New Tech, New Math, New Methods In Massive Upgrade of Census Bureau Construction Programs
[ Article originally appeared in www.census.gov ]
Think satellite imaging, artificial intelligence, new data sources and new methodology replacing survey collection methods that date back to the 1950s.
Modernization will lessen the burden on local governments to report building permits or builders to disclose building starts and status reports. No more relying on sample designs that only provide data at the state level.
Instead, satellite images will soon give us instant updates on projects at every stage of construction from start to finish down to the county and metro level for the first time.
The changes will modernize nearly every aspect of the construction data the Census Bureau releases, including New Residential Construction (NRC). A Principal Federal Economic Indicator (PFEI), the NRC data are one of the most timely and leading official measures of the current condition of the U.S. economy.
While use of satellite imagery and new data sources are still in development phases, other improvements are already on the launch pad.
The first estimates from a new sample design are set to be published with the February 17 release. The local and county level estimates from the redesigned Monthly Building Permit Survey (BPS) are scheduled to be published for the first time February 24.
To appreciate this massive upgrade, it’s vital to understand just how important new residential construction statistics are to measuring the U.S. economy and just how outdated construction data collection and processing had been — until now.
What Are Indicators?
The Census Bureau produces 13 of the PFEIs, which answer questions like: How many imports and exports were there this month? What were hospital revenues this quarter? How many homes were built?
In addition to counting every U.S. resident each decade, the Census Bureau is responsible for collecting economic data and it does that primarily through sample-based surveys, including the BPS and the Survey of Construction (both of which feed into the New Residential Construction (NRC) PFEI.
Focusing exclusively on the construction sector, the surveys work in concert to report monthly measures of construction activity, housing inventories and other key inputs to the quarterly Gross Domestic Product. These data provide a window into the health of our economy.
It All Starts With a Permit
Since 1959, the Census Bureau has published monthly NRC data, which measure the construction of single-family and multi-family homes in the United States.
There are basically three main construction phases — the permit, the “start” and the “completion.” Construction begins with a building permit issued by a county and city or other local jurisdictions.
Through the monthly BPS, the Census Bureau collects data on how many homes are authorized to be built. A permit does not guarantee a house will be built but is an indication that it is likely to be built.
Building permits are the foundation of the Census Bureau’s construction data programs and are generally viewed as a leading indicator of construction activity.
Up until now, the Census Bureau has relied on roughly 9,000 state and local governments and other jurisdictions to provide monthly building permit data. That provides a representative sample — not a complete count.
But when the BPS modernization is completed, the Census Bureau will receive permit data from third-party sources, decreasing the state and local governments’ burden to report. This should reduce the Census Bureau’s reliance on voluntary reporting and improve the effective response rate.
Most notable — and not only to statisticians and math enthusiasts — the monthly BPS is also changing how it processes data points.
Currently, the Census Bureau’s sample used to produce the monthly BPS estimates only represents the state level.
By going from a representative to a “cutoff” sample, the Census Bureau will be able to produce monthly estimates for every jurisdiction in the United States, down to the local level.
What’s a cutoff sample? The new BPS cutoff sample design focuses data collection on places with the highest level of permit activity and relies on modeling for places with little to no permit activity.
These cutoff sample estimates will generate complete county and metro area coverage as well.
The change will provide a much clearer picture of construction activity, a shift tantamount to going from the Flintstone’s grey rock television to a 4k ceiling-mounted projector.
It will also allow for more detailed data visualizations and make it easier for jurisdictions of all sizes to compare their construction activity to others of similar size.
From the Flintstones to the Jetsons
While permits are an indicator of future construction activity, “starts” provide the first sign of actual construction activity taking place. The Survey of Construction (SOC) measures the number of homes in the process of being built and/or completed.
What constitutes a so-called start? Generally, the SOC labels the excavation for footings or a home’s foundation as a start, and installation of finished flooring as a “completion.”
So how does the Census Bureau know if a construction project has begun and what stage it is in?
Since the 1950s, the Census Bureau has contacted builders to ask about the status of their projects. If a builder didn’t reply, Census Bureau sent a field representative to the construction site to talk to someone and survey it in person.
Pair the cost of dispatching people to construction sites with the current statistical methodology, and you end up with limited ability to scale the survey affordably.
Here’s how the Census Bureau’s construction branch will emerge from the Stone Age.
After collaborating with Statistics Canada and further academic research, the Census Bureau team designed a way to survey construction sites by satellite. When fully in place, satellites will take thousands of photos of places that have a building permit from their orbit high above earth.
Those images will then be fed into the Convolutional Neural Network (CNN), which classifies the aerial images. Census Bureau staff and contractors used thousands of aerial images of buildings to train a computer program to identify three stages of construction: start, incomplete, completed.
The process of taking third-party permit data, surveying those sites by satellite and analyzing the images with artificial intelligence will produce accurate estimates while meeting monthly indicator timelines at a cost equal to or less than existing methods.
And with the renovation of the construction surveys and indicators complete, the Census Bureau will be able to expand its surveys, remove some reporting burden on state and local governments and builders and owners, and provide more granular detail about construction activity in the United States.
Back To News