New broadband maps are coming soon. Congress passed a law directing the FCC on exactly how to build the new maps, and the FCC is moving expeditiously to bring them to life. It’s called the Digital Opportunity Data Collection (DODC) effort. I’m going to point out some areas of concern, but I trust the new FCC under Jessica Rosenworcel to see this through to implementation, making slight additions or corrections where necessary. For years she’s championed good data and will undoubtedly work in the public’s interest in bringing these new maps to life. With that as background, I offer these suggestions: (1) being careful not to start undercounting, (2) using 2020 Census data, and (3) incorporating as much transparency as possible — to make sure we don’t all end up disappointed.
First, let’s try to size the problem. We know the Form 477 over-counts because if one house is served in a census block, every house is considered served. As I, and others, have written about previously, closing the Homework Gap isn’t just about “passing” every American with broadband if they can’t afford it or don’t sign up. It’s about actually getting broadband service to all Americans as a utility — like water and electricity — so they can share in the economic opportunity. Here we’re talking about the “access” numbers because that’s, well, unfortunately how the FCC has measured progress previously.
Dr George Ford of the Phoenix Center did the most thorough analysis of the problem, finding that there may be “4 million homes said to have broadband that may not”, or put another way, the 477 data overstates access by 3.3 percentage points — 1.9 points in urban areas and 11.5 points in rural areas. In the above chart, the red bars stay the same (the ACS numbers aren’t affected) but the blue bars are reduced by these amounts.
We have a big plan to fix this problem. The FCC opened the Digital Opportunity Data Collection proceeding, and after a long history of “scuttling efforts to improve broadband mapping” USTelecom — an industry group of ISPs — decide to “lead the charge” for better maps. Through an extensive record with the FCC, USTelecom, under the name the “Broadband Mapping Consortium”, and including at times presentations by the consulting firm CQA , advocated their position for the creation of a Broadband Serviceable Location Fabric. Congress eventually weighed in with very specific direction for the FCC, passing the Broadband DATA Act which became law March 23, 2020. The Broadband DATA Act is substantially similar to the USTelecom proposal, and mimics language around the creation of the Broadband Serviceable Location Fabric, among other things.
Suggestion 1: Focus the Fabric on “primary” residential serviceable locations, and have a plan to ensure it hits the target. This will avoid undercounting access.
My concern is that in an effort to fix a well-understood problem — over-counting access by ~3.3 percentage points — we’re dismantling the whole system and rebuilding it from scratch, with the potential to swing the error in the other direction to under-counting.
The new statutory requirement is that the FCC create this Fabric of all “broadband serviceable locations” in the United States. Currently, the Homework Gap is the number of Americans without access to broadband divided by all Americans. The Fabric becomes the new denominator. Now the Homework Gap will become “serviceable locations without access” divided by “total serviceable locations”. With billions at stake in public funding to close the Homework Gap, the details of this equation are crucial.
The Third Report and Order in the DODC proceeding tries to settle unanswered questions. It defines a “residential location” as “all residential structures” then goes on to specifically include some type of residential locations. The Order predicts the Fabric will include all individual structures to which broadband Internet access service can be installed. The FCC included more descriptive examples of where this dataset might go awry in earlier comment rounds: how to exclude “dilapidated houses and sheds” (para 173) , or “the farmhouse and each garage, barn, chicken coop, storage shed, etc.” (para 101). A BMC letter updating the Commission on their Fabric pilot includes discussion of identifying a “primary” structure (para 6) among all the structures on a single parcel, but the DATA Act nor the latest FCC Order institute rules to that effect.
CQA’s presentation of their technology (see slides at end) is convincing in its ability to identify structures from satellite imagery12; however, its ability to determine what is a “primary” residential structure versus, for example, a barn or large chicken coop, was not convincingly addressed. They say “parcel and land attribute data, address data, and other sources” is used in identifying serviceable locations, and that “these multiple data sets are available for the entire country”. Yet in their presentation to Congress, they talk about the obvious data challenges standardizing parcel and tax assessor across every county and jurisdiction. To the extent the Commission does use algorithmic identification of locations, the inputs and accuracy of the model should be well understood publicly, and the vendor contractually committed to claims that underpin the quality, such a national coverage of detailed datasets like parcel locations and tax records.
For the Fabric to be successful in measuring and closing the Homework Gap, it will need a precise count of “primary” residential locations — one that can be matched up with the 2020 Census data.
Suggestion 2: Ensure interoperability with the 2020 Census
In the Second Order and Third Further Notice the Commission asks for comments about how to ensure the quality of the Fabric. While there was a mention of using 2020 Census data for quality control, interoperability with Census data didn’t make the final rules. Over the summer of 2021, the Census will release its detailed block level demographic data, including population, housing units, and households. This dataset is indispensable as reference data to the Fabric and new DODC program. It is critical that the Fabric’s locations include the Census block that contains them. This will allow the Census’s official counts of housing units and households to be compared to the Fabric, and the many population models based on Census block data to be compared against ongoing updates to the Fabric.
Likewise, it is important that the provider shapefiles use the same basemap/reference data and geocoding solution as the Fabric. CQA made this point to Congress. In the figure below, a hypothetical provider creates their access polygons by buffering street segments where they provide service. But since the provider only uses address, and doesn’t use the same geocoder as the Fabric, the shapefile they report (light green) doesn’t include the rooftop geocode from the Fabric (green dots).
This isn’t dissimilar from suggestion #1. Calculating the Homework Gap by associating the Fabric’s geocode of a location with the providers’ shapefiles, we’ve disassociated this important calculation from the Census block. It becomes very difficult — and inaccurate — to see whether an entire Census block has access to broadband. At a minimum, the FCC should aggregate the Fabric locations and the provider shapefiles at the block level, producing a similar dataset to the Form 477 data, and release that publicly.
Suggestion 3: Transparency is important to getting this big change right
This new Fabric will be the basis for billions of dollars in public funds to close the Homework Gap through RDOF and hopefully other programs, and deserves the same rigorous public scrutiny as any other Commission dataset — if not more. However, the Second Report and Order says the Fabric is likely to be subject to licensing agreements that will “limit the extent to which it will be made available” (para 85) and that it will not be subject to “routine public inspection”.
While the Commission hasn’t implemented the DODC process yet, a proprietary serviceable location dataset was used in the Rural Digital Opportunity Fund Auction. Free Press has raised questions about why road medians were funded in that auction, and I’ve speculated that there could be a problem in the underlying dataset that put eligible locations in those medians — but we don’t know because we don’t have access to the data.
USTelecom/Broadband Mapping Consortium offered that the dataset would be available in map form, but the public would not be able to download the dataset. That’s a start, and of course the FCC should do that. It might even answer what happened in the RDOF auction. But it won’t allow for the type of rigorous review that is necessary to compare the Form 477 data with the new Fabric.
In conclusion, the FCC deserves credit for moving quickly to improve its broadband maps so funding decisions are made with more accurate data. I offer these suggestions with the hope the FCC will use their power through the contracting and implementation process to ensure the final product is the accurate, granular, and verifiable dataset we all want to see.