Case Study: Tracking Human Mobility with Mobile Data Amid COVID-19 Lockdown
Did Parisians change their behavior when social distancing regulations were implemented? Here's how one organization conducted detailed research using data on human mobility to check the accuracy of anecdotal news media reports.
- By Artem Berehovyi
- August 12, 2020
In the midst of the COVID-19 pandemic, as countries one by one started closing down in response to health threats, international media turned the spotlight on Paris. They reported that the first few days after the introduction of social distancing measures, Parisians were reluctant to change their lifestyle and sacrifice the everyday joys of life. Large crowds were spotted in the streets going about their business without having to explain the necessity of being out.
Media reports are oftentimes based on anecdotal episodes which may not always reflect a real situation. There was a chance some people were dissatisfied with the lockdown measures and even some of them decided to ignore them, but we believe that such claims should be backed by actual data. This motivated us to conduct more detailed research using data on human mobility.
To monitor the shift in the Parisians' behavior during the lockdown, Aspectum, a business intelligence company, together with its partner Predicio, a location-based behavior intelligence company, conducted an analysis of weekly patterns during working periods. The analysis revealed the changes in citizens' movements within Paris exceeding 1000 m (ca. 3,281 ft) from March, 6 to March, 26. The time frame was broken into three weeks: 3/6 - 3/12, 3/13 - 3/20, and 3/21 - 3/26. The mobility changes were tracked using location data from mobile devices. Together with visualizations of the findings, the Aspectum map helped define patterns and clusters of high and low values as well as outliers.
Tracking Major Behavior Changes
Thus during week 1, there was a slight change in the number of Parisians' journeys longer than 1000 m. The only change we discovered from the data was that people migrated from the city center to residential areas. However, during week 2, the activity of Parisians outdoors dropped significantly — by 70 percent.
Also, the researchers examined changes in the crowding of some of the most popular places, namely the famous tourist attraction Île de la Cité, one the most popular shopping malls Galeries Lafayette Haussmann, Gare du Nord train station, and Parc des Buttes-Chaumont, a recreation area within a residential district. This survey was conducted with areas of interest (AOI) analysis using the points' locations in specific areas that allowed researchers to uncover local movement.
The analysis revealed some drastic changes in the activity within the aforementioned places in the middle of March:
- Île de la Cité: activity decreased by 60 to 75 percent compared to the previous week
- Galeries Lafayette Haussmann: activity significantly dropped by 95 percent
- Gare du Nord: activity decreased by 70 to 80 percent compared to the previous week
- Parc des Buttes-Chaumont: due to its being a recreation area within a residential district, the movement outside the perimeters of the park soared by 200 percent up to 500 percent, an increase due to people moving about in the residential area bordering the park
Data Gathering and Analysis
The survey was built on mobile location data provided in an aggregated, raw format by Predicio with the help a software development kit (SDK), installed in apps that shows a user's mobile location. The company sourced GDPR and CCPA-compliant mobile location data directly from SDK integrations with mobile apps. It was gathered from a large variety of differently themed apps with the average accuracy 14 meters indicating that the analysis covered a diverse group to ensure the reliability and representativeness of the findings, yet leaving a certain margin of error.
This research illustrates the Exploratory Data Analysis to analyze data sets and summarize their main characteristics, often with visual methods, not the modeling based on the ML techniques such as supervised, unsupervised, or reinforcement learning. When performing this analysis, it was extremely important to establish the utmost accuracy and precision. That is why researchers drew on the following statements:
- There is always a certain level of error up to 14 m for the GPS location, especially in the urban environment.
- The location points for movement data are not equal in terms of time or distance. For instance, when a car is on the road, every two minutes a segmented line is created with each segment's length of 1000 m. But when a person is walking, a line is created every minute with each segment being 50 m long.
- The data includes a variety of outliers in moving distances, which creates the need for preprocessing.
To make sure the analysis would provide reliable results, our research team addressed these issues while performing calculations in the following manner:
- We used segments intersections for calculating mobility changes instead of points' locations
- We cleaned outliers to eliminate their influence on results
In addition, the volumes of data played a key role in data verification. Thanks to our partners, we had a chance to analyze tens of millions of data points. Of course, it was impossible to be sure each point was precise and accurate, but large volumes of data allowed us to dismiss this imprecision. With each data point, horizontal accuracy given by the device is delivered and the precision of every single data point is verified.
Even though such extensive data enhanced the accuracy of complex analytics, the analysis did not require special training. All researchers needed to transform location data into insights were strong spatial thinking, creativity, and basic statistical knowledge for "reading the data."
Also, such hard skills as knowledge of SQL and experience with core GIS software were involved in the research.
Making the Most of SQL
SQL skills allowed researchers to execute analyses significantly faster with more flexibility. PostGIS Spatial extension in PostgreSQL became the key component of all experiments involving location data in volumes as big as millions of rows. PostGIS extends capabilities of PostgreSQL to increase its management capabilities by adding geospatial types and functions to enhance spatial data handled within a relational database structure. It adds spatial functions such as distance, area, union, intersection, and specialty geometry data types to the database. That's the primary reason Aspectum provides a SQL interface with the support of PostGIS functions.
In turn, core GIS software allowed proving concepts and visualizing the results. However, knowledge of GIS software is not essential for working with Aspectum. Data specialists did not need Python or JavaScript coding skills to display results. Aspectum provides visualization features in its user interface. As a result, it doesn't take a team of GIS specialists to conduct such research. A typical business analyst has all the needed skills.
Data Indicates Social Distancing Objective Met
The findings of the research showed that even though at the beginning the majority of Paris' citizens struggled to readjust to the confined lifestyle, with time the social distancing behavior prevailed in the city. Data proved that the French authorities achieved their objective, as Parisians started to avoid moving around the city and rely on social distancing as the most effective way to stem the spread of the virus.