Source-linked AI summary
Mobility Changes in Response to COVID-19
Michael S. Warren, Samuel W. Skillman
TL;DR
The paper addresses the need to measure how COVID-19 interventions and voluntary behavior changes affect human mobility. It uses anonymized or de-identified mobile-device locations and mobility statistics to examine changes across geographic areas. Large mobility reductions are detected in the United States and globally, and state- and county-level data are released openly for further analysis.
Problem
Measuring mobility changes is important for assessing COVID-19 containment strategies and forecasting disease spread.
Method
The paper computes population mobility statistics from mobile-device location reports, including filtered max-distance mobility and normalized regional baselines.
Results
Large mobility reductions are detected in the United States and globally, including drops to less than 20% of normal in California, Illinois, New York, and Washington.
Takeaways & Limitations
Freely available state- and county-level mobility data can be combined with pandemic growth rates to improve models of interventions.
Takeaways & Limitations
The analysis assumes sampled devices fairly represent population behavior, but vendor access changes or reporting patterns can skew mobility statistics.
Abstract
from arXiv · showhide
In response to the COVID-19 pandemic, both voluntary changes in behavior and administrative restrictions on human interactions have occurred. These actions are intended to reduce the transmission rate of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). We use anonymized and/or de-identified mobile device locations to measure mobility, a statistic representing the distance a typical member of a given population moves in a day. Results indicate that a large reduction in mobility has taken place, both in the US and globally. In the United States, large mobility reductions have been detected associated with the onset of the COVID-19 threat and specific government directives. Mobility data at the US admin1 (state) and admin2 (county) level have been made freely available under a Creative Commons Attribution (CC BY 4.0) license via the GitHub repository https://github.com/descarteslabs/DL-COVID-19/
I. INTRODUCTION
COVID-19 interventions aim to reduce contact between infected and noninfected people, and their effects may be measurable through aggregate daily mobility. Mobile-device geolocation offers a way to sample movement, but privacy concerns remain even after anonymization.
- COVID-19 responses include restrictions and closures intended to reduce contact and disease transmission.These measures include travel and work restrictions, quarantines, curfews, event cancellations, and facility closures.
- Aggregate geospatial mobility statistics may measure the effects of interventions on people’s daily movement.The paper identifies mobility measurement as an input for forecasting disease spread and assessing containment effectiveness.
- Smartphone and mobile-device geolocation reports provide a mechanism for sampling individual movement.
- Anonymization replaces true identities with random identifiers, but location traces still create privacy risks through observations of whereabouts.
II. METHOD
The method processes large-scale mobile-device location data into daily, locally timed mobility statistics, aggregates them geographically, and publishes results in streaming-friendly formats while addressing sampling bias.
- Over 50 TB of mobile-device location data were analyzed using cloud computing and about 50,000 CPU hours.
- Reports are filtered by position accuracy, converted to local time, collated by device, and restricted to nodes with sufficient daily coverage.The accuracy threshold is 50 meters; nodes with fewer than 10 reports or under 8 hours of daily reporting are removed.
- Daily mobility is characterized using maximum-distance, bounding-box, and convex-hull measures computed from device positions.Maximum-distance mobility uses Haversine distance from the day's initial location, while the other measures convert reported areas into equivalent linear distances.
- Processed devices are reverse geocoded to countries and administrative regions, enabling statistics for areas such as US states and counties.
- Reported results focus on m50, the median daily maximum-distance mobility, and normalize it against a region-specific earlier weekday baseline.The baseline is the median weekday m50 from 2020-02-17 through 2020-03-07 for the US analysis; the normalized index is dimensionless and can be converted to percentage change.
- Outputs are stored as NDJSON and CSV so large datasets can be processed incrementally or used with geographic information tools.NDJSON supports one-record-at-a-time processing, unlike a whole-file GeoJSON FeatureCollection parse.
- The statistics assume device sampling represents population behavior, but vendor access changes or reporting correlations can produce systematic mobility errors.The analysis mitigates these risks by using multiple datasets and independent validation observations.
III. RESULTS
Mobility reductions appeared internationally and across US states and counties as COVID-19 spread and restrictions took effect. The results also show local disruptions and contextual variation in mobility patterns.
- US states: March 13–16: Mobility reductions were detected across all six selected US states.The paper reports higher normal m50 values in rural Texas than urban New York during the first week of March.
- US counties: 12:01am March 17: Mobility initially declined in Alameda, Contra Costa, and Santa Clara counties after the six-county Bay Area shelter-in-place order took effect.The order expanded statewide on the evening of March 19.
- US states: 5.2 km to 31 meters: New York’s median maximum-distance mobility fell between March 2 and March 23.The paper interprets this as half of New York individuals spending most of the day within 100 feet of their initial position.
- US states: Less than 20% of normal: California, Illinois, New York, and Washington reached this normalized mobility level, while Florida and Texas fell to about 30%.The mobility index is normalized to 100 using the February 17–March 7 baseline period.
- US counties: March 13: Champaign County experienced a mobility spike believed associated with students leaving after University of Illinois on-campus classes were cancelled.A similar student-related jump was observed in Monongalia County after West Virginia University activity changed.
IV. CONCLUSION
The study detects dramatic COVID-19-related mobility changes in the US and globally, and provides US state- and county-level mobility data for continued analysis.
- Mobility changed dramatically in the US and globally during the COVID-19 pandemic.
- Mobility data are freely available at US state and county levels under a CC BY 4.0 license.
- The largest counties in New York are examined through changes in their mobility index.
- A March 13 mobility jump in Monongalia County is associated with students leaving West Virginia University.