Hi everyone, we noticed that the num_devices_residing values in the home_panel_summary.csv file of Monthly Patterns differ slightly from the values of that variable in the home_panel_summary.csv located in the monthly Neighborhood Patterns.
We did calculations over calendar year 2019 and used “Monthly Places Patterns (aka “Patterns”) Dec 2020 - Present” for Monthly Patterns and “Neighbourhood Patterns” for the Neighborhood Patterns.
The discrepancy is small on average (Neighborhoods on average has about +34 more devices more per month, per Census block group). Some block group discrepancies are big though (+7305 devices). Just wanted to point it out and to see if someone can explain the discrepancy.
The discrepancies, we found, add up, though. In the monthly patterns version of home_panel_summary.csv, adding across all census block groups we find 32MM devices residing. In the neighborhood patterns version, the same calculation delivers 51MM devices. We calculated these numbers using the Sept 2019 month. Which is the most reliable file to use when we want an accurate count of number_devices_residing?
Hi @Alex_Zentefis_Yale_University, sorry for the delay here - I personally don’t know but after speaking with @Nick_H-K_Seattle_University, he mentioned it might be:
> both data sets only count devices observed making some sort of visit, which would be everyone in neighborhoods but only devices observed in a POI at some point in patterns
Perhaps @Lauren_Spiegel_SafeGraph could weigh in here and either confirm or deny the theory
the home panel summary is meant to be used with the product it is delivered with. so if you are analyzing Neighborhood Patterns, use the Neighborhood Patterns one. For Monthly Patterns, the monthly patterns one.
Just so that I have this correct: the 20 million more devices that we noticed in the Neighborhood Patterns over the Monthly Patterns seems like a big difference. Is this difference due to the recorded POIs in a neighborhood belonging to a much smaller set than the entire neighborhood itself, which makes it harder to count a device visiting a POI than visiting somewhere in the neighborhood in general? In other words, is @Nick_H-K_Seattle_University’s theory is correct?
With the Home Panel for Monthly Patterns, if a device that doesn’t visit a POI in the period is not listed as a device in the Home Panel won’t that mean that any normalisations with understate mobility drops? For example say the actual devices SafeGraph gathers data remained unchanged in 2020 but 20% of devices just don’t visit a POI in April. If I normalise visits by the num_devices_residing in the home_panel_summary, then I wouldn’t I be incorrectly inflating April visits by 20%?