Before the Census Bureau terminated all three Census Advisory Committees, the National Advisory Committee was scheduled to meet on May 2 and 3. So today we are sharing one of the recommendations that we would have offered during the public comment period of that meeting.
One of the key ways the Census Bureau has measured accuracy in the decennial census is through a post enumeration survey or PES. The Department of Commerce Office of the Inspector General (OIG) recently released a report in which it indicated several major problems or issues with the 2020 PES.
While the OIG report noted several challenges and limitations of the 2020 PES, it overlooked one important problem: the method greatly underestimates the net undercount of young children. In the 2020 Census, the PES found a net undercount of 2.8 percent of the population ages 0 to 4 compared to a net undercount of 5.4 as measured by Demographic Analysis (DA).1 In the 2010 Census, the PES estimated an undercount of 0.7 percent for young children compared to 4.6 percent result using DA. In comparing DA and PES for estimating the coverage of young children in the Census, the Census Bureau concludes,
“While both methods have their strengths and limitations, DA is a better approach for assessing the Census count of young children because the estimate comes primarily from birth records, which are considered 100% complete in the United States.”
Thus, the low net undercount estimates for young children produced by the PES method can be seen as inaccuracies or errors.
Census accuracy metrics for young children are particularly important because young children had a much higher net undercount rate than any other age group in the 2020 Census and the net undercount of young children has worsened steadily from 1.4 percent in 1980 to 5.4 percent in 2020.
The problem with the PES method for counting young children is something called “correlation bias”. This basically means the kinds of people likely to be missed in the census are also likely to be missed in the PES. There is a similar problem with the Black population, but the Census Bureau corrects the correlation bias for that demographic group. The method used to correct for the correlation bias in the Black population count is not available for young children, but some demographers have suggest using a different method called the “child/woman ratio.”
Fixing the correlation bias problem in the PES is not simple or easy. In the past, the Census Bureau often asked one of their Advisory Committees for help thinking about a solution for this kind of problem. Now that the Advisory Committees have been disbanded, that is not a possibility.
While the DA method provides reliable estimates for the net coverage rate for young children, we still need a PES that provides accurate data on young children. The PES is the only method that provides data for what are called “components of coverage” such as
- Omissions
- Erroneous enumerations
- Whole-person imputations.
These kinds of detailed data would help identify more specifically the kinds of problems experienced while trying to accurately count young children. The high net undercount of young children could be because of high omissions rates, low erroneous enumeration, not enough whole-persons imputations, or some combination of those three elements. With DA, we only get the net undercount—which shows us that nationally the count of young children was 5.4% lower than it should have been. But we don’t know how many children were missed and how many were included when they should not have been, for example because they were counted at two locations. So, we don’t know if the census missed 10% of all young children but doubled counted nearly 5%, or whether it missed only 6% and double counted less than 1%. We also don’t know how many children were “imputed”, or all their demographic features were added (relationship, sex, age/date of birth, Hispanic origin, and race). While the Bureau did produce tables with these figures for children ages 0 to 4, given the overall inaccuracy of the PES estimates for the coverage of young children, these tables are not thought to be reliable.
A redesigned PES could allow the Bureau to better measure the count of young children.
The Bureau announced in 2023 that it was planning to redesign the PES as recommended by the PES staff and experts on the Base Evaluation and Research Team (BERT). In its response to the Inspector General report (included in the report), the Bureau said that while it disagrees with the Inspector General’s findings, it “will consider the suggestions in the recommendation for future coverage estimation activities. The Census Bureau is committed to transparency and accuracy as demonstrated by the 2020 PES. We are currently researching substantial innovations to the coverage estimation program that will improve the already high degree of transparency and accuracy.” We urge the Bureau to include innovations that would improve the PES ability to measure the count of young children.
We also note that a more accurate PES could allow the Bureau to make corrections for the total state population in the blended base approach to post-census population estimates. The Bureau announced in 2023 that the 2020 PES was inadequate for this purpose but that it would continue to try to improve the PES for 2030 in part with the goal of being able to use a better PES to improve post-census population estimates. This matters for young children because the total state population, as captured in the population estimates, is a key factor in the formula that determines the single largest stream of funding to states for human needs programs, the Federal Medicaid Assistance Percentage or FMAP. All five programs which receive funding under this formula are important resources for serving children: Medicaid, Children’s Health Insurance Program (CHIP), the mandatory portion of childcare funding, foster care and adoption.
- The PES involves taking an intensive sample survey right after census data collection is completed, then matching the sample responses to the Census responses to see who was missed and who was double-counted. The DA method uses current and historical vital statistics and international migration data. ↩︎
