I’ve read the docs on normalization and correcting sampling bias, but are there different approaches necessary for different elements of the social distancing data?

Like many, I’m interested in identifying areas where potentially risky behaviors (ie, more physical contact with more people) are increasing–say within a single state. I’ve read the docs on normalization and correcting sampling bias, but are there different approaches necessary for different elements of the social distancing data? eg, completely_home_device_count versus full_time_work_behavior_devices versus destination_cbgs–these are very different kinds of measures built from SafeGraph data, and I imagine they have different levels of sensitivity to underlying changes in the panel