How many people in your county have been in a COVID-19 hotspot in the last two weeks? How many other people are you being exposed to when you visit a public place such as a restaurant or a park? Using geolocation data, Chicago Booth’s Jonathan Dingel and his coauthors have created a pair of indexes that provide a daily, county-by-county snapshot of potential exposure to COVID-19.
We produced two different indices that are designed to measure movements between places and social contexts in venues like bars and restaurants.
The first index is called the Location Exposure Index. What we do is we follow the movement of phones between different counties in the United States. So New York City and Seattle were the parts of the US that were hit first by the pandemic. A very interesting question is simply to say: How many phones were in New York last week but now are in my county this week? Because the COVID-19 disease has an incubation period of something like one to two weeks, which means you care about where people have been and been potentially exposed before they started having symptoms.
This kind of tracking of movement between different places in the United States leverages cell-phone data because you wouldn’t have been able to produce statistics like this using traditional data sources. That Location Exposure Index describes movement between counties, and that let us produce a snapshot on a daily basis of where phones in one particular place might have come from across the country.
Of course, we’re looking at phone-movement data, so we don’t know if they are infected with the disease or not, but this describes potential exposure in terms of a bunch of people that were in Seattle in the middle of March and have shown up in my city on April 1, for example. We’re publishing those numbers for the 2,000 largest counties in America. So we’re not publishing information about the movements of devices in places that only have hundreds or maybe a thousand residents.
These pieces of information don’t describe individuals in a way that would violate their privacy. It describes the average behavior of a phone for a geographic unit that’s sufficiently large that we’re confident we’re preserving individual privacy. We as the researchers have no idea about the identity of the phone user. It’s just a randomly generating advertising ID—that’s a series of numbers and letters that don’t mean anything. But because we follow the phones’ movements over time, we want to make sure that we publish this information in a sufficiently aggregated way that everyone remains anonymous and has their privacy protected.
The second index that we developed is called the Device Exposure Index. This describes overlapping visits to venues like parks or restaurants or retail stores in terms of, if we see a cell phone visit or patronize a store, how many other distinct cell phones patronized that same store on the same day?
And thinking about the potential spread of the disease in terms of being in close physical proximity to other individuals, this again kind of captures how much social contact is happening in a city or a county. What we’ve seen in these data, of course, and this has been documented in a variety of different ways, is a huge fall off in movement and a huge fall off in social contact.
By the end of March, heading into early April, there have been really big drops in both our Location Exposure Index and in the Device Exposure Index. And those drops are obvious if you look at the national average, but what we did was publish these data publicly so that people can start exploring the variation in terms of changes in social contact and changes in movement in different places in the United States with different potential timing.
For example, our indices allow you to investigate whether stay-at-home orders or shelter-in-place policies introduced by state or local governments caused changes in behavior in the places that were passing those policies compared to behavior in other parts of the United States.
What we see in our Location Exposure Index is that counties or states that are really far apart have dropped to very little travel. In our smartphone-movement data, travel between Hawaii and the mainland United States, for example, has fallen to very, very low levels. You could have also corroborated that with more traditional data sources, like the Transportation Security Administration publishes daily checkpoint totals of how many people pass through an airport. And that series is very correlated with our movement data, revealing people that would be flying to Alaska or Hawaii, say.
In addition to that decline in movement between very far away places, you can zoom in and look at a level of detail that you wouldn’t be able to see in a traditional data source. Say, what fraction of people who live in New Jersey and work in New York City are still going into the city every day versus staying in a residential location on the other side of the river.
So the smartphone data can be corroborated with traditional data sources in some way, but then pushed much farther in terms of providing us greater detail. And so that is potentially really valuable for thinking about a variety of questions in the context of the pandemic. Other researchers have used our data set to investigate whether shelter-in-place orders were in fact effective at reducing movement, or whether if we wanted to announce travel restrictions, what the optimal arrangement in terms of defining mobility zones that potentially restrict travel, what those would look like.