Has anyone tried using a more stable subset of devices to produce these patterns?

Hi all, sorry if this has been discussed before (did a search and did not find anything). We are using SafeGraph’s POI data in conjunction with Veraset mobility data to create a pattern-like dataset for Canada. What I’m seeing across different regions is that large fluctuations in panel’s composition seem to coincide with peaks and valleys in the daily POI visit patterns (even after normalizing with the daily device counts of each region). Has anyone tried using a more stable subset of devices to produce these patterns? e.g. only using devices that are present at least for a minimum number of days per month etc… This might result in some sampling bias favoring more mobile devices or apps with more frequent updates but I’m wondering if it could alleviate some of the fluctuations.

Hi Amir, we’re in the process of uploading Canada weekly patterns. I’ll post an update on the channel once it’s ready :smile:

Awesome, thanks! Can you elaborate on availability of historical data for Canada’s weekly patterns? how far back would this go?

We’ll share Nov and Dec for the years 2019 and 2020

Changing the sample used to one that’s only of more consistently-seen devices would require a change at a pretty low level - as far as I know it hasn’t been tried. You might see if you can get anything by using visits from the visitors_home_cbg column - this only includes visits from devices for which the home location can be determined, which requires visibility over a longer period of time, and links more closely with the number_devices_residing normalization variable, if that’s what you’re doing

Thanks @Nick_H-K_Seattle_University, currently there is no pattern data for Canada (looks like we will have November and December data soon however most models would need data back to January 2020). I am deriving monthly home locations (at cbg level) from raw GPS data and using the number of daily devices “residing” in a cbg (we call them DA in Canada) to normalize.

Amir, we just uploaded 6 weeks for 2020. We planned to release 2019, but we caught some irregularities and chose to stick with 2020.

Could you please take a look and give me some feedback (if any) so we can release more data soon? :smile:

Thanks @Francisco_Utrera, I will have a look right away.

cc @Parker_Ence_Veraset