taken by the government, (during the period between March 1 and June 27), the mobility of people, and the evolution of the pandemic. The study has investigated the behavior of different sources of private data, and how they can be used in a complementary way to provide a better understanding of the mobility. Based on this data, the study analizes the mobility patterns in the 17 autonomous communities of Spain for four months. Results indicate a high degree of compliance with mobility restrictions policies by the population, sustained for a long period of time (seven weeks). The methodology used, as well as the conclusions drawn, can be a useful tool to react to future mobility crises.

The work had the participation of researchers from the Barcelona Supercomputing Center (BSC), the Universitat Politècnica de Catalunya (UPC) and the Comparative Medicine and Bioimage Centre of Catalonia (CMCiB).

Data

A highlight of the research is that mobility data comes from tracking applications of Facebook (Facebook Data for Good Program) and Google (Google Mobility assessment), installed by users on their mobile phones. This data is only captured for those users who have given their explicit consent. Although in this data there are segments of the population that are over-represented, such as that of young people, the sample is broad and representative (it is estimated to be close to or greater than 1% of the entire population), and therefore their estimates are considered reliable.

This is data published by these companies in an open manner, which both scientists and governments use as a source of analysis. This study analyzes their peculiarities, identifies and clarifies their anomalies, and shows how they can be used in a complementary way to provide a more complete vision of mobility.

The study shows how an absolute measure like the one provided by Facebook (which measures the volume of people who do not move) can be useful for epidemiological purposes. For example, identifying Sundays as the days when mobility is most minimized, making them candidates for the first measures of de-confinement, and for the mobility of the population at risk. Instead, for a relative measure such as that provided by Google, the study suggests its use for socio-economic purposes. For example, identifying Fridays as the days when normality is most disturbed, making them candidates for actions of moral and psychological support (official communications, cultural actions, social events).

Mobility and the pandemic

The mobility study confirms that the level of confinement that was reached in Spain was very high for seven weeks (March 15 to May 2), close to the limit that guarantees the fulfillment of essential services. To make this estimate, the study compares mobility during the two weeks of hard-lockdown and economic hibernation with mobility during the rest of the confinement, obtaining a difference of between 10% and 15% in absolute mobility. Although the relevance of these numbers from the epidemiological point of view is unknown at the moment, the researchers consider it possible that the phase of hard-lockdown contributed to raising awareness in society, increasing compliance with restrictions.

To get an approximation of the impact of confinement on the COVID-19 pandemic, the researchers cross-reference the mobility data with the cases detected 14 days later. The study shows a clear correlation between both measurements, the end of the exponential phase of the curve overlapping with the declaration of the alarm state. These data also indicate that the onset of de-confinement coincided with the dates on which the pandemic reached a controlled status.

The new normality

The study ends on June 27, a week after the alarm state in Spain was lifted and the new normality began. This last month of data shows a gradual recovery from normality, accelerated in the most rural and sparsely populated regions. Upon reaching the new normal, weekends show mobility rates similar to those before the pandemic. On the other hand, working days show clear reductions, mainly in Barcelona and Madrid, where mobility has decreased between 10% and 15%.

Dario Garcia Gasulla, researcher from High Performance Artificial Intelligence group at BSC, explains that “our work focuses on mobility data, which is not the same as physical proximity. Although weekend mobility rates are already at pre-COVID levels, it is possible that the type of mobility is different and reduces physical proximity.”

• Link to article Private Sources of Mobility Data Under COVID-19 https://arxiv.org/abs/2007.07095

Evolution of mobility in the 17 autonomous communities, from March 1 to June 4. On the vertical axis, reduced mobility according to Google, compared to pre-COVID dates. On the horizontal axis, reduced mobility according to Facebook, in absolute terms. The animation shows the drastic reduction of mobility (top right) with the declaration of the state of alarm. Mobility began to recover on May 1, with the generalized deconfinement measures.

This article was first published on 23 July by BSC.