Accounting for Sampling Bias in COVID-19 Mobility Data

Amanda Coston and Daniel E. Ho share their recent findings from an independent audit of potential sampling bias in SafeGraph mobility data: Mobility Data Used to Respond to COVID-19 Can Leave Out Older and Non-White People. They found that certain groups are underrepresented, and offer potential ways of addressing this bias in future datasets.

This study was made possible by SafeGraph, who’s team members have been transparent, helpful, and open to this research - as well as understanding sampling bias more generally.

‍

Sampling Bias: Older ages underrepresented in mobility data

This study was the first independent audit of demographic bias in SafeGraph’s smartphone-based mobility dataset. Researchers attempt to find unintended sampling bias and examine how COVID-19 high-risk groups are underrepresented.

Specifically, they examine how SafeGraph mobility data underrepresents older age groups, as they use smartphones significantly less than younger age groups. These people are missed - and therefore not properly represented - in SafeGraph data, which is tracking mobile devices.

How Placekey enabled research on sampling bias in mobility data

SafeGraph made this data available knowing that sampling bias is always something that is important to consider - and hard to avoid. Understanding the importance of accounting for sampling bias, SafeGraph welcomed this scrutiny to improve future datasets and help researchers use these datasets more effectively.

To learn more about how SafeGraph and Placekey are continuously improving and evolving how their datasets can be used, join the SafeGraph community and share ideas with others.