Locating people in times of Covid-19
Finding a common understanding and an unbiased representation of Location Technologies and Location Data has never been easy. And when the subject hits the mainstream media, as location data is currently doing in relation to Covid-19, it can easily be misrepresented and cause confusion. Therefore, I wrote this article to try and clarify how mobile location data is collected and how it could be used in a Covid-19 context. Everything that follows is only my opinion, and I have nothing to sell.
People can be located in absolute terms (point on a map), and in relative terms (proximity to other people). For the absolute location there are 2 methods by which individuals can be located via their mobile phones. We can use a Smartphone App, or information from the mobile networks that serve them. Each method has fundamental advantages and disadvantages.
- Using Location Data from Smartphone Apps
The standard way to obtain a user’s location is for them to download an App and then configure the privacy settings to allow Location extraction and use. The App could be a modified and updated existing App, or a completely new App.
The only exception to this would be if the underlying Operating Systems (from Apple, Google) were themselves modified to support such location functionality. This may be happening as I write, given the news of a joint development by Apple/Google of a contact tracing function using Bluetooth, as described in section 6.
When looking for existing location data the most popular Apps would likely be Maps, followed by those for Weather, Dating, Public Transport, Fitness, Mobility, some Games, and not that many more. In the broad and densely populated world of mobile Apps, surprisingly few really use location at scale.
Apps use an identifier known as an Advertising ID that is unique to the smartphone and which would have the location data associated with it. It is not related to the SIM card holding your mobile phone number. As the name suggests this ID was developed for the Advertising industry and is used extensively to profile and target users of Apps.
2. Location accuracy, and the challenge of Scale
Phones and Apps generally use a hybrid location method, taking all received radio signals (GPS, Cellular, WiFi, etc.) and crunching them into a best guess of the device position. If there is a strong GPS signal resulting from a good view of these satellites, then sub-10m accuracy is possible. But when indoors, or in areas such as shopping centres, GPS signals are weaker and WiFi and Cellular signals would have to be used. Any location accuracy then achieved would be a function of WiFi and Cell density in that specific area, likely 50m-1000m or more.
With Smartphones there is a challenge to get location data at scale, as users have to actively and correctly download or update an App, and ensure their privacy settings allow them to be located. Scale is important because more users bring more certainty in more places to any analyses and conclusions. There are 2 workaround solutions to access lots of existing phone location data quickly, both of which have appeared in the Media recently. Google have a huge quantity of phone location data from their Maps App and other Services, and have made this available in anonymised form to health professionals. This data shows user location and time, and can show population movement (footfall) to determine the effect of lockdown measures and potentially analyze social distancing observation. The categories of places measured are those already used by Google and the Advertising industry for profiling and targeting. E.g. Parks, Retail and Recreational areas, Grocery stores and Pharmacies, Office areas, Commuter stations, Residential areas, and many more.
The other solution would be to get the data from the many companies in the Advertising industry which are already processing similarly huge amounts of phone location data. The data is generated by multiple Apps and each location is assigned to an Advertising ID and its associated Smartphone. This location data is aggregated together into bidstreams and sent to many parties in the advertising ecosystem. It is generally specific to a single country. Both the advertising industry and Google would have this location data in non-anonymised form, as they need the association to unique devices and users for their advertising. The data can be stored and therefore historical movement since the start of the year could be available.
3. Using Location Data from Cellular Networks
The challenge with Mobile Network location data is the opposite from Smartphone Apps, in that the scale from large numbers of users exists — but the location accuracy is not as good. Cells in a mobile network typically vary in size from 100m to many km. Whilst busy city centres can occasionally return 100m accuracy, as you leave these areas the accuracy degrades, quickly sliding out to 500m, and then beyond as we reach the suburbs. There are various ‘triangulation’ or multi-cell calculation methods described, based on signal strength or timing information from 3 or more cells, but they have generally not been implemented. This means the location accuracy is usually a direct function of the size of the cell that your phone is using at any given time.
Network location data is usually available for all subscribers belonging to a mobile network, and users in this network from other countries (roamers). The location is recorded against the user’s phone number. As there are usually more than one mobile networks in any country (typically 3) you would need to combine each of them to get close to a 100% population coverage. A further issue is that mobile network location data is usually anonymised and the user is neither known nor contactable. The current Covid-19 situation should justify lifting this anonymity under specific conditions, such as only the location data and an ID being exposed for a specific time period.
4. Simple Covid-19 use cases using location data
Both Smartphone App and Network location data can show absolute locations as points on a map, and therefore infer general mobility patterns of phone users. Simple Covid-19 use cases could include:
• Population movement and mobility in response to lockdown guidelines. The (anonymised) data can show whether people are generally out and about, at home, or elsewhere.
• Density of people in public and pre-defined places and potential risks of contagion
• Movement in and out of defined areas
• Communication and Alerts to those in specific areas at times of high risk
These are the basic applications that most mobile location data can support.
5. The return of Bluetooth
Smartphone Apps can also invoke and manage Bluetooth functionality, a technology with an interesting history. Although it successfully supports multiple applications today (headphones, speakers, smart watches, connected cars), Bluetooth was not so successful in Beacon form when acting as a location technology. This was partly due to the need to maintain and share a global reference database indicating precisely where all these Beacons were. As Beacons could appear and disappear easily, and some companies were reluctant to share their Beacon ID’s and locations, it never really worked as a location technology. However, what Bluetooth does have is a short-range proximity confirmation, whereby 2 devices are deemed in close proximity to each other if they can receive each other’s Bluetooth signal. This may have a use in a Covid-19 contact tracing context.
6. Proximity use case to identify contact with Covid-19 carriers
To take this a step further we can consider what could be done if users are tagged with a Covid-19 status, such as positive, negative, suspected, or similar. With such a status indicated we could see where Covid-19 carriers had been, where they are, and which other people they may have been in close proximity to.
The effectiveness in predicting exposure to Covid-19 carriers would be a function of the number of locatable users (more = better), and the granularity of the location data. Location granularity is normally a direct function of the underlying technologies used. Therefore good 10m-50m accuracy from an App is clearly much more usable than the 100m-1000m from a Cellular network. But sub-10m from Bluetooth may be the best solution of all, as this technology can operate on a proximity basis, where the absolute locations of the users are irrelevant and do not need to be known.
This option has been developed by an EU organisation known as the Pan-European Privacy-Preserving Proximity Tracing (PEPP-PT), who have turned their attention to Covid-19 and Bluetooth technology. Their solution regularly sends its ID out via Bluetooth and listens to the ID signals of other nearby users. If two users are within range of each other for a time period sufficient for a potential infection, they exchange their IDs and store them locally in encrypted form. If a user then falls ill with the Covid-19 virus, his smartphone subsequently forwards the encrypted ID’s of those users with whom he had contact, to a central system. The system could then warn the respective contact persons via smartphone notification of the possible infection and recommend preventive quarantine. For such an Application, scale is even more critical as with more users of this App more cases can be identified and dealt with, so it needs a large % of the population to quickly become users.
But this is now drifting beyond my knowledge. Hopefully this helps some understanding, and that there are professionals and experts now working on it. Best of luck.
Disclaimer. These are 100% my own views and in no way represent any company I have been associated with.
About the Author. I’ve spent the last 20 years working with Location Technologies, Location-Based Services, and Location Data. This has been from Product Management positions for companies operating across the Ecosystem, and in roles such as: Consumer Services Provider, B2B Advertising Solution Vendor, Location Technology Vendor, Navigation Solution Integrator, Mapmaker.
