Census Undercounted Some Groups, Overcounted Others
Antiquated counting methods lead to misallocation of resources.
NPR (“The 2020 census had big undercounts of Black people, Latinos and Native Americans“):
The 2020 census continued a longstanding trend of undercounting Black people, Latinos and Native Americans, while overcounting people who identified as white and not Latino, according to estimates from a report the U.S. Census Bureau released Thursday.
Latinos — with a net undercount rate of 4.99% — were left out of the 2020 census at more than three times the rate of a decade earlier.
Among Native Americans living on reservations (5.64%) and Black people (3.30%), the net undercount rates were numerically higher but not statistically different from the 2010 rates.
People who identified as white and not Latino were overcounted at a net rate of 1.64%, almost double the rate in 2010. Asian Americans were also overcounted (2.62%). The bureau said based on its estimates, it’s unclear how well the 2020 tally counted Pacific Islanders.
The long-awaited findings came from a follow-up survey the bureau conducted to measure the accuracy of the latest head count of people living in the U.S., which is used to redistribute political representation and federal funding across the country for the next 10 years.
Other estimates the bureau released on Thursday revealed that the most recent census followed another long-running trend of undercounting young children under age 5.
Despite following long-term trends, the 2020 Census was conducted under highly unusual conditions:
While the bureau’s stated goal is to “count everyone once, only once, and in the right place,” miscounts have come with every census. Some people are counted more than once at different addresses, driving overcounts, while U.S. residents missing from the census fuel undercounting.
Disruptions from the coronavirus pandemic and interference by former President Donald Trump’s administration raised alarms about the increased risk of the once-a-decade tally missing swaths of the country’s population. COVID-19 also caused multiple delays to the bureau’s Post-Enumeration Survey that’s used to determine how accurate the census results are and inform planning for the next national count in 2030.
During the news conference announcing the follow-up survey results, Census Bureau Director Robert Santos — who, before becoming the agency’s head, told Bloomberg CityLab that he believed the census was “being sabotaged” during the Trump administration to produce results that benefit Republicans — acknowledged “an unprecedented set of challenges” facing the bureau over the last couple of years.
“Many of you, including myself, voiced concerns. How could anyone not be concerned? These findings will put some of those concerns to rest and leave others for further exploration,” Santos, a Biden administration appointee, said during the news conference announcing the follow-up survey results.
The bureau said previously that it believes the census results are “fit to use” for reallocating each state’s share of congressional seats and Electoral College votes, as well as redrawing voting districts.
What’s the problem then? That’s what the Census is for, no?
Census numbers are also used to guide the distribution of an estimated $1.5 trillion each year in federal money to communities for health care, education, transportation and other public services.
Oh, that’s right.
Some tribal, state and local officials are considering ways of challenging the results for potential corrections that would be factored into future funding decisions.
The report the bureau released on Thursday only provided a national-level look at the count’s accuracy, and the agency says it’s planning to release state-level metrics this summer.
“There are a lot more states for us to check and review and look through,” said Timothy Kennel, assistant division chief for statistical methods, during a webinar before Thursday’s release.
So, leaving aside interference from Trump, this whole thing is bizarre, right? Even though the Constitution specifies a decennial count for the purposes of allocating House seats, and by extension the Electoral College (each state gets an Elector for its two Senators and one for each of its Representatives) the Census Bureau is constantly sampling the population and putting out revised information. Indeed, it’s so transparent at this point that a simple Google search will give you the revised information without even clicking a link.
We do all of this with statistical sampling techniques that have been perfected over more than a century. Yet, the Supreme Court has ruled that it is impermissible to do that for the actual apportionment count.
Because I’m getting old, I thought that controversy was over the 2010 Census and I spent quite a while trying to unearth my takes on the matter here at OTB. I was unsuccessful because the dispute was over the 2000 Census and the SCOTUS ruling in question came out three years before I launched the blog.
The Supreme Court yesterday rejected the federal government’s plan for using a controversial counting method to estimate portions of the nation’s population in the 2000 Census, ruling in a case that carries enormous political and economic consequences for communities around the country.
By a 5 to 4 vote, the justices said federal law prevents the Clinton administration from supplementing the Census Bureau’s traditional procedure of trying to reach every household with statistical estimates that would be used to determine the nation’s population and divide seats in Congress among the states.
But beyond the crucial apportionment purpose of the census, the court did not foreclose allowing “statistical sampling” for other important purposes, such as the drawing of political boundaries within each state and the allocation of federal funds for everything from road construction to housing for the poor. And the Clinton administration made clear yesterday it intends to move forward with plans to provide population estimates that states could use for everything other than determining how many congressional seats each state would receive.
I honestly don’t remember my contemporaneous take (perhaps my co-blogger and then-teaching colleague Steven Taylor does) but I suspect I was suspicious of Clinton’s intentions. Regardless, in hindsight, it seems clear that, to the extent we want to make important decisions based on where people actually live, statistical sampling is far superior to trying to get an actual count of people. We’ve known for a long time that minority and poor subgroups are harder to count directly. And, for obvious reasons, illegal aliens, the homeless, and other marginalized groups are next to impossible to count by traditional methods.
When we know there was an deliberate effort by the administration to undercoumt minorities, why would we leave this aside?
@Stormy Dragon: Because it’s a one-off problem that we’ve already discussed while it was ongoing and the Census Bureau has assured us had no meaningful impact on apportionment. What’s more important is the systemic problem that recurs every decennial.
@James Joyner:
Do you really believe that?
Deliberate attempts to skew the census results for partisan advantage in redistricting is a FAR bigger problem than noise unintentionally introduce by poor statistical controls.
@James Joyner: “Because it’s a one-off problem that we’ve already discussed while it was ongoing ”
Sure, we discussed it. We discussed it and discussed it. And Trump and Wilbur Ross kept doing it, and there was never any penalty for their obvious criminal behavior, not least of all perjury. So if there’s a Republican in the White House in 2030, they’ll do it again.
There’s that old meme of a dog sitting at a sidewalk cafe, fire blazing all around him and he says “This is fine”. Always makes me think of Dr. Joyner.
@Stormy Dragon: @becca: @wr: The report doesn’t disaggregate the effects of COVID and Trump, so it’s hard to say how much he mattered. Further, I’d guess the general atmosphere he created around deportation contributed more to a Latino undercount than whatever thumb he had on the scale with the Census.
Regardless, the story remains the same: the Census has an incredibly good picture of US demographics but we’re unable to use that picture because of an antiquated way of doing the count that matters for apportionment.
@becca:
He’s a proud example of dedication to the status quo as an end in itself that we’ve come to expect from years of military training.
@Stormy Dragon: Just to be precise–JJ has years of being a Government Civilian–who are far and away invested in the status quo than uniformed military members. His bio says he separated from the Army as a Junior officer.
The Military, especially the higher up one ascends–is mostly about adapting the Force or Organization for the current reality and setting it up for the future. Unfortunately for many Military leaders (who will be in place for only a few years before moving to the next assignment) their plans are often slow-rolled and squashed by the Government civilians that work for them. I often refer to Government Civilian middle-management as “The Frozen Tundra” to new ideas.
@Jim Brown 32: But I’m a civilian at a PME institution and a policy wonk. Our civilian faculty are both the institutional memory (the uniforms rarely stay more than three years and most leave after one or two) that can explain why things are the way they are and also really big change agents. We pushed toward more China during GWOT, well before Obsma’s “pivot,” and toward Great Power competition well before Mattis.
As to this issue, I’m arguing AGAINST the way we’ve always done it. Census professionals have fantastic data based on the actual count and supplemented with sampling. I’m in favor of using that. Which actually insulates against subversion attempts like Trump’s.
@James Joyner: Futzing with the sampling data is likely way easier than futzing with the raw counting. Getting sampling right, without bias, is very hard.
The only reason the sampling isn’t futzed with is the Trump administration didn’t bother — either because it doesn’t matter, or because it’s more pleasing to rub “socialists’”collective noses in open partisan power plays.
@James Joyner: To be clear, what frustrated me about Government Civilians was also the very thing that made them valuable so certainly no disrespect.
They do, as you say, provide institutional memory and enduring subject matter expertise while the green suites are doing all the ancillary training, PME, and deployments—Perfect for well defined and understood problem sets. For ill-defined problems and emerging requirements? Frozen Tundra 😉
@Jim Brown 32: There’s a lot of deadwood in the civil service, alas. Not that there isn’t in the ranks of the O-5s who will never command and are just putting in their time to get to 27 years. There’s a lot of brainpower and creative energy brought in by the rotating uniformed professionals.
The problem is that being a great Infantry officer, fighter pilot, or whathaveyou often doesn’t translate to utility as long-range planning or other staff functions. And, by the time they acclimate to the building enough for their downrange experience to be useful, they tend to rotate back to the operating forces.
For the last decade or so, I’ve thought the obvious answer is something like the German general staff system, where officers are assigned based on their aptitudes and utility rather than to check a complex series of boxes. But that comes with its own challenges.