On April 7, Facebook launched new tools that use “aggregated” and “anonymised” data on population movement, collected from users, that can show whether social distancing measures are working or when outbreaks are most likely to occur next. The tools, developed under the Facebook Data for Good program, offer Disease Prevention Maps, which show movement patterns at a region or country level that researchers can use to understand how diseases like COVID-19 spread.
- Co-location maps show the chances of people in one area coming in contact with people in another area, based on commute patterns for instance, to show chances of COVID-19 transmission to newer areas.
- Movement range trends uses data of people using Facebook on their mobile to show whether people are staying put in one location or visiting many parts of a region/county. Movement maps can also show if social distancing policies are being followed or enforced.
- The social connectedness index shows friendships across states and countries, which can help epidemiologists forecast the likelihood of disease spread, as well as where areas hardest hit by COVID-19 may need support.
A co-location map from Italy. Source: Facebook
This map shows how connected East Village, New York is to the rest of the US. Source: Facebook
Network coverage maps are among the maps being opened for research, it shows where people on Facebook have cellular connectivity. “All data is de-identified, aggregated, and visualized as a polygon of network coverage by speed,” Facebook said.
Google announced a similar project last week, releasing data based on “aggregated, anonymized data” from Google Maps to track how people’s movement habits have changed since lockdowns were implemented across the world to contain the COVID-19 pandemic. For instance, visits to restaurants, cafes, shopping centres, etc. have reduced by 77% in India, after the Janata Curfew imposed on March 22, followed by a nationwide lockdown on March 25.
Unlike Google, that has made data available for 131 countries, and state-specific data for the US, it is unclear how many countries or locations Facebook’s maps are available for. Also, researchers have to email at diseaseprevmaps@fb.com to get the aggregated data from Facebook.
Researchers can get access to the aggregated data by emailing diseaseprevmaps@fb.com 🙂
— Eugenia Giraudy (@EugeniaGiraudy) April 6, 2020
Privacy measures: Facebook did not specify any privacy-protecting measures it is taking. , but clarified in its announcement that all Facebook data is aggregated at the state or country levek, and do not show individuals’ patterns.
Survey to gather data on symptomatic cases
Apart from these tools, Facebook is also launching a survey developed by Carnegie Mellon University’s (CMU) Delphi Research Centre, which is designed for public health officials to figure out where lockdown guidelines should remain or be expanded. The Delphi Research Group is one of two Influenza Forecasting Centers of Excellence designated last year by the US Center for Disease Control and Prevention.
The survey is aimed at gathering data about US residents who are symptomatic for COVID-19 at a county level — to generate data about how to respond to an outbreak — by generating heat-maps of self-reported symptoms for instance. “People have been discouraged from visiting physician offices and hospitals,” associate professor Ryan Tibshirani noted. “The only way to get this is with the survey,” he said. The Delphi group is working with Google, which ran an initial survey last week, and is reaching out to more companies.
“If this works, it can produce a weekly map that may help more accurately forecast how many cases hospitals will see in the days ahead, as well as an early indicator of where the outbreak is growing and where the curve is being successfully flattened.” — Mark Zuckerberg
Privacy measures: Facebook is providing the data, but is not involved in the survey, Tibshirani clarified. The company will share a with a random ID number with CMU, which they will return once an individual completes the survey, without the actual replies. Facebook will then provide a statistic known as a weight value that will help CMU correct for any sample bias.
COVID-19 Mobility Data Network to help local officials evaluate social distancing measures
Facebook is also working with over a dozen universities and researchers including IIT Bombay, IIT Madras, and IIT Tirupati in India, and Princeton and Harvard among others, as part of the COVID-19 Mobility Data Network. Researchers in the network seek aggregated anonymised information on movement from Facebook, Camber systems, and Cuebiq.
Unlike Movement range trends maps, where the maps are provided solely for research uses, the Mobility Data Network researchers will use the aggregated data provided by the above-mentioned three companies to provide daily situation reports on social distancing policies to local government health officials. The network is connected with departments of health at the city, state, and country-level, they claimed. They support the local measures by evaluating the effectiveness of social distancing measures, identifying potentially high-risk zones, and for planning the roll-back of restrictions.
Broad guiding principles for protecting privacy: The network laid out a few principles for the project, including some aimed at protecting privacy. Some of the principles are:
- Use of the data, including data sharing, aggregation, and analysis, for COVID19 response must speak to a clear need articulated by public health authorities, and for no other purpose
- Data use must in compliance with existing laws and adhere to best practice principles of data governance.
- The network will only share analysis with government health officials, and will not share underlying data sets with governments, third parties, or the public at large.
- Research participants will have to embrace “appropriate legal, organizational, and computational safeguards” to minimise and/or manage data privacy risks associated with this research effort in general and the use of aggregated data in particular.
