I’m running into a visit attribution issue with the Patterns data

Hi @Ryan_Fox_Squire_SafeGraph: I’m running into a visit attribution issue with the Patterns data. Non-essential retail businesses (clothing stores, florists, book stores, etc.) in NYC were closed by executive order on March 22. Visits to these points of interest should drop to 0 (or very close to it). In fact, visits attributed to non-essential retail points of interest dropped to ~23% of February baseline levels as of the week of 04/06 (the low-point). Visits to these points of interest had recovered to 35% of baseline levels as of the week of 05/25, before the City started to reopen. What might explain the apparent continued presence of visitors at closed businesses, and what if anything can we do to cleanse the data? In particular:

  1. Is there anything in SafeGraph’s visit attribution algorithm that would filter out visits to apartments/offices/etc. above POIs?
  2. Am I correct to read the visit attribution white paper as saying that pedestrians walking by would be filtered out and would not be counted as visits?
    I assume some of the attributed visits are actually nearby loiterers misattributed as visitors or visitors to nearby open POIs misattributed as visitors to closed POIs, but it’s hard to believe that either is the full explanation (especially since I know the visit attribution algorithm should minimize misattribution to neighboring POIs).

If it helps: here are a few non-essential retail POIs where visits did not drop anywhere close to 0:
sg:0368bf01ca98400c86d0a999925fc230
sg:065f98f8bffa4a77ad52bcc4a17eb1a5
sg:13fb2e783b3542059de1e140a0a521a2
sg:02157d3de5174318b0b14bc6b661abf9
sg:1c5b0bfa408347f9b1a450d191b88bf2

Hi @Ezra_Cohn_NYC_OMB glad you found the visit Attribution White paper
thanks for providing examples

to answer your questions
for Ezra’s questions

  1. Yes, when we think a device lives there it will not be counted as a visit.
  2. yes, a device that is waling by will not show a spatio-temporal clustering pattern associated with actually stopping and visiting a place. That said, it can be very difficult to distinguish someone who stops outside the door of a POI for 10 minutes without going inside vs someone who goes inside.
    I wrote up some additional info here:
    SafeGraph Community FAQs - Google Docs

Thank you, @Ryan_Fox_Squire_SafeGraph–this is quite helpful. Are devices that typically spend daytime at the location filtered out as well?

@Ezra_Cohn_NYC_OMB No we don’t filter out common day time locations. Previously we attempted to do a common-day-time, work time filtering, but we moved that last month because it wasn’t very reliable.

Got it. Thank you, @Ryan_Fox_Squire_SafeGraph!