A working paper authored by researchers at the U.S. Census Bureau is the latest of a number of analyses, including others done by career Census staff, that highlight the downfalls of asking a citizenship question on the 2020 survey, something the Trump administration has sought to do.
“Our results imply that survey-sourced citizenship data produce significantly lower estimates of the noncitizen share of the population than would be produced from currently available administrative records,” the paper, produced by researchers at the the Census Bureau’s Center for Economic Studies, said.
Commerce Secretary Wilbur Ross, who oversees the census, announced he was adding the question in March to collect data for the Justice Department’s Voting Rights Act enforcement. Numerous Census scientists and other experts had privately advised that using existing administrative data would be more accurate, according to internal documents released after the announcement as part of litigation over the question. Ross was also warned that the adding the question risks an undercount among immigrant populations.
The new paper, titled “Understanding the Quality of Alternative Citizenship Data Sources for the 2020 Census,” is perhaps the most extensive public-facing report backing up that assessment.
While there were some risks of discrepancies in using administrative records, that approach would still produce better quality data than asking a citizenship question on the 2020 Census, the paper suggested.
“The evidence in this paper also suggests that adding a citizenship question to the 2020 Census would lead to lower self-response rates in households potentially containing noncitizens, resulting in higher fieldwork costs and a lower-quality population count,” it said.
Among the analyses the paper conducts to come to this conclusion was a comparison between the response rates for the 2010 American Community Survey — which asked a citizenship question — and the 2010 Census, which did not.
The American Community Survey, taken on a rolling basis, is directed at far fewer households than the decennial census but asks many more questions.
The paper looked at households that received both surveys, and compared the response rates, which were lower for the American Community Survey than they were for the decennial Census. To distinguish whether the relatively lower ACS response rate was being driven by the citizenship question, or some other reason — like the length of the ACS questionnaire — the paper separated out the response rates of the households with at least one noncitizen, who were likely more sensitive to the question. It found that, the difference in response rates was 13.8 percentage points among all-citizen households, and 18.9 percent among households with at least one noncitizen.
The difference grew once the paper’s authors weighted the comparisons.
The paper acknowledged that there are factors other than the citizenship question driving this difference between the two types of households, and explored other models to adjust for that possibility. However the paper’s authors also offered reasons why the nonresponse rate on a 2020 Census with a citizenship question may be greater than what their models were predicting.