Spoiler alert. We have, so it would seem.
When the 2020 census was finally released, we knew within hours that we could no longer count on the decennial census as the reliable base data it had once been. If the pandemic wasn’t enough to affect the quality, the bureau went down a path which deliberately resulted in data which was damaged at the block group level and made virtually unusable at the block level. Our chief demographer Matt Needham spent several months working to undo the damage, and we invented such terms as ghost blocks, mermaid blocks, baseball teams, and Lord of the Flies blocks.
We decided that we needed to lessen our dependence on the Census Bureau and direct our energy towards constructing a very accurate building inventory to use as our base. We summarized the goal in a hypothetical conversation where we fast forward to late 2030 and I ask Matt if he had downloaded the 2030 PL-94 data, and his response is ‘oh, was that today?’. The census, while still important, would no longer serve as the decennial anchor for small area data. Instead, we would need to use a broad range of big data sources from parcels to imagery.
The ideas we hatched at the time seemed rather aggressive, but in just a few years we have progressed to the point where we are already integrating them into our modeling systems – well ahead of our original goal.
So, what was that plan? In short, we would shift our focus from the census block level to the property level. To be effective, we would need to shorten the time lag between when changes occur on the ground and when they show up in data files such as the American Community Survey.
What are the data elements we are using or now building?
- Parcels
We own a national parcel database which is updated quarterly, allowing us to track changes in ownership, parcel geometry, and land use. Those changes provide valuable indicators into corporate land assembly, new housing developments, and changes in ownership and tenancy. - Building Permits
On a monthly basis we are tracking building permits by type – new residential construction, ADU, general residential permits, and non-residential. While not available in all jurisdictions, permit activity is the best single sign of imminent residential development and one of the few for identifying ADU construction. - Street and Address Inventories
Compiling a complete and always up to date roster of addresses nationwide is a key component of our strategy. The strategy involves using multiple sources to identify both new and infill addresses. So, we are tracking changes in street networks, address ranges, and summarizing lists to identify brand new addresses. For new multifamily buildings, using apartment and unit numbers intelligently can yield good estimates of how many units are present. The Census Bureau publishes a good file with block level dwelling counts that provides additional useful data. - Postal Change
While we have always used the USPS ZIP+4 rosters, we also utilize a private delivery service roster that tends to track actual change better. We have also always understood that adding a new ZIP+4 doesn’t mean growth has occurred – it usually means that growth might occur in the future since the postal service tends to be an early leading sign. However, in conjunction with parcel changes and building permits, we are in the process of constructing some very effective models that predict when a new ZIP+4 will come to fruition. - Image Analytics
With high quality imagery, it is both possible and practical to identify new construction at all phases – from initial land preparation through the day when the fleet of white Silverados are replaced with the family SUV. Imagery remains expensive, and we have no particular desire to count the moose population in Alaska, so judicious selection of the imagery to process is necessary. Our Hotblocks project being introduced this summer allows us to focus on those areas with active development. - Textual Analytics
Most large projects get talked about by city councils, planning commissions, local newspapers, builders and developers. Our targets cover everything from long term infrastructure projects (trunk sewers, water projects, fixed rail lines, and proposed freeways) to announcements of builder sales offices. Much of this chatter is not well georeferenced, but in conjunction with parcel level intelligence, projects can be identified, classified, and incorporated into current, short term, and longer-term projections. We have a pilot project underway that should yield methods that are consistent and can be applied to a growing list of target areas.
As we progress down this road, we will no doubt find other data sources to integrate, as our data ferrets are always on the hunt. The result? More robust current year estimates and greatly improved projections – our plan is to shift to 2, 5, and 10 year projections.
Have we gone crazy? Absolutely. But quantum leaps require vision, monetary and computing resources, and some very bright minds. We know we have all four.