A First Look at the Evolving National Broadband Map
Maps of broadband availability are critical inputs for targeting public programs aiming to address disparities in digital equity, such as the $42B Broadband Equity, Access, and Deployment (BEAD) program.
These maps aren’t easy to produce: no one entity knows where each of the thousands of Internet Service Providers (ISPs) provides service, and historically the United States has relied on self-reported data from ISPs to determine which parts of the country are served or unserved by broadband. Federal broadband funding is often tied to the resulting maps, making this approach ripe for misreporting and inaccuracy.
In 2020, the Broadband DATA Act called on the FCC to create new, more granular, maps of broadband availability in the US. Although the new National Broadband Map (NBM), released in November 2022, still relies on self-reported data from ISPs, a key feature of these new broadband maps is that they are constantly evolving. Congress directed the FCC to develop a challenge process by which state, local, and tribal governments, along with members of the public, could contest ISPs’ claimed service availability.
So, in addition to the two major updates per year when ISPs report the locations they currently serve, the National Broadband Map changes every few weeks. This could be due to a number of factors:
- ISPs submit their reports late, after the deadline. The BDC process for ISPs is both new and more complex than the previous Form 477 process, so this isn’t unreasonable for the first couple years.
- ISPs submit modifications to their reports, either fixing mistakes or updating their estimates of coverage. Mistakes are common in these submissions — we’ll have more to say on this topic in the coming months, but our initial analysis suggests that at least 10% of ISP filings have some form of systematic issue impacting data quality.
- The FCC is correcting errors in their data processing pipeline. As with any large, new data processing pipeline, there are going to be data quality issues.
- Challenges to ISPs reported service availability, either due to challenges by individuals reporting service is unable in their location or via bulk challenges by governments (e.g., the Commonwealth of Virginia appears to have challenged providers who indicated coverage of a location in the FCC’s maps, but not in Virginia’s state broadband maps).
The last of these means that these broadband maps will always be evolving, by design, between the bi-annual major releases of the map. Our question is how has the map been evolving so far?
tl;dr:
- The National Broadband Map is changing, and has added a net of 2M served locations between Dec 2022 and Mar 2023.
- ISPs modifying their own previously submitted reports seems to drive most changes at this point.
- Challenges do appear to have meaningfully reduced the total number of served locations, but just five states make up three-quarters of this change.
Where has the map been changing?
At the highest level, since Dec 2022, more than 20 million records have been changed in the National Broadband Map: around 11.5 million locations had service removed by a provider, around 6.4 million had service added, and around 3 million had service modified.
Here’s an animation depicting how the NBM has changed over the last few months, in terms of absolute number of records changed.
Remember, a few caveats on this data (for more details, I describe our methods at the end of this post):
- We’re missing New York’s December 2022 data, unfortunately.
- The Feb 5, 2023 image depicts changes since Dec 12, 2022 — a much longer period than the others.
With those caveats out of the way, three features immediately jump out. First, the map changes with every snapshot — even if some snapshots have less change than others. Every state, Washington DC, and Puerto Rico have seen changes (though the other territories, notably, have not).
Second, we can often see outlines of states in the map, suggesting that the source of these changes are happening at a state level. Why is this the case? I’m not entirely clear. One possibility is that these state-level changes reflect the results of state-level challenges (e.g., by state broadband offices). Anecdotally, this seems to have been the case for our state of Virginia, where I noticed availability challenges wherever ISPs reported service on the FCC map, but not in Virginia’s state broadband map. But in the case of Vermont, these changes seem to be administrative in nature, due to a large number of locations across the state changing provider IDs. In still others (like the case of Mississippi, which we’ll discuss below), it’s plausible that operators are making state-level corrections to their submissions.
Third, the map isn’t really changing that much, on the surface. Each snapshot of the National Broadband Map consists of more than 870M records; the largest change we see in a single snapshot is 8.6M, barely 1%. But nearly half of these records are from satellite providers, and what really matters is the marginal impact of these changes — how do they impact our understanding of who has access to reliable broadband?
Changes to served/unserved designations
Let’s turn to the multi-billion dollar question: which locations are served versus unserved. There are three forces acting here: growth in locations served due to late reports, growth in locations served due to providers updating their data, and decreases in locations served due to challenges. Let’s take a look at how these stack up.
I’m considering all “underserved” locations as unserved here, for simplicity; more broadly, I agree with what others have written calling the “underserved” category a distinction without a difference for the purposes of BEAD.
Two million additional locations are “served” nationally, but this varies by state.
Overall, the National Broadband Map has seen a net increase of 1.98M served locations, but this varies significantly from state to state. Illinois and Puerto Rico lead the way with more than 1.7M and 620k locations added, respectively. On the flip side, California has seen more than 240k locations removed, with Mississippi and Washington state following with 63k and 50k (our state of Virginia is in fourth place, removing 27k locations).
New ISPs only account for 289k newly-served locations.
As of this writing, there have been 73 new ISPs that have shown up in the National Broadband Map since December 2022, collectively accounting for just over 2.5M serviceable locations. At the same time, only ~289k (11.4%) of these locations became “served” as a result of these late additions. Every state, with the exception of Hawaii and Louisiana, had at least one new ISP, with most new providers providing unlicensed fixed wireless (30) or fiber (19).
Modifications to reported coverage account for most of the remaining increase.
That means the remaining net increase of 1.7M locations that became served since the map was published came from providers who were updating their coverage. These changes come from true modifications, where a provider actually revises their reported service availability at a specific location. Others come via additions, where a provider adds locations to their reported service area.
- Comcast in Illinois is the biggest example of a mass true modification. Indeed, the bulk of the nationwide net increase came from an update that Comcast made in Illinois to their reported max service speeds in 2.7M locations. Previously, they had reported providing 300/10 service to these locations, which does not meet the definition of “served”, whereas by Feb 2023 they had increased this to 1000/25, which does. This accounts for nearly all of the increase in served locations in Illinois.
- More common were increases in reported coverage. In Puerto Rico, for example, one provider (Liberty) added service at more than 900k locations, while they dropped service at just over 2600. Crucially, this provider didn’t modify any locations that were included in their initial report — they mostly just added locations that weren’t initially included, unlike what Comcast did in Illinois.
The latter of these appears to be much more common. Ignoring Illinois (which is dominated by the large Comcast modification), only 255k locations have been modified since Dec 2022, compared to nearly 6 million locations added.
512k locations once claimed as served become unserved.
Perhaps more interesting are the locations that drop off the map, since these are the places that are likely to have been challenged. As before, there are a few canonical types of changes we see.
- Coordinated, state-level challenges. A few states, such as California, Virginia, and Washington, show consistent, ongoing declines in served locations over multiple snapshots of the map.
- Isolated challenges. Every state has at least a few locations move from served to unserved during the timeframe we analyze (good examples include Florida, Tennessee, and South Carolina).
- Provider updates. Consider Mississippi’s changes as of Feb 5, 2023. Here, it turns out a single provider, C-Spire, removed about 330k locations that they had previously reported as served with 1Gbps symmetric fiber; of these, this modification meant that around 64k locations were no longer considered “served”. Curiously, these locations seem to follow roads. I happen to know that C-Spire is a mobile operator that also offers fixed broadband via fiber. Perhaps these changes were a correction on C-Spire’s part — most mobile operators have fiber plant used for their mobile network that isn’t easily “broken out” to provide fixed service. I could easily imagine their initial filing mistakenly including their long-haul fiber plant when calculating their fiber-to-the-premises footprint.
The counts of these newly-unserved locations gives us some insight into how effective states have been with service availability challenges. Let’s assume that changes in served locations are evenly distributed across states. We can compare the amount of newly-unserved (i.e., successfully challenged) locations to the total number of locations [1] in each state, capturing the intuition that larger states should see more locations become unserved if the challenge process is working. The below figure captures this idea of states over- or under-performing expectations in terms of the number of locations that have moved from served to unserved since Dec 2022.
So what?
The NBM is supposed to change over time; that it does isn’t surprising. But understanding how the map is changing within a release gives us insight into how effectively the challenge process is working to make the maps reflect the reality.
The National Broadband Map doesn’t just change due to challenges. We expect challenges to impact map coverage, but so far, the biggest changes (especially when it comes to served/unserved status of a location, are most likely due to modifications of submitted availability reports. BDC is a new process, and it’s not surprising that we’re seeing churn in the input data to the map in its first year.
Served locations vary meaningfully over time, and they don’t always trend downward. It’s also a reminder to those of us working with this data that the NBM is a moving target, and depending on your analytical goals or the state(s) you care about, choosing a different bi-weekly release can yield quite different results. For example, if you cared about the state of connectivity in Illinois and used an early release of the map for your analysis, you’d be undercounting served locations by almost two million locations — nearly 50% of total locations in the state!
Challenges are here, they’re just not evenly distributed. We see evidence of change in the map that suggests challenges are having an impact on reported availability. But a handful of states seem to be driving most challenges, such as Virginia and California, and an even smaller set is impacting availability. There isn’t likely something that’s uniquely bad about reported service availability in these states, so what this says to me is that there are many locations that have overstated availability due to some combination of a lack of challenges, poor quality challenges, or misreporting (intentional or otherwise) by service providers.
What’s next?
Of course, the big “diff” that will matter is when the next edition of the National Broadband Map drops in a few weeks. Will successfully challenged locations reappear in the new map, for instance? What about successfully challenged fabric locations — how will we compare the maps to see what has changed if a location ID maps to a completely different location? We’ll look forward to digging into that in the coming weeks.
We’d love to talk to you if you’re trying to challenge broadband availability where you are: email us at spinlab@vt.edu.
Thanks to fellow SPIN Lab member Syed Tauhidun Nabi for his help in gathering the December 12, 2022 snapshot of the National Broadband Map.
Methodology
The FCC releases updates to the NBM roughly once every two weeks; we refer to each of these biweekly releases as a “snapshot”. We took a snapshot of the publicly-released bulk downloads for the entire US in Dec 2022, in the weeks immediately following the NBM’s initial release. In early Feb 2023, we realized for the first time that the NBM was being updated regularly, and have captured every snapshot released since then. Note that all of these snapshots are for the July 30, 2022 BDC filing deadline, using the corresponding version of the Fabric.
Note that due to a data processing error, we do not have data for the state of New York from December 2022; we either remove New York from our analysis or substitute in its Feb 2023 data where appropriate. We also don’t have a copy of the original November 2022 map. If you do and would be willing to share it, please get in touch!
We compute the pairwise changes between each subsequent snapshot of the NBM — e.g., comparing the Feb 5, 2023 snapshot to the Feb 16, 2023 snapshot — to produce a “diff” of the map at each stage. This allows us to see what changed at each point in time. These diffs capture locations that have been added, dropped, or modified between each pair of releases.
We assume that drops from the map stem from service availability challenges, and that all other changes (additions or modifications) are the result of updated filings by ISPs. These assumptions aren’t perfect: ISPs presumably can remove locations on their own, and challenges could result in modifications to services offered or, in principle, even add new locations to a provider’s served locations. Unfortunately, without access to the challenges themselves, it’s not yet possible to validate our assumption.
Finally, we used a slightly more generous definition of “served” here than what NTIA has proposed for BEAD. Specifically, we assume that any terrestrial technology providing 100/20 speeds counts as served — even unlicensed fixed wireless, which NTIA doesn’t consider. Without getting too far in the weeds, the June 30, 2022 data collection period for the NBM didn’t distinguish between unlicensed fixed wireless and shared spectrum fixed wireless (e.g., CBRS), and I understand NTIA will consider the latter to meet their definition of “Reliable Broadband Service”. More fundamentally, unlicensed fixed wireless under the right conditions can absolutely provide reliable broadband service, regardless of what NTIA says: it’s important to remember that unlicensed wireless these days includes millimeter wave systems a la Terragraph and even next-generation 5GHz systems like Tarana, both of which can provide gigabit-class service in real-world conditions.
[1] We don’t actually have access to the underlying Fabric, so we just use the total number of location IDs that show up in the National Broadband Map for each state here, which misses any wholly-unserved Fabric locations.
