Using cell phone location data to explore Philadelphia mobility patterns throughout the pandemic

1 Introduction

As the coronavirus spread in Philadelphia this year, mobility patterns changed. With this shock came fewer commutes to work; we changed where we shopped, where we dined and how we traveled. In order to understand the consequences of these changes, we use GPS data from mobile phones to track travel patterns before and during the pandemic.

To do so, we collect data from SafeGraph, a provider of mobility data from iPhone and Android smartphones. Note that SafeGraph gathers data on a representative sample (10%) of the population across the country, so our indicators are not the true number of visits or journeys, but a sample. The data model is explained in Figure 1.1. The number of visitors is the count of devices arriving at a point of interest (POI) while a connection is an origin-destination line between a Census Block Group and a point of interest. A flow is also a connection, but with a weight measuring the number of visitors traveling between origin and destination.

1.1 Key definitions

Each visit is a mobile device entering into a point of interest; these include parks and museums, restaurants and bars, or offices and hospitals. In Figure 1.2, the distribution of these venues and businesses are mapped. Each point of interest is classified by its type, which SafeGraph provides. ¹ We can see that most businesses cluster in Center City or nearby but no businesses cluster more than leisure—which is restaurants and bars.

1.2 Distribution of points of interest

To demonstrate how connections form a mobility network throughout Philadelphia, Figure 1.3 connects origins to destinations by month.

1.3 Connections over time

This analysis comprises different spatial scales: Citywide, Neighborhood and Point of Interest. We can look globally, across the city, to explore trends throughout; we can also think locally, dividing the city up into cells or neighborhoods to probe variations within the city. Finally, we can look at individual businesses or venues. Below, we attempt to understand patterns at each scale in order to understand how mobility has shifted since the onset of the pandemic.

2 How have visitations changed?

In this section, we explore trends in visits, defined as the count of devices traveling to a point of interest or area, beginning with the city as a whole. To analyze the business environment for chain stores across the city, Figure 2.1 ranks stores by the number of visitors they received and animates this change throughout the pandemic. Visits to dollar stores rise gradually throughout the year; another important shift is away from non-essential retail towards essential businesses like pharmacies. Starbucks and Wawa occupy top spots for the first several weeks of the year but when the shelter-in-place order occurs, visits fall and they are replaced by essentials like RiteAid and ShopRite.

2.1 Comparing foot traffic by brand

In Figure 2.2 we aggregate visits by industry, grouping by classes like leisure (restaurants and bars) and tourism (museums and theaters). The pandemic has curbed visits to each class of business, but hit particularly hard is leisure and “other”, which includes offices. Tourism is regaining visitors while shopping and grocers are not, perhaps as many switch to digital commerce.

2.2 Industry trends by category

Next, we look at how mobility varies across smaller geographic units to determine whether or not the pandemic is impacting some parts of the city more than others. We find large disparities between the best- and worst- performing neighborhoods. Figure 2.3 explores trends across neighborhoods; those dominated by office work, like the Navy Yard along with Logan Square and Center City, saw precipitous declines in visits, but neighborhoods with strong amenities and residential communities have recovered. This suggests that economic activity may have shifted away from Center City.

2.3 Neighborhood trends

Best	Worst

Figure 2.4 presents an alternative approach for visualizing neighborhood visitations, plotting the rolling average of visits for each neighorhood (in red) compared with the citywide mean trend (dotted lines). Many neighborhoods in the Northeast and Northwest are rebounding while University City, Center City and Old City are still down substantially.

2.4 Rolling averages

With visits data at each point of interest, it is possible to explore these dynamics at smaller spatial scales. Figure 2.5 aggregates visits to 500 meter squared grid cells and visualizes them for each month between January and August. Visits fell in most cells during the worst months of the pandemic, but the business district has regained visitors each month.

2.5 Visits by month in gridded units

Businesses are not evenly distributed across the city, however, so understanding business activity requires a unit of analysis that respects commercial corridors—zones where businesses cluster together—of which the city has designated 279. We look at visits to restaurants and bars within commercial corridors below. The largest corridors are Market West and Market East, on either side of City Hall (boxed on the map), with 1712 and 1263 restaurants and bars respectively, followed by Old City at 654 and another in University City with 493: most of the business activity is concentrated in a few locales.

2.6 Commercial corridors

Figure 2.6 maps percent change from January to August, while Figure 2.7 plots the trend in the top and bottom 10 of these corridors over time, the greatest reduction in visitors is in Center City and at the Sports Complex. Plazas like Oxford and Levick, home to a supermarket, and City and Haverford see the smallest impact.

2.7 Corridor trends

Best	Worst

3 How have connections change?

This section looks at connections to points of interest within the network to see how to see how interconnectedness is changing for different points in the City over time. Figure 3.1 replicates the animation from Figure 1.3 as a series of monthly maps. Again, the network changes dramatically as the pandemic sets in, but the decline in connections is most evident in Philadelphia’s central business district.

3.1 Connections by month

Here we take a closer look at some key network connections in Philadelphia. Figures 3.2 and 3.3 plot connections to the Comcast Center and Reading Terminal Market, respectively. Economic activity in these two key hubs decline dramatically during the worst months of the pandemic.

3.2 Comcast Center in focus

3.3 Reading Terminal in focus

These destinations are critical to Philadelphia’s office and tourist economies, but what about the brands that serve thousands on a daily basis? We can see the differential effect of the pandemic simply by looking at the change in visits to Target stores and Planet Fitness gyms. Both see similar visitation activity prior to the pandemic and they have a similar number of locations—12 Targets and 14 Planet Fitnesses. Connections to Target held steady throughout the pandemic, but the same cannot be said for Planet Fitness.

3.4 Connections to Target stores

3.5 Connections to Planet Fitness gyms

We can track changing connections for all brands in the city. Figure 3.6 shows the number of connections certain brands lost between January and August selecting the top and bottom 10. It shows that essential business and fast food restaurants saw comparably less of a decline. The map shows the locations of brands for context.

3.6 Relative brand connections

Rankings	Locations

4 Possible consequences

Might the pandemic further exacerbate Philadelphia’s deep socio-economic divide? If communities of color and low-income communities are disproportionately comprised of essential workers who cannot work from home, then Census tracts with higher rates of non-white residents should exhibit greater rates of outflows relative to communities with fewer minorities. Figures 4.1 and 4.2 test this outflow proposition using income and race, respectively.

4.1 Did trips change with income?

In March, at the onset of the pandemic, more Philadelphians were leaving their home tracts. In April, all communties see reductions in trips. However, the higher the median income, the greater the reduction in trips, though the correlation is weak. We plot the relationship between race and mobility in Figure 4.2. There is no correlation here in any month but March.

4.2 Did trips change with race?

5 Conclusion

The purpose of this report is to examine mobility patterns in Philadelphia before and during the pandemic using cell phone mobility data provided by SafeGraph. The visualizations built from these data show large declines in mobility at the onset of the pandemic. While we cannot access economic indicators for particular neighborhoods and businesses throughout the City, as a proxy, these mobility data suggest that many industries and corridors have experienced a tremendous loss in economic activity in recent months.

The next stage of this work is to develop some interactive, web-based visualizations that can help stakeholders in Philadelphia understand and plan for a return to ‘normal’ mobility patterns.

---

1. If that description contains “restaurant” or “bar”, we classify that as leisure. Anything educational, from tutoring to public, private or charter schools to tertiary education, we call that education. Tourism includes museums and parks.↩