@Ryan_Fox_Squire_SafeGraph @Lauren_Spiegel_SafeGraph @ross_epstein_safegraph per ross’s suggestion over pm to bring this query over to the group chat, is it possible to use safegraph to measure visitation to workplaces?
Hi Alan, it’s not the best use of our data at this moment. We don’t have polygons for office type workplaces. We have starbucks, universities, parks, etc. We like to say we have every place people spend money or time outside of work or home.
So, to determine work, we have to use heuristics like where are people spending 9-5 M-F and NOT seen at other times. By doing this we created a “work panel” which is used internally for some of our products but is very sparse since it’s a strict definition of work.
I’m working on creating a social distancing index dataset that would get at whether people are leaving their homes to go to any place including work. Would that help?
Not sure about Alan, but the social distancing indexing dataset you describe would be super helpful for our purposes
@Lauren_Spiegel_SafeGraph that would be helpful indeed. however, however sparse it is, the dataset you described could still be useful for one of my projects. please consider providing it. thank you.
@Lauren_Spiegel_SafeGraph correct me if I’m wrong, but isn’t there a good amount of observations for Insurance Agencies and other brokers (e.g., State Farm, Edward Jones)? Obviously not an exhaustive set of workplaces, but may be enough for what @Alan_Kwan is looking for.
We’re actually trying to filter out “work visits” right now (though not that effective since we’re not finding that many workers based on our methodology). The best way to actually determine workers is to look at the bucketed dwell time and see who is staying abnormally long.
thanks lauren, that is so helpful. to provide some color, i am interested in a cross-firm, cross-industry measure and less interested in specific types of firms (e.g. insurance brokers, although that is a very thoughtful suggestion, thank you). my main goal is causal inference, so its less important to get the exact headcount at an office as much as it is to get the fraction of people who stopped visiting. for example, i am interested more in the fact a 1000 person office, where observe 100 workers, we only observe 12 now, even if the 100 is off.
@Alan_Kwan, I am proposing to add two columns to our new Physical Distancing Metrics dataset that gets at the worker issue. See the last two columns here and let me know your thoughts.
that would be very helpful. would you be able to do it at the level of the office? this seems like it’s at the level of a census block.
i know its awkward to measure for a 50 floor building.
we don’t have the locations of offices so we can just figure out proxies for work like behavior
okay, thank you. this is already tremendously helpful. thanks so much.