Predicting new FCC Broadband Maps with Current Data

The FCC is diligently working on new broadband maps. The stakes are high since all of the IIJA broadband funding requires the new maps to exist before the money can be allocated to states. The new maps will fix a well-known flaw in the current “Form 477” maps, which is Census blocks are either served or not served - there’s no concept of a partially served block. Here, I use a simple technique to make an estimate of what the new maps might look like, using only data currently available.

The innovation of the new maps will be that they'll use a “fabric” of locations to show the “network edge” of broadband availability at the address level, instead of for the entire Census block. To find the network edge using current data, I take a simple approach: for a given Census block, do all its neighboring blocks also have broadband? If the block is surrounded by other blocks with broadband it can’t be the network edge. However, if a neighboring block doesn’t have broadband, that might be the network edge.

Above is the outskirts of Reading, Pennsylvania. The red areas don’t have access to 100/20 broadband according to Form 477 data. In the current data, both green and yellow areas are considered to have access to broadband. I’m simply making the assumption that if the red areas don’t have access to broadband, there’s a good chance its’ neighbors are partially covered, which turns them from green to yellow. Here’s what it looks like for all of Western Pennsylvania:

For the commonwealth of Pennsylvania, 4% of the population doesn’t have access to 100/20 broadband using current Form 477 data. Using this method, the 4% number is the same. We add 4.6% of the population that lives in areas that might be partially covered. If these blocks had 50% coverage, that would add 2 percentage points to the number of unserved Pennsylvanians

The most computationally intense part of this is the spatial query to find each blocks’ neighboring blocks. There are 11 million 2010 Census blocks in the United States. On average a block has ~7 neighboring blocks. If it’s helpful to anyone, I’ve put the national file of neighboring 2010 blocks here. It’s 72 million rows.