Measuring the Effectiveness of Social Distancing Policies
Location data shows that New Yorkers’ response to COVID-19 has led to major changes
By Irving Wladawsky-Berger
One of the most interesting projects in the MIT Connection Science group I’m affiliated with is The Atlas of Inequality. The project,— led by visiting professor Esteban Moro and professor Alex ‘Sandy’ Pentland — uses aggregated anonymous geolocation data from digital devices to estimate where different groups of people in U.S. cities spend their time. Over time, the project will be expanded to cities around the world. The data show the significant income inequality among people in those cities, not just by neighborhoods as you’d expect, but also in the restaurants, stores and other places beyond their neighborhoods which they visit every day.
A couple of weeks ago, the Atlas research group started to apply its geolocation data and methods to analyze the effectiveness of the social distancing policies adopted in the New York metropolitan area in response to the Covid-19 pandemic. These policies include school closures, bans on non-essential gatherings, limiting restaurants to take-out orders, and strict stay-in-place measures. There’s no way to empirically measure the impact of these social distancing measures in real time on the spread of Covid-19. You can only measure their impact retrospectively, or simulate what might happen in the future based on past data.
However, you can empirically address a key set of important question: How have these social distancing policies changed mobility and social behavior? How does social distancing behavior vary across the New York metropolitan area? How does the behavior vary across the variety of demographic groups in the N.Y. metro area? and overall, how well are people following these social distance measures?
The initial findings reveal that N.Y.’s social distancing policies have led to major changes in where people spend their time and how they interact with each other.
“Distance travelled everyday dropped by 70 percent from a weekend average of 25 miles in February to 7 miles” toward latter part of March; “the number of social contacts in places decreased by 93% from 75 to 5,” where social contact is defined as being within 25 meters (82 feet) of each other for at least five minutes. “The number of people staying home the whole day has increased from 20% to 60%” and “social distancing policies have greatly reduced relative differences between different demographic groups as nearly everyone’s mobility and social contacts has been dramatically reduced.”
The changes in distance travelled and social contacts became significant only after non-essential business closure measures were put in place on March 22. Retail food and essential supply stores are now the most common places for social contacts. After those measures were introduced, about 5.5% of New Yorkers started to spend time in places beyond the metro area, including New Jersey (37%), upstate N.Y. (23%), Pennsylvania (9.8%) and Florida (6.7%).
More detailed findings can be found in the draft report. Let me briefly describe the sources of the data used in the analysis as well as the methods used to preserve data privacy.
The kind of central government directives that were deployed in China to combat its Covid-19 outbreak aren’t applicable in the U.S. and other free-market democracies. In these countries, it’s thus important to turn to sophisticated data analysis methods that are compliant with privacy policies.
The primary data source for the Atlas of Inequality project is anonymized location data from a variety of applications on smartphone devices. The data comes from Cuebiq, a geolocation-based intelligence and measurement company, and in particular, from Cuebiq’s Data for Good initiative which makes its data available for academic research and humanitarian programs.
For the N.Y. social distancing analysis, “Cuebiq collects anonymized records of high-resolution timestamped GPS points from users who opted-in to share their data anonymously across the U.S. from January 1, 2020 to March 25, 2020.” Mobility data is only extracted from those users who opted in to share their data through a GDPR and CCPA compliant framework. Residential and work areas data is then aggregated to the Census Block Group level, allowing for the demographic analysis while obfuscating the exact location where the anonymous users live and work.
“The data we received is constructed from the sequence of pings reported by devices,” explains the report. “This results in a dataset of the public places where many people have stayed (with high spatial accuracy) corresponding to the points of interest that people typically visit and the most likely census tracts of where these device owners live and work.” A stay is defined as a place where an anonymous user stopped for at least five minutes. The analysis is limited to data from people who were active during February 17 to March 9, and for whom there is location data reporting that they stayed in their home Census Block Group more than 10 days. The dataset includes information on 567,000 people.
There’s much, much more to be done. “The next empirical question is how effective are these social distance policies at reducing the spread of the coronavirus,” notes the report in conclusion. “With high-resolution anonymous mobility data, we can study the effect of mobility-related policies on population-level behavioral responses and how the coronavirus spreads differently across places with different policies.
Not only can high-resolution, but anonymous and aggregated mobility data monitor adherence to social distancing policies, it can also inform epidemiological models based on real time contact matrices.”