Update: Comparing the New FCC Fabric to the Census

Focusing on where the Fabric overestimates locations

Not long after the new FCC maps came out, I published a comparison of the FCC Fabric locations against the number of housing units according to the Census. The takeaway from that analysis was that the FCC’s Fabric actually seems to have too many locations, not too few. In America’s least dense 2,148 counties, there are 25.5 million Broadband Serviceable Locations compared to 24.5 million housing units according to the 2020 Census. Since then, the FCC released a file that contains the number of “units” (usually housing units) in the Fabric, which provides a more direct comparison to the Census. The result is the same: there are 158 million “units” in the Fabric and 140 million housing units in the 2020 Census. In the least dense 2,148 counties, there are are 30 million “units” in the Fabric and 24.5 million Census housing units.

The chart below shows the difference between the Fabric count of units (housing units + business locations) and the Census housing unit count. A higher percentage difference means there are more Fabric units than Census housing units. What we see is that as counties get more rural (moving to the right in the chart), the Fabric increasingly has more locations than the Census. In the least dense counties, the Fabric routinely has 40% more locations than the Census. We hear a lot about missing locations in the Fabric; we don’t hear as much about over-reported locations.

In some states, the difference is dramatic. In the least dense 206 Texas counties, there are 2.4 million BSL housing and business units compared to only 1.8 million Census housing units. Some of that difference will be rural business locations that aren’t in the Census; but not nearly all 630,000 of it.

It only takes a little bit of scrolling around the FCC map to see the real life implications. Here are some examples in Edwards County, Texas, west-northwest of San Antonio, where there are 3,282 housing and business units in the Fabric but only 963 in the Census. First up, a roadside in Rocksprings, Texas:

Next is a sand road that is slightly more ambiguous — you can imagine how AI thinks there’s a structure there — but the “road” isn’t connected to the house which is suspicious.

Here’s another. A sand road to nowhere. Unless more recent imagery suggests a house has been built in this area, there’s no structure here.

I can’t claim to know these areas. But to be clear, Texas will get BEAD funding for all of them. Of course they won’t have to build to them because they don’t actually exist.

Texas is the state with the largest difference, but it isn’t the only place this is happening. In Nebraska, the Fabric thinks there are 492,097 units in the 89 least dense counties (excluding the most dense 4), which is 430,244 unique Broadband Serviceable Locations, but has only 365,504 housing units according to the Census. Businesses are not counted in the Census, but are BSLs. I don’t think that explains these large differences in rural areas.

As a summary, I think it’s presumptuous to assume the FCC Fabric underestimates locations in rural areas. In fact, I think it’s possible — even likely, given current data — that the Fabric over-represents rural locations compared to the ground truth. I still think the right approach is to keep moving forward with the NTIA’s allocation. It will never be perfect.