-
These columns are relatively new and experimental for SafeGraph, and there is reason to expect these columns to under-report as a fraction of device_count. We generally do not collect data for devices throughout the day, so only a subset of devices have enough data to be eligible for columns like full time work and delivery behavior. However device_count has no requirement for sufficient data, so there is a selection bias for devices to be included in the *_behavior_devices columns
-
We expect that these counts are directionally correct, and valid for looking at trends over time, but not as reliable for estimating exact percentages as a fraction of device_count