Mlab is an online, open-source speed test which publishes their data to Google
BigQuery - it is maintained by Code for Science and Society. This data is
aggregated over a number of speed tests performed by users themselves and are
pretty likely to suffer from some sampling bias, not to mention some relatively
sparse data in rural areas where density is lower. Additionally we might see
some issues arise from non-Broadband related bottlenecks (routers, clients, etc).
Still, at some point, it might be worth spatially joining this over some areas of interest to see what kinds of discrepancies exist (especially if we are unable to find an FCC dataset that has contractual speeds instead of advertised speeds).
This is data of form 477 submissions by ISP's with fields for census blockgroup, advertised up and downspeed, and type of service. According to the explanation page, contractual up and downspeed should be present but is not! That's a real shame since both are likely to be inflated, but contractual is likely to be much closer to the real thing.
These data are reported December and June - June 2019 data are the last time contractual up and downspeed were reported. Since this is relatively recent, we can use this for analysis purposes, with the hope that internal docs can be updated when/if the Biden FCC starts reporting contractual speed again (fingers crossed).
Interesting project from Microsoft comparing actual downspeed/upspeed from anonymized microsoft devices, aggregated to the county level. Obviously this is just windows devices, but probably a more representative dataset than M-lab. Since this is aggregated at county level, we wouldn't need a fancy spatial join, we could join these together using the first 5 digits of the census GEOID.
Only available in state by state chunks. Sure would be convenient to get all of the US in one fell swoop but it seems we can only get that down to blockgroup. In any case, if we could get all the data in state chunks, we can have a congressional area or county or other GEOID input which will fetch the proper FCC data, join against the proper TIGER data, join the proper census/acs data, all of which will (hopefully) be less computationally intense than doing it nationally.
Documentation for how to read 12-digit GEOIDS.
Description of TIGER shapefile columns and methodology. Brief description of columns: [ColumnDescriptions]!(img/reference/TigerCols.png)