When it comes to watching people move, it’s a brave new world out there.
A new article in the Lancet journal Digital Health, co-authored by Direct Relief’s Vice President of Research and Analysis, Andrew Schroeder, talks about how Covid-19 has brought a surge of interest in population mobility data.
The data, collected from cell phones, can offer valuable insights into how people move, whether they’re adhering to shelter-in-place recommendations, and even how policymakers might (cautiously) work to safely roll back social distancing requirements.
Companies have increasingly made this data available to help in the fight against Covid-19.
But there are challenges, Schroeder said, ranging from the vastness of datasets to the comparatively few people qualified to analyze them.
Schroeder sat down with Direct Relief to discuss the new article, what can be learned (and not) from watching population movement, and how that knowledge can be used to help the world live more safely with Covid-19.
Direct Relief: How can we use data from cell phones to learn about social distancing? What does the data tell us?
Schroeder: I think the most important thing you can learn from looking at how people and their devices move around, is the rate at which people have stayed in one place or have moved to any number of other places over a certain length of time.
One example is what’s called co-location. Co-location is something calculated principally through Facebook, but it could be applied to any number of others. But it looks not only at whether you’re moving, but at the probability over a certain period of time that people from two different places spent a certain amount of time in the same place.
And that can tell you something interesting when it’s connected to, let’s say, a disease model, or even just descriptive information about the number of disease cases that have occurred and the kinds of risks that people might have. The more we add up that information, the more we can start to see these large-scale trends that are meaningful for epidemiology in near real-time patterns.
We’ve seen this now play out over months, where it’s become part of the basic, assumed landscape of information for how we understand events like these. This wasn’t really the case before Covid-19. It’s kind of a new world in that sense.
Direct Relief: So what changed? How did all of this become available?
Schroeder: Well, the main thing that changed was that Covid-19 was such a huge event, affecting so many people simultaneously, with such a large economic impact. At the very same time, you had a set of health policies being imposed basically all over the place, kind of at the same time – restrictions on mobility to reduce disease incidence.
There was a very high incentive for a lot of governments and other agencies to know what was going on. When we look back now on what happened over the last six to eight months, we can see that spike in demand very clearly.
And we can see that it’s associated with what I now think of as the Big Bang for digital human mobility data. All of a sudden, a huge number of providers that have collected this information for many years – the collection of it isn’t new – began to see that there was an immediately relevant use case for their data.
That’s one of the unusual things about the society we live in right now, which we’re kind of aware of – but I think probably not to the degree that we could be in terms of epidemiology – which is that many of us are being tracked at a very granular level all the time.
We should be wary of that, but we can also leverage that for really important things like controlling Covid-19 if we do it the right way.
Direct Relief: I’m just curious: How detailed does the information get? Do we keep track of people within feet, within blocks, within square miles? How granular do we get?
Schroeder: There’s really these two different wings of mobility data. One is cell phone data records and the other is GPS traces. Both of those are actually very granular when they’re in the hands of the companies that are tracking the data.
When it comes to a researcher, though, we don’t actually want to know about your point-specific location as you moved around. It’s not relevant for understanding population and health trends to understand that you were in the corner of A Street and B Avenue at a particular time. It’s important to sum it up into a unit that is analytically meaningful.
There are two kinds of units that are really meaningful. One is administrative, so the lowest level would probably be a census block. And the other would be like a grid square. Basically, if you overlay a grid onto a map or physical space, then you would just add up everyone into that grid.
The standard for this, which has really been set up by Facebook’s Data for Good program, is 600 meters on a side.
So pretty big, but small enough that when you start to see a lot of 600-meter blocks all together, and see how they vary over time, you actually get a pretty granular view.
The data itself that is being collected is quite specific inside that 600-meter block, but that’s not actually what gets transferred to researchers.
Direct Relief: What are some of the challenges of gathering and making sense of this kind of data?
Schroeder: There are many. Probably the biggest is the speed of my laptop.
These are huge datasets. Big data – I kind of hate the term – it’s usually defined by not only [the fact] that you have a lot of data, but that it changes frequently. It’s not enough for you to know just one slice of it. You need to know the next slice and the slice after that and the slice after that, and how they work connected together.
And the faster that those slices come at you, which is known as data velocity, the harder it is to manage it.
We’re co-directors of the Covid-19 Mobility Data Network, which is a group of infectious disease researchers, principally, and public health response agencies from all over the world who study this stuff and who use it for public health and crisis response.
And that in and of itself becomes complex because you have a bunch of different sources and a bunch of different policies and a bunch of different [data] agreements, all of which need to be managed in some way, or you can’t actually access the data at all.
This is actually one of the motivations for writing the [Lancet] paper in the first place, that data is collected in subtly different ways.
That becomes methodologically complex, and it limits the number of people that actually are qualified to do it. Which is a problem, because the number of problems that are addressable by the data is very large, but the number of people that are qualified to analyze the data, to solve the problems, is actually pretty small.
I think that’s an open question for the future: How do we increase the scale of expertise? How do we make it so that more people are more able to actually do valuable things, inside ethical parameters, with this kind of data?
Direct Relief: At the end of your Lancet article, you said that metrics could play a role in the rolling back of social distancing measures. Can you tell me a little bit about what that would look like if it happened?
Schroeder: Social distancing, physical distancing, was never intended to be in place for very long.
It was intended as a mechanism to buy time so that certain changes or advances that could be made that would benefit people over the longer term, while minimizing casualties to the virus. You have to relieve them.
We’ve seen this throughout the United States, where, to be quite blunt, it was done in an extremely poor fashion. A very large number of people have been killed as a result of extraordinarily poor policies for changing social distancing very fast.
But if we were to apply a very specific and granular lens, we could make much more specific decisions about how to close and open pieces of society, rather than closing all of society.
And [we could be] doing that in a way that’s mindful of the lived reality of these people, and their jobs, and their businesses, and their government agencies, and the day-to-day life that we were living prior to February of 2020.
If we got rid of this idea that there is actually going to be a day when the virus is gone and we go back to the way things were – we’re probably going to live with this virus for a reasonably long period of time.
If you wanted to live with that, and still conduct daily life, and not actually have hundreds of thousands to millions of casualties from it, which is an unimaginable humanitarian catastrophe already in some ways, you have to get very specific about how you deal with closing, opening, changing behaviors, and following whether or not there was behavior change.
I would say that we’re not good at it right now as a society and we need to get better at it because the virus is probably not going anywhere.