Big Revisions in the Census County Data
Whenever the Census Bureau releases new estimates of population, media outlets pounce on estimates of the fastest and slowest growing places in the nation. It’s an interesting angle, but not the one that interests me most. The reality is that these first-estimates have large errors in them due to incomplete statistics and preliminary inputs. It’s only after several years that the estimates stabilize, and, even then, once we get a new Census, intercensal revisions are very large.
As I showed in a previous post when we got Vintage 2017 state data, Census made large changes to the migration component which resulted in a larger estimate of the foreign-born population. For this post, I won’t focus too much on specific components of change.
But let’s assess how big the changes were. Here’s a map of the sum of the percent changes in population between Vintage 2016 and Vintage 2017, for estimate years 2010–2016.
See a pattern?
Neither do I. While that map doesn’t look quite randomly distributed, it’s hard to pick out really clear regional trends in it.
So let’s look just at counties gaining or losing at least 5,000 person-years between vintage 2016’s 2010–2016 estimates and vintage 2017's.
Here we can see a few trends. First of all, big cities do well for the most part. There’s lots of blue on this map partly because the Census Bureau did raise the aggregate estimate of population a bit, but also partly because they raised it primarily among foreign-born people, and the foreign-born are disproportionately located in cities. Ergo, cities show up strongly.
But how big was the typical revision to a county’s population estimates?
Here I show several indicators. The brown line is the average absolute magnitude of change. 2016 population estimates for the average county were changed, positive or negative, about 0.5 percentage points based on the 2016 estimate. The magnitude of change shrinks as you go back in time to 2010. The aggregate revision, however, i.e. the total revision to national population, was near zero, and positive: the U.S. population trajectory was raised slightly. However, the average county had negative revisions, because the positive revisions came in a small number of more populous counties, while many less populous counties saw downward revisions.
But look at those error bands! Within our 95% confidence interval, you get everything from -2% revision of population, to about 1.8%. It’s not at all unusual for the Census Bureau to revise a county’s population by a whole percentage point or two, in a context where population growth for counties is usually 1% or less. So revisions can easily shift the growth path of a county by the equivalent of multiple years worth of growth.
Are these revisions unusual? It turns out, they are a bit unusual. Here’s the average absolute value of the percentage revision to population for every county in each year and vintage.
In past years, the revisions to the most recent year, for which data is the most volatile, have averaged around 0.2 to 0.25%. In the 2017 vintage, they were double that. Throughout the time series, for data-years equivalently lagged from the vintage year, V2017 had an average change size about 0.2% or 0.3% higher than was typical. This difference shows up across numerous counties and states and doesn’t appear to be single-source to a specific region, but rather can be attributed to changes in Census Bureau methodology, particularly their change in domestic migration data sources for people over age 65, for whom they now use Medicare data.
So these revisions are big; unusually big.
I won’t bore you with tons of details. Instead, I want to pivot to a few specific examples to show what Census revisions for counties can look like.
Case Studies of Census Revisions
I recently wrote an article arguing that Cincinnati is having a far more impressive Rust Belt Renaissance than Pittsburgh is. How do those claims stack up in the new county data?
Here’s Allegheny County’s revision history back to V2011, and also a line showing what my previous post projected for Allegheny’s Vintage 2017 numbers:
So in 2017, Pittsburgh’s population was revised *upwards* in many back years. But the 2016–17 population change was much worse than I had expected. So the revisions were good, but the growth trajectory deteriorated vs. a state-based benchmark. Pittsburgh appears to have lost 100% of its post-2009 growth, having enjoyed a purely cyclical boom as young people faced constraints on homeownership, so deferred to renting and continued education, both of which favor university-dense urban environments like Pittsburgh.
What about Cincinnati (i.e. Hamilton County)?
A similar story is true for Cincinnati: upward revisions through 2016, but then the V2017 figure undershot by projections for 2017 itself.
But there’s a difference between these two graphs too. Pittsburgh’s graph is going down. Cincinnati’s is going up.
Here’s percent change in each county’s population since 2007:
So it looks like the divergence I pointed out was real. Whereas Pittsburgh was enjoying a cyclical boom, Cincinnati’s growth seems to be a bit more robust to the shifting economic cycle.
Let’s look at some other counties. What about San Francisco County?
San Francisco got a very good 2017 revision: a big step up in back years, and continued substantial growth in 2017. In other words, we should all be revising our mental estimates of San Francisco county’s growth upwards.
What about New York City?
Again, we’ve got a big, positive revision for NYC. 2016 population was revised upwards by 77,000 people, which is a huge adjustment. But growth is slowing down. NYC barely eked out growth in 2017, and the population of Brooklyn actually fell, from 2,651,000 in 2016 to 2,649,000. It’s a small change but, still, it suggests the recent years of booming growth in NYC may be about to take a turn.
Indeed, all five NYC counties are in the top 11 counties with over 250,000 people with the largest percent revisions to population. So Census’ revisions seem uniquely beneficial to NYC’s population profile, and yet even so, they show population stagnation in 2017.
These are just a few examples. If you have a question about a specific county or area, let me know and I can graph out the revision history. But for now, it’s sufficient to show that these revisions can sometimes be very substantial.
I’m an Advisor at Demographic Intelligence, the nation’s leading producer of rigorous national- and regional birth and marriage forecasts. I’m also a Research Fellow at the Institute for Family Studies, a Senior Contributor at The Federalist, and I write periodically for Vox’s Big Idea column. I’m a native of Wilmore, Kentucky, a graduate of Transylvania University, and also the George Washington University’s Elliott School. My real job is as an economist at USDA’s Foreign Agricultural Service, where I analyze and forecast cotton market conditions. I’m married to a kickass Kentucky woman named Ruth. I am not paid one penny by anybody for this blog post.
Follow me on Twitter to keep up with what I’m writing and reading. Follow my Medium Collection at In a State of Migration if you want updates when I write new posts. And if you’re writing about migration too, feel free to submit a post to the collection!
Check out my Podcast about the history of American migration.
DISCLAIMER: My posts are not endorsed by and do not in any way represent the opinions of the United States government or any branch, department, agency, or division of it. My writing represents exclusively my own opinions. I did not receive any financial support or remuneration from any party for this research.