Using Facebook data to understand changing mobility patterns

Throughout the coronavirus (COVID-19) pandemic the Data Science Campus has been exploring new data sources that can provide insights into mobility. Previous blog posts have covered our use of aggregated mobile phone location data and the Google COVID-19 Community Mobility Reports.

Facebook Data for Good is another source of mobility data that we have been investigating and initial insights from our work are included in the ONS publication “Comparing behaviours and economic activity during lockdown periods” published on 19 March 2021. The Facebook data provide timely insights that can rapidly help us understand patterns of change in how people are moving around the country and the impact of the lockdown restrictions. The data we receive is aggregated, de-identified and re-scaled, so that no individual or individual business can be identified.

The data is provided by the Facebook Data for Good programme to WorldPop at the University of Southampton. WorldPop analysts carry out aggregation, re-scaling and analysis of the data that they receive via the COVID-19 Mobility Data Network. These data have already been used in an experimental analysis on monitoring the population density changes in coastal towns.

We receive the aggregated, de-identified and re-scaled data on Facebook app users in the UK, who have location services activated on their smart phones. It provides a daily time series, starting from 10 March 2020.

The location information uses the Bing Maps Tile System to allocate users to tiles within a regular grid across the UK. Based on the tile that a user spends the most time in during an eight-hour time period, the data include measures of:

  • population density – the relative change in the number of users per tile
  • population flow – the relative movement of users from one tile to another

We have been experimenting with the best way to aggregate the tiles to administrative geographies such as local authorities, which we can link with other types of demographical data published by the ONS. This has enabled us to create a dataset of the changing relative numbers of journeys between local authorities.

The Facebook data offers some different insights from some of the other data sources we have. This includes higher spatial resolution and eight-hour frequency. We also want to explore these data alongside our other mobility data sources, because we know that the different data sources will have different strengths, weakness and biases, and using a range of indicators helps us to triangulate the various sources. We want to fully understand these differences and are continuing to explore how the Facebook mobility data can be used to complement our other sources.

The coronavirus pandemic has resulted in rapidly changing patterns of population movements and new data sources are needed to understand these changes. The Facebook data show patterns of the reduction of movement following changes in policy, such as the introduction of the tier system and the national lockdowns in England, in a more timely way than more robust traditional data sources.

All data provided to the ONS are strictly de-identified and in an aggregated and re-scaled form so that no individual can be identified in any of our work, and only relative changes are analysed. We also carried out an ethical review to ensure that the use of these data was appropriate.

We are grateful to Facebook Data for Good and the WorldPop team at the University of Southampton for providing the data and helping us understand and analyse it. We are exploring what further insights we can make available and hope to publish these soon.

Cathy Atkinson, Senior Data Scientist