Hi All, apologies if this has been asked previously, but I couldn’t find answers to these two questions:
- In the Neighborhood Patterns dataset, does the “device_home_areas” variable include visitors that might have travelled for reasons such as commuting?
- What is the difference between the “device_home_areas” variable in the Neighborhood Patterns dataset and the “destination_cbgs” variable in the Social Distancing Metrics dataset?
Hi Matthew, thanks for the response! Just as a follow up, I am curious to better understands what qualifies as a visit (beyond purely the one minute presence criterion you outline in 2)). Perhaps, my question can best be broken down into a series of scenarios:
- If I live in home census block group (CBG) A and visit a friend in CBG B who lives in a residential location in CBG B, will this be recorded as a visit?
- If I live in CBG A and travel to work in an office building in CBG B, will this be recorded as a visit?
- If I live in CBG A and travel to a commercial location in CBG B, will this be recorded?
Hi @Kamen_Velichkvo_The_Wharton_School, I can add to what @Matthew_Dawidowicz said and answer your scenarios.
Each row in Social Distancing Metrics and Neighborhood Patterns corresponds to a CBG. In SDM,
destination_cbgs tells you where the row’s CBG visited to. On the flip side, in Neighborhood Patterns,
device_home_areas tells you where the row’s CBG had visitors from.
Both those datasets are aggregated at the CBG level. In contrast, the Monthly/Weekly Patterns datasets are aggregated at the POI level. So Monthly/Weekly Patterns would only see a visit from situation (2) and (3), assuming the office building and commercial location is in SafeGraph’s POIs.
Hi @Ryan_Kruse_MN_State and @Matthew_Dawidowicz,
First of all wanted to thank both of you for the responses and insights. I think all of this has clarified things for me, but I think I might still be struggling with defining the full scope of Safegraph data.
I am interested in understanding exactly what pings from cellphones are being recorded and reported.
• Does the Neighborhood Patterns dataset report any ping that has made a stop longer than 1 minute within a given census block group (CBG) regardless of whether the location of the ping is in a residential, office, or commercial location, or even a sidewalk, park, or street?
• In addition, if I am driving from cbg A to cbg C via cbg B and I stop at a traffic light in cbg B for more than 1 minute, then will the data pick up my traffic light stop as a visit in cbg B? What about the same route but if I am walking instead of driving and decide to stop on the sidewalk in cbg B and take a 1 minute phone call?
I do apologise if these seem oddly specific questions, but I am just trying to imagine different situations and decide whether the method under the “How was it created?” section of the Neighborhood Patterns documentation (Neighborhood Patterns | SafeGraph Docs) would capture these situations.
@Kamen_Velichkvo_The_Wharton_School No apology necessary! You are welcome.
First, there may be some value in looking at the visit attribution whitepaper, even though it’s more for Monthly/Weekly Patterns than Neighborhood Patterns.
SafeGraph does not make public exactly which sources the pings come from. It’s important to understand that even if SafeGraph had the exact locations of every device in its panel at all times, the panel is just a sample of the entire population—so there is always some uncertainty.
Here are some answers to your specific questions:
- Yes, as long as the ping is part of a “cluster” of pings lasting > 1 minute. In Neighborhood Patterns, it does not matter where in the CBG the cluster of pings happens. (For Monthly/Weekly Patterns, those clusters must be attributed to a POI.)
- Theoretically, yes it’s possible both those stops would become a cluster of pings that attributes a visit in CBG B. However, in reality SafeGraph doesn’t have continuous pings from the devices, so many of those “stops” might not even result in a single ping.
Hi Ryan, thanks so much for the response! I looked through the attribution whitepaper; it was tremendously helpful!