Where are people staying at home?
New mobility data highlights where people are working outside their homes during the pandemic
By Andres Crucetta, Dylan Halpern, Qinyun Lin and edits by Susan Paykin
In the context of COVID-19 and other infectious disease pandemics, where we go and who we interact with matters a great deal. Mobility data can help us better understand, examine, and quantify these trends. In simplest terms, mobility data captures people’s movements. While understanding each person’s movement would be especially helpful, this kind of information is not readily accessible and also raises important privacy, security, and ethical concerns. However, such data is available at aggregated levels (e.g., a Census block group, county, or state). This aggregated mobility data can be informative, showing, for example, how communities respond to varied local, COVID-19-related health guidelines (e.g., stay-at-home order and mask mandate).
In this week’s post, we provide an overview of several public mobility datasets, with a particular focus on Safegraph’s social distancing data, which is now included and available for exploration on the US COVID Atlas.
Safegraph Social Distancing Data
We’re excited to share that Safegraph mobility data has recently been added to the Atlas. Safegraph is one of the leading companies with location and foot traffic data. As part of their COVID relief efforts, they released a dataset with Social Distancing Metrics. It works by capturing the GPS location of anonymous mobile devices. They first determine the nighttime location of a person and then track how far they have moved from home and what type of mobility patterns they exhibit. Some interesting metrics provided include include the number of people who are working part-time and full-time, the number of people who could be delivery drivers, and the number of people who’re dwelling at home during the day. Safegraph makes the information available at the Census block group level, which we aggregated up to the county level for inclusion in the Atlas.
The Atlas now features three main mobility variables to explore. In addition to aggregating up from block level to county level, we removed weekends and holidays from the data to highlight work behaviors and activity, providing a clearer view of how the pandemic has changed mobility.
- Percent Full Time on Workdays: This variable measures how many people are working at a workplace full-time (6–8 hours) in each county.
- Percent Part Time on Workdays: This variable measures how many people are working at a workplace part-time (3–6 hours) in each county.
- Percent at Home: This variable measures how many people are staying at home for the full day in each county.
Preliminary exploration of this mobility data highlights a shift in activity happening sometime in mid-August. Before this point, we generally observe more intense lockdown behavior: fewer devices exhibiting part-time and full-time work behaviors and more at home. After August, and through today, we see more devices active across work categories and fewer at home.
Importantly, the differences between communities’ travel patterns reflect a number of conditions, including different guidelines, communities’ varied compliance to their local guidelines, and different workforce demographics. In terms of the last point, for example, some communities have more essential workers than others. To help better interpret the travel patterns in different places, we also include the percent of essential workers to the Atlas (see Figure 6), along with the Safegraph social mobility data.
Note: We use the definition of essential workers from CMAP Illinois, a regional planning organization for the state of Illinois, which considers essential workers professions such as community and social services, healthcare workers, food services, construction, and production.
Exposure Indices from PlaceIQ Movement Data
The PlaceIQ Exposure Indices were born as a research project across several U.S. research institutions. This group collected location data from PlaceIQ and computed a series of indices that track how exposed were phone users to each other when visiting a particular venue. Outdoor recreational areas, places of worship, airports, and universities were excluded from the venue category. The most common venues visited were restaurants.
Some interesting indices presented by this dataset include the Device Expose Index (DEX), and Location Exposure Index (LEX). The county-level DEX, for example, captures, for an average phone device residing in a given county, how many other devices also visited any of the commercial venues visited by that device, within a single day. The county-level LEX captures what fraction of phone devices that pinged within a given county today, have been active at least once, in each other county within the past 14 days. Note that the daily county-level LEX is an approximate 2000-by-2000 matrix, describing the share of people in a given county who have been in other counties during the prior two weeks.
Each index has its own focus: the county-level LEX measures movement between counties, while the county-level DEX measures the average exposure of devices to each other within commercial venues, for each county. Figure 1 visualizes the county-level DEX on November 30th, where a darker color means a person may face more exposure when visiting a commercial location such as a grocery store or a mall.
Google Mobility Report
The Google Mobility Report shows movement trends by region across different categories of places. This includes includes Parks & Recreation, Workplace, Residential, and Transportation.
Google captured the location data of users who willingly turned on their location history over Jan 3, 2020, until now. With this data, Google set a pre-COVID 6-week baseline in which they tracked how users would have moved without the pandemic. This includes regular commutes to work, time spent at home, and visits to parks. It then compared this baseline with the constantly updating data from the COVID-19 pandemic.
This dataset is one of the most interesting ones because of its wide coverage and the fact that it captures key categories of mobility such as workplace and residential movements. As seen in Figure 2, in February there are hardly any changes in workplace movement compared to January. In Figure 3, we start to see drastic changes from the baseline with an up to 76% reduction in movement compared with January 2020.
Mapbox Movement Data
The Mapbox Movement data focuses primarily on device movement through space at a much finer scale than the other datasets discussed here. The data is available at down to roughly a block-scaled tile, internationally, so it potentially provides very detailed information on where and when people are moving. Mapbox collects this data as part of their licensing for applications utilizing their software development kit for making interactive maps. Trip data are collected with a random segment removed from the start and end of the journey for privacy protection, and monthly and daily activity indices are generated based on total trips through a given geography. This dataset has broad coverage and a genuinely robust sample size, but we decided that the metrics of the final dataset from Safegraph discussed below better suited our needs.
Patterns of mobility in the pandemic raise more questions than they answer — but that’s a good thing.
The many factors influencing mobility and decisions are core to understanding the community impacts of COVID — from state unemployment policies to school openings, from mask orders to transit access. Understanding these patterns helps us to unpack more local context of the pandemic and learn how, where, and when it unfolded.
For more information on the COVID-19 pandemic and to explore the Safegraph and Essential Workers maps, check out USCovidAtlas.org
