Afcacia

Researchers and health organizations around the world are advocating the importance of accessing aggregated and anonymized mobility data tools to help them better track and respond to Covid-19. These tools are critical to large scale public health responses because they help provide a more accurate, real-time picture of what is happening.

Social media giant Meta, which holds data of more than 2.5 billion people, recently launched new disease prevention map tools to support the Covid-19 disease tracking and research efforts in what it calls Data for Good project.

This was done through a virtual press call attended by journalists in the Europe, Middle East and Africa region and technology journalist Faustine Ngila took part as Laura McGorman, Facebook’s Public Policy Manager, Data for Good, answered critical questions from the media.

1.     What does Data for Good aim to achieve and why is it important?

We are launching new geo-location maps that reveal the probability that people in areas with disease outbreaks will come in contact with new populations, greatly improving insights on the actual chance that the disease will be spread by human to human contact. The maps will also reveal trends in staying near your home, which will generate statistics at a county and regional level that show whether populations are generally staying in one part of town, or whether populations are on average are visiting many parts of a city or town. These are new maps on social connectedness, which share information on the rate at which friendships exist across states and countries. Data on social ties also play a role in forecasting the likelihood of disease transmission.

2.     Experts say that weak anonymization methods might allow data to be traced to an individual. How do you ensure the data you use for your maps are fully anonymised? When you share data with researchers how do you protect individual privacy?

In building data products, we take care to protect people’s privacy. When handling data from non-public sources, our data teams use methods including preventing re-identification, which includes processes used to prevent a person’s identity from being connected with information. We also use aggregation, which draws insights from the total number of people in a particular geographic region, rather than from individuals. And then smoothing, which combines population estimates for sparsely populated regions with nearby areas to avoid the possibility of re-identification. We also follow legal procedures like data-sharing agreements, which ensure that our data is shared only with specific organizations that will use it for specific purposes.

3.     Telecommunication companies are talking with governments about sharing location data to assist with contact tracing. Are you planning to do the same?

There is no agreement to share people’s location data with governments. We already support researchers and NGOs with our Disease Prevention Maps, built with aggregate data.

4.     Is the Data for Good initiative GDPR-compliant? How are you getting consent from users before sharing their data with researchers and institutions?

Both recent statements by the European Data Protection Board and Irish Data Protection Commissioner on COVID 19 have underscored that GDPR is no barrier to industry processing and even sharing data, provided that appropriate privacy measures are implemented.

5.      How many Data for Good partners do you have in Europe?
We have a range of partners, with more being added. We have researchers working with our mobility data in every region of the around the world such as Harvard School of Public Health, Colorado State University, University of Colorado Boulder, the Institute of Disease Modeling, the Bill and Melinda Gates Foundation, the World Bank, Direct Relief, National TsingHua University in Taiwan, Wadhwani AI and IIT Tirupati in India, University of Pavia and Politecnico di Milano in Italy, Universitat Politècnica de Catalunya in Spain, University of Southampton, London School of Hygiene and Tropical Medicine and Oxford Big Data Institute in the UK, Tecnológico de Monterrey in Mexico and Universidad del Rosario in Colombia

6.     Where has Facebook disease prevention map data been used by authorities and to what end?
The World Bank used Facebook’s Disease Prevention Maps to look at how prepared existing health care facilities in Spain are to accommodate patients. While modelling the rate of infection is an extremely complex and data sensitive process, it is still nonetheless extremely important to estimate the degree to which facilities might need extra beds.

Since the onset of the Covid-19 pandemic, we have onboarded dozens of trusted partners who can use Disease Prevention Maps to combat the spread of the disease. These partners range from universities in the US like Harvard School of Public Health, to those in Taiwan like National TsingHua University, or in Italy like University of Pavia, as well as nonprofits and other institutions such as Direct Relief, the Bill & Melinda Gates Foundation, and the World Bank.

The Institute of Disease Modeling at the Gates Foundation used our mobility data to forecast transmission of Covid-19 in King County in Washington State in the US, the World Bank used our publicly available tools to determine where additional beds might be needed in Spain, and National TsingHua University used our collocation maps to better understand transmission rates in Taiwan.

7.     What type of user data is Facebook providing to its Data for Good partners?
Facebook’s High Resolution Population Density Maps aren’t built using Facebook data and instead rely on combining the power of machine vision AI with satellite imagery from Maxar and census information from countries, which are already anonymized and de-identified data sources. Maps that use Facebook data, like Movement and Network Coverage, also include only de-identified and aggregated data from the Facebook platform.

8.      Has Facebook been asked by governments in Africa and Europe to provide user location data and what is its position on providing such data?
There is no agreement to share people’s location data with governments. We already support researchers and NGOs with our Disease Prevention Maps, built with aggregate data.

9.     Why are you not providing data to governments like Google?

That’s a question for Google. I’m not in a position to speculate on their decisions. Public sector officials have access to our publicly available tools and resources. We encourage governments and health authorities looking to combat the spread of COVID-19 to work directly with our research partners who study our Disease Prevention Maps through our Data For Good program.

10.   Google just released mobility maps, what is the difference with Facebook’s Disease Prevention Maps? 

Google has released public reports that show the rate of change in people visiting particular points of interest, such as retail shops and grocery stores, at a country level. Our maps measure whether communities are likely to come in contact with each other generally, whether they are generally staying home, as well as information on social ties, and are at the regional level, which is more granular. Google’s maps are also released in pdf format publicly, whereas Facebook shares daily data on aggregated mobility with researchers in machine readable format so that this information can be directly analyzed to inform the public response.

11.   Why are you not making all your data public and available in a legible format?
Some Data for Good products are publicly available, such as our high resolution population density maps, which are built with satellite imagery and census data. These are built with already publicly available datasets and are simply improved through Facebook AI capabilities.
For datasets built with aggregated user information, we make these insights available to nonprofit organizations and research institutions with data licensing agreements. Access to these datasets is administered through a password protected mapping portal and follows strict access control guidelines.

12.   What data are you using for these maps?  
The primary source of information for Disease Prevention Maps is aggregated insights from people using Facebook on their mobile phones who have opted into a feature called location history. The social connectedness index is an index created from the likelihood of friendship across different geographic boundaries.

13.   What control do users have over this data?
We only collect precise location data from the user’s device if the user chooses to share it with us through Location Services. If a person has not turned on location history, we will not build a history of precise locations received through Location Services.
[If pressed on location data]: We have higher level location information based on things people do on Facebook – includes things like your stated location on your profile, and your network connections. Where we use more precise level location data from people’s devices, this comes from Facebook users who’ve opted into share their precise location with FB through the device level setting, and enabled the “Always” collect option, and users who have, in addition, opted into location history (which is a choice to enable FB to store that location data).
We then take that information and aggregate it within the geographic areas, which we call “tiles” or administrative boundaries (boundaries around counties, cities, or states).

14.   What kind of health data is included in these maps?
There is no health data included in these maps. Disease prevention maps focus using inputs such as aggregated mobility data to improve the way health organizations respond to outbreaks and administer health commodities. These groups will continue to rely on epidemiological data from Ministries of Health and groups like the World Health Organization — our role is to fill gaps in other kinds of population data that are needed to effectively respond.

15.    Does Facebook know or map where people who have a disease are located?
No, there’s no health data of any kind included in disease prevention maps. This effort is instead focused on filling population-based data gaps that health institutions face to do their work better — not about sharing health data.

16.    How can you control access to the data tools so they aren’t used for harm?
Data for Good at Facebook builds products with both Facebook data and non-FB data and the permissions associated with each product type are different. In both cases, access control is designed to protect individual privacy and the safety of vulnerable populations. Some Data for Good products are publicly available, such as our high resolution population density maps, which are built with satellite imagery and census data. These are built with already publicly available datasets and are simply improved through Facebook AI capabilities. For datasets built with aggregated user information, we make these insights available to nonprofit organizations and research institutions with data licensing agreements. Access to these datasets is administered through a password protected mapping portal and follows strict access control guidelines.

17.   What are your initial analyses showing?
Facebook provides the aggregated mobility data and colocation probabilities to our health research partners and the epidemiologists themselves that build disease forecasts. For example, the Institute for Disease Modeling recently utilized Facebook aggregated mobility data for the State of Washington to determine the rates at which mixing among workforces had been similarly disrupted to inform transmission forecasts.