Product updates, industry-leading insights, and more
A Network Analysis of COVID-19’s Impact on Mobility Patterns
May 01, 2021 by
In this seminar, Mohsen Bahrami shares research conducted by him and his team at MIT Connection Science (Hasan Alp Boz, Aaron Nicholas, Nina Mazar, Burcin Bozkaya, Selim Balcisoy, and Alex Pentland).
Their study explores how different neighborhoods are disproportionately affected by COVID-19 through a network analysis perspective. To do this, they examine census block group data from New York City, breaking it down by borough. Using Placekey location data and Safegraph foot traffic data, they explore how census block groups (CBGs) intersect with points of interest (POIs) and each other.
We summarize the seminar below so you can reference the main takeaways. We still think you should go watch it though!
Understanding how COVID-19 impacts mobility patterns: a network analysis perspective
To understand mobility patterns at a neighborhood level, Bahrami and his team investigate network structure and centrality metrics. By studying how CBGs interact with POIs and each other, they found mobility patterns were impacted by socio-demographic features of a neighborhood, geographic location, and access to POIs.
While other studies have proven that different neighborhoods are disproportionately affected by the pandemic based on socio-demographic features, they wanted to examine how networks interact to further understand COVID’s impact.
They combine Placekey POI identifiers, Safegraph mobility data, US census data, and NYC open datasets to analyze this behavior. This data allows them to analyze how different census block groups (CBGs) interact - and intersect - with each other. This behavior can help in identifying the spread of infection, as well as understanding the impact it has on different groups.
Their findings confirm the disproportionate impact of the pandemic, and show that changes to mobility behavior depend not only on socio-demographic features of the residents in a neighborhood, but also on geographic location and access to POIs.
6 key takeaways from this seminar on the pandemic’s impact on mobility patterns
While this doesn’t cover the entire scope of the seminar (psst… go watch it above!), here are the five main takeaways:
1. COVID-19 affects neighborhoods differently
Time in seminar video: 2:38
By nature, mobility is a critical factor involved in COVID-19 based on the need to limit the infection and spread. Mobility is directly impacted for those that contract COVID-19 as well as the general public, due to social distancing guidelines and store shutdowns. This has a subsequent effect on the economy.
Bahrami and his team understand that networks are the backbone of the economy, which is why they chose to analyze how different networks affected (and were affected by) the dynamics of mobility patterns. This can then be used to predict future economic patterns and mobility behavior.
For their study, they chose to analyze New York City specifically, as it’s a large, densely populated area, and represents a diverse range of demographics. The main comparisons they examined were deciles of university-level education, income, and minority populations.
2. COVID-19 restricts how census blocks groups interact with each other
Time in seminar video: 5:50
To analyze this relationship through a network perspective, they needed to develop two types of networks:
1. POI-CBG bipartite network: to study the change from a point of interest (POI) perspective.
2. CBG-CBG network: using the POI network, they create a system that analyzes visits at each POI, allowing analysis of where and how frequently CBG groups intersect.
When we analyze betweenness (which measures the importance of a node in a network), we see that education and income show very similar trends. As you can see, when the pandemic hit, these higher education and higher income groups play a less significant role in the network.
Over time, we have seen these balance out, but not return to pre-pandemic levels. This shows that groups encountered similar trends in how they were impacted by mobility, but higher education exhibited less mobility, and had a lower impact on the network (by causing less spread).
3. Heavy disparity of mobility patterns is seen between different groups
Time in seminar video: 11:34
When examining networks pre-COVID and post-COVID, we see that the ego-net dissimilarity grew significantly when the pandemic first happened (indicating that behavior changed drastically). Then we see the levels balancing out, but still remaining disparate in comparison to pre-COVID numbers.
We see that both top and bottom income deciles responded similarly to the outbreak. However, we also see that the top income and bottom income groups go back to normal differently. Lower income groups (presumably due to the need to work, inability to work from home, inability to take time off, and other economic pressures) go back to normal mobility behavior faster.
Overall, the pandemic outbreak quickly and significantly impacted mobility behavior, while the process of returning to normal took significantly longer.
4. Digging into demographics data: higher income = less mobility
Time in seminar video: 13:30
Now that we know these dissimilarities exist, we want to examine them closer to determine where they are, and explore their demographic distribution for deeper insights.
The closest relationships show that low income groups disproportionately maintained their mobility patterns from before the pandemic hit. This low income decile also had the highest rates of infection. Higher education and income consistently had the lowest rates of mobility and infection, but this in part is likely due to their ability to work from home and avoid public transit.
5. Predicting census block groups that could be spreaders
Time in seminar video: 18:19
They can then take this information and predict potential spreader census block groups (CBGs) based on the frequency that different groups interact.
First, they find the CBGs that are in the top COVID cases quartile (accounting for a lag between when patients contract COVID and are actually admitted). Second, they find the neighbors of those CBGs in the mobility network (going back the two weeks, based on when they would actually have contracted COVID). Third, they analyze the frequency of visits to various POIs along with demographic distribution data.
Ultimately, this data lets them analyze the behavior of various census block groups, as well as their behavior in relation to each other and the POIs around them. This information can inform future policy and decision-making to help better support the neighborhoods most affected and help institutions prepare properly.
6. Staten Island: understanding an outlier
Time in seminar video: 24:44
Staten Island was a statistical outlier, with higher CBG interaction than other areas, and more outside travel. This didn’t align with the other data, showing that higher educated and higher income groups generally showed less interaction.
This could be in part because Staten Island is geographically isolated. However, their findings showed that there was an extremely low number of POIs per person, indicating that there may not be enough POIs within the area, forcing the local population to travel elsewhere to get access to goods and services.
Mohsen Bahrami’s backstory & research
Mohsen Bahrami is a Postdoctoral Associate at MIT Connection Science, where he specializes in computational science. His research areas have included data-driven behavioral analytics with the aim of predicting individual and group behavior using large-scale datasets from a variety of industries. With a recent focus on how human mobility, social interactions, and peer influence affect health and economic outcomes, he is a perfect fit for this study.
How Placekey enabled research on the impact COVID-19 had on mobility patterns
Placekey’s universal identifier capable of mapping POIs provided information about the quantity, density, and distribution of POIs in specific boroughs of New York City. Safegraph’s foot traffic data provided information about the location and mobility of devices within the areas of the census block groups.
Using these solutions together helped Bahrami’s team understand the POIs where census block groups intersected the most. This data can be used to understand - and predict - mobility behavior related to POIs. Pairing this with NYC census data allowed them to leverage this data for deeper insights.
To learn how you can benefit from location and foot traffic datasets made possible by Placekey and Safegraph, join the Placekey community and talk to others just like you!
Get ready to unlock new insights on physical places