The 2020 Census provides a new source of "ground truth" for unserved locations

Every calculation of the “Digital Divide” has a denominator. We can assess the population that is unserved by broadband, the residential housing units unserved by broadband, or with the new FCC maps, “broadband serviceable locations.” My analyses are based on census block-level housing unit projections for 2019 (based on the 2010 Census), published by the FCC. In April 2021, the Census published the 2020 decennial block-level housing unit counts, which updates the “ground truth”, and it’s worth examining how these numbers change the “Digital Divide”. Nationally, the 2020 Census has 140 million housing units, a small 0.6% increase from the 2019 projection. However when we looked at housing units unserved by broadband according to the June 2021 FCC Form 477 data, there are 19% fewer unserved housing units.

To be clear, nothing changed. The areas that are unserved according to the 2019 projections (based on the 2010 Census) represent 3.6 million housing units, but the same areas only represent 2.9 million housing units in the 2020 Census data. The two data sets aren’t perfectly interchangeable. The 2010 blocks could have been split into multiple 2020 blocks, or multiple 2010 census blocks could be merged into a single 2020 block. The Census provides a “crosswalk” table between the two.

Let’s look at an example: This census block, around Joshua Tree National Park in Riverside County, California, had 21 housing units according to the FCC’s 2019 projections. It was split into four blocks in the 2020 Census, but none of them have any housing units. What’s likely happening is the same problem that beset RDOF and I wrote about at the time: in an effort to update projections for county-level population growth since 2010, somehow Census blocks that shouldn’t have population were given population. The 2020 Census corrects that back to zero.

Here’s another: three Census blocks in San Diego county had a combined 57 housing units according to 2019 projections. They were combined into one Census block for 2020 and have 43 housing units.

It wasn’t a sure thing that 2020 Census data would be able to be used in this way. The Census Bureau applied “differential privacy” to many of the block-level fields in the 2020 Census. It has been controversial. Differential Privacy adds “fuzziness” to the block-level data in an effort to obscure individual Census responses. Some — including privacy advocates such as myself — would argue that the Census Bureau is trying to solve a non-problem. It is unlikely that an “attacker” would try to reverse engineer individual Census responses because the prize doesn’t yield any PII, only racial and age information about the household. On the other hand, the effect of differential privacy raises serious questions about the usability of the block-level dataset for some data fields. Luckily, however, housing units is considered “invariant” and is accurate at the block level, as confirmed by David Van Riper at the Minnesota Population Center. I haven’t seen the FCC’s new broadband maps, but I hope they have Census 2020 blocks on them. That would certainly make it easier.

Considering how different the new FCC “broadband serviceable locations” maps will be from the existing Form 477 block-level data, and the grumbling about the maps which has already begun, the FCC and CostQuest need “ground truth” to compare against. It won’t work if there’s nothing to compare a “broadband serviceable location” against, and it won’t work if the comparison is apples-to-oranges. What will work, in my view, is a comparison of residential broadband serviceable locations as identified by the FCC and CostQuest against the 2020 Census housing unit counts.