On March 19, the governor of California’s order for all citizens to stay at home except “as needed to maintain continuity of operations of the federal critical infrastructure sectors” went into effect statewide. Over the prior 24 hours, thousands of California residents bustled from their homes to take advantage of the remaining time. They went shopping for groceries. They took care of last-minute errands. They visited friends and family. All this activity was in preparation for the most severe movement restrictions in the history of the state, in response to the threat posed by the Covid-19 novel coronavirus.
California had to impose drastic movement restrictions, known in public health circles as “non-pharmaceutical interventions,” or NPIs, in order to create more physical distance between people, slow the spread of the virus, and reduce pressure on already stressed hospital and primary care systems. Without vaccines or treatments available, NPIs, like physical distancing, are the only tool available to mitigate the virus and “flatten the curve” of new cases.
Californians followed these orders at high, but varying, rates throughout the state. Urgent questions presented themselves. Was it possible that some of that increased mobility which occurred in the day immediately prior to the stay at home order may have played an inadvertent role in spreading the virus, simply by temporarily increasing the number of individual points of contact? How could one know if that was the case? Why did people in some places appear to move more than others? And if it turned out to be the case afterward that the messaging of distancing orders played some role in viral spread, might it also be possible to design future distancing orders, which might have to be re-imposed periodically depending on the course of the pandemic, with greater precision?
Mapping Physical Distancing
We know this uptick in movement took place in the lead-up to March 19, in the aggregate, because of a new set of measures and maps being generated from the distributed activity of hundreds of thousands of mobile devices using the Facebook app with location history enabled. Facebook compiles this information into “disease prevention maps” which locate large numbers of users into blocks of space which are 600 meters on a side and updates the distribution of those locations every eight hours throughout a crisis event.
Direct Relief and many other researchers have been using these types of maps for a couple of years to do things like monitoring evacuations from wildfires and coastal flooding. During Covid-19 response, they have become even more important as a constant barometer of the progress of non-pharmaceutical interventions to slow the pandemic.
Two new measures from Facebook Data for Good are being used to track changes in population mobility. One, called “relative mobility,” looks at the frequency with which users have traveled outside their home location compared to a baseline frequency from the month of February, before physical distancing policies began to be applied across the country. Another, called “staying put,” looks at the proportion of all users with location history enabled who remained in their home location for an entire 24-hour time period. Whereas “relative mobility” calculates a percentage increase or decrease in the amount of movement occurring compared to a previous “normal” period, “staying put” is an absolute measure of the proportion of users on any given day who have significantly constrained movement patterns. The two measures are highly correlated, but they are not identical.
Based on these two measures we can see that practically everywhere across the country has seen some level of decline in mobility since the middle of March. Americans have measurably slowed down in response to distancing orders. However, there is considerable heterogeneity in the data. As of mid-April, in California for instance, only 10 counties out of 58 are below the state average for mobility reductions. These 10 are concentrated almost entirely on the coast – in the Bay Area, Los Angeles, Orange County, San Diego and Riverside. More rural areas have seen reductions, but far less than the major urban centers. This gap between urban areas and the rest of the state has increased in California over the past two weeks at least, and in some states for longer.
Rural and Urban Relative Mobility
The gap between urban and rural rates of mobility has echoed throughout the country.
States that contain higher proportions of rural areas than the national average, from Arkansas and Alabama to Idaho and Montana, have higher overall rates of relative mobility, compared to baseline, than their more urbanized counterparts. The average rate of mobility reduction in New York and New Jersey, for instance, remains more than three times greater than in Idaho, Montana and South Dakota. This gap raises important concerns about relative risk exposure over time, even if more rural areas are not quite at the forefront of the case curves yet.
It is important not to jump to conclusions about exactly why rural areas may be exhibiting lower rates of mobility reduction. This is a multi-dimensional problem. For instance, rural areas simply have lower rates of population and infrastructure density, which means people need to travel further to reach services like grocery stores and pharmacies. Likewise, rural areas tend to have fewer jobs which lend themselves to being virtualized; if you’re working in agriculture or warehousing and logistics, you do need to show up for work in person.
These kinds of in-person contacts are one reason that rural areas remain at considerable risk of Covid-19 infection; lower rates of distancing adherence may prove to be of increasing concern over time in terms of public health impacts and health system pressures.
Like the rural-urban divide, evidence also exists of correlation between areas with higher percentages of people over the age of 65 and lower reductions in rates of mobility. In many states, including California, the higher the proportion of people over 65, the more one’s county tends to move around relative to baseline. In part, this may be because urban areas tend to have higher proportions of younger people, who also may tend to work in occupations that are easier to do remotely. It may be due to the probability that older people live in areas of lower density, which is also correlated with more movement.
Nevertheless, it’s still the case that this correlation exists, and that it remains a cause for concern, given that age above 65 remains among the strongest predictors of the likelihood that individuals will not only become infected with Covid-19, but also become hospitalized and experience acute complications.
Helping the Public Sector by Coordinating the Covid-19 Mobility Data Network
As Facebook’s CEO Mark Zuckerberg argues, the new types of mobility data being created now for real-time analysis of society are potentially a kind of “superpower” for pandemic response, but only if we can get them quickly into the hands of people who can use them to set and modify health policies.
For the past several weeks, Direct Relief has helped to bring insights and information concerning aggregated rates of population mobility to the attention daily of health officials and policymakers throughout the U.S. and other parts of the world. In California, Direct Relief is participating daily as part of the data team informing the governor’s Covid-19 response policies. At the same time, Direct Relief has helped to create and co-coordinate, along with colleagues at the Harvard T.H. Chan School of Public Health, the Covid-19 Mobility Data Network, which links together dozens of infectious disease experts and data scientists in a voluntary effort to provide operational analytical support to key public decision-makers throughout the Covid-19 response.
Since the beginning of March, teams of researchers in this network have supported public actors from New York City and the state of California, to Massachusetts, Florida, Michigan, Illinois and Kentucky. Important new work was done in Seattle, Washington, in collaboration with the Gates Foundation and IDN at the outset of the crisis to begin understanding how the reduction in mobility might be reducing mortality from the virus.
Researchers across the Covid-19 Mobility Data Network have been supporting in Italy, Spain, and the UK in the same fashion, while booting up support across India, and increasingly, in Latin America as well, from Santiago, Chile, to Mexico City. The number of locations requesting and receiving support continues to expand almost every day. The number of new data sources also continues to expand, allowing researchers to compare Facebook’s location data with related data from ad tech firms, traffic patterns, financial transactions, and points of interest data which help to fill out the picture of real-time social activity.
Privacy and Data Protection Remain Crucial
The main reason it’s important for nonprofits like Direct Relief, and for the academic research community more broadly, to work with this type of data to inform public actors is to ensure that insights from this data can quickly reach the right people in positions to use them, without violating privacy and data protection laws and policies.
There is no need to spy on everyone in order to use research-based data agreements to improve understanding of who may be following physical distancing guidelines as part of pandemic mitigation. These agreements were created for the express purpose of allowing vital data to be used without creating conditions where private data is simply being funneled unfiltered to government actors.
As we move into a probable future over the next 18-24 months where distancing guidelines may have to be relaxed and re-imposed on a semi-regular basis in order to reduce pressures on health systems, aggregated human mobility data will be one of our best and in some cases only guides as to whether these policies are working and having their intended effects.
Direct Relief in other parts of its program activity works to keep health workers safe through increased access to personal protective equipment, to expand access to testing, and to keep frontline health centers operational even while they’re experiencing massive economic shocks. But all these efforts depend, for their long-term success, in part on reducing case counts and working together with public sector and private tech company colleagues, like those at Facebook.
Analysis of aggregated mobility data will continue to be in itself one of the most important contributions we can make to the broader fight against Covid-19.