Hi everyone. I was interested in comparing some of the SafeGraph data with real-life events that have been well documented and for that I looked into a march that happened June 6th 2020 in Philadelphia. According to several news outlets/AI count using aerial footage, there were between 50,000-80,000 people in that march, but no noticeable spike in the number of visitors can be seen in the SafeGraph data for the CBGs where the march happened. Is there an explanation for this?
In other words, from the SafeGraph data alone it would seem like that was just a regular day in terms of foot-traffic to the CBGs, but there is well documented evidence (e.g.: https://twitter.com/BradfordPearson/status/1269337799523733504) that there was increased activity in the area. Is this discrepancy perhaps due to some way in which data is collected from phones? For the sake of clarity, I’m taking CBG visits extracted from the Neighborhood Patterns Dataset.
+1 here. I seem to recall that the foot traffic data from neighborhood patterns for that protest and for the specific CBGs in that video did not show any out of the ordinary visits. Is it possible that SafeGraph didn’t record activity from those 10,000+ devices?
Hi @Austin_L_Wright_Univ_of_Chicago! This is a plot of visits that happened only on the Saturday. The largest protest happened Saturday June 6th, which is the orange point in the graph. There was another smaller protest on the Saturday prior to the orange dot, but not on other weeks as far as I’m aware. Each dot was extracted from the stops_by_day field for each CBG
Hi @Leonardo_Ferreira_Guilhoto_University_of_Pennsylvania, @Austin_L_Wright_Univ_of_Chicago, @Jorge_Upenn. I think something also worth considering is how SafeGraph treats movement. From this post:
> In theory if someone is moving continuously driving or walking (even slowly) our algorithm will not cluster that as a “stop” in space-time. However 1 min is a very low threshold and in some cases it is possible that we will get false positives if someone is stuck in traffic or stops at a really long traffic light, etc. It’s not perfect.
Of course I can’t say for sure, but I would expect at least some of the protesters are not appearing in the counts because of this. Interested to hear your thoughts
Also, this paper by @Dhaval_Dave_Bentley_University, @David_Van_Dijcke looked at protest behavior. They found that net stay-at-home behavior increased following protest onset, which is a bit counter-intuitive, but it may agree with what you’re seeing. The reduced visits from non-protesters outweighed the increase in visits from protesters. You could perhaps explore this further by using Weekly Patterns to see if there were significant changes in visitor_home_cbgs for the dates of interest (where visitors were coming from).