I have a question about the number of devices. I’m doing research in the Bronx, NY (FIPS = 36005) and I downloaded all of the weekly home panel data and summed the “number of distinct devices observed with a primary nighttime location in the specified census block group” for the Bronx. The median is 88,013, which is only 6% of the Bronx population of 1,439,725. In a Safegraph article on sampling bias, a chart showed the number of devices in the Bronx = 352,829, which is 24.5% of the population. I’m trying to figure out the difference – is the number of “home” devices much smaller than total devices? Am I overlooking something obvious? Thanks in advance for your help.
Is the number of “home” devices much smaller than total devices? Am I overlooking something obvious?
@Nevin_Cohen_CUNY_School_of_Public_Health What is the basis for calculating the median of “number_devices_residing” ? I assume there are multiple census block groups in Bronx and for the total number of devices residing in Bronx you would need to sum all of them up.
For each week (from 12/31/18 through 6/15/2020) I summed the devices for each CBG in the Bronx (to get the number of weekly devices for the Bronx), and then took the median of all the weeks.
OK I will check and get back to you.
Can you also point to me where is the chart showing the number of devices in Bronx as 352,829? Thanks!
The table is in this document under the section about Sampling Bias at the County Level: Google Colab
Hmm, I see. So there is an actual decrease in number of devices…
Thanks so much, Jessica. Does anyone else have an idea why the number of devices in the Bronx is so much lower in 2019 (and 2020) than in 2018?
The SafeGraph panel changes month to month and year to year based on a variety of factors. It is not a stable panel. It looks like we just had significantly more devices in 2018 compared to 2019 and 2020.
If you are comparing data across time periods it is critical that you take into account changing sample sizes over time (as reflected in home_panel_summary and/or normalization_stats). I have some additional resources about this listed here: https://docs.safegraph.com/docs/data-science-resources#section-panel-normalization-for-longitudinal-analysis-sampling-bias-corrections-and-extrapolation
and am working on additional demos for longitudinal analysis, but this is always challenging.
Okay, thanks Ryan.
@Nevin_Cohen_CUNY_School_of_Public_Health please let us know if you are able to figure this out or if you run into trouble with any of your analysis.