STANDARD DEVIATIONS: The First Assessment of 2020 Census Accuracy

By William P. O’Hare

Standard Deviations blog posts represent the views of the author/organization, but not necessarily those of the Census Project.

The Census Bureau’s release of the 2020 Census apportionment counts on April 26, 2020, provides the first opportunity to compare the Census counts to the estimates from Demographic Analysis. Demographic Analysis (DA) estimates are prepared by Census Bureau staff to provide an estimate of the population they expect to find in the Census, and they have been used to assess Census accuracy for more than half a century. The DA estimates are based largely on data from birth and death certificates along with Medicare data and data on immigration from abroad. The DA estimates are widely believed to be very accurate for data at the national level.

The Middle Series of the DA estimates for the total population released on December 15, 2020, was 332,601,000 (U.S. Census Bureau 2020) and the Census count of the resident population in the 2020 Census was 331,449,281 (U.S. Census Bureau 2021).   Thus, the Census count was 1,151,719 below the Middle Series DA estimate which results in a net undercount rate of 0.3 percent.  

Keep in mind this is a NET undercount figure. It does NOT reflect the number or rate of people missed in the Census. The NET undercount is basically a balance between those missed and those counted more than once (O’Hare et al. 2020). Data on omissions and erroneous enumerations in the 2020 Census won’t be available until data from the Census Bureau’s Post-Enumeration Survey are released (probably in 2022).

Figure 1 provides the 2020 Census coverage estimate along with similar Demographic Analysis assessments from 1950 to 2010.

While the net undercount of 0.3 percent in the 2020 Census, is a slight increase from the 2010 Census coverage rate, the net undercount of 0.3 percent is not very different from the coverage rates in the 2000 or 2010, and 2020 Census coverage looks much better than the coverage rates of censuses of from 1950 to 1990. By this measure, it appears the 2020 Census is not quite as good as the 2000 and 2010 Census, but it is not the train wreck some folks predicted. However, more analysis is needed to provide a solid assessment of the 2020 Census quality.

Despite this relatively good news about the overall accuracy of the 2020 Census, it is important to remember what is really important about Census accuracy is differential accuracy: that is the relative accuracy of census counts among subgroups and across places. The overall accuracy for the total population at the national level can mask important differences among groups. As shown in Figure 1, the overall coverage rate in the 2010 Census was quite good, but that masks the fact that the net undercount rate for the black population was 2.5 percent and it was 1.5 percent for Hispanics (O’Hare 2019). For young children (ages 0 to 4) the net undercount rate in the 2010 Census was 4.6 percent and for Black and Hispanic young children, the net undercount rates were in the neighborhood of 6 or 7 percent (O’Hare 2015).

The analysis presented here suggests that the 2020 Census coverage was relatively good in terms of the overall national population, but the real assessment of census accuracy will be seen when data become available to assess the coverage of subgroups.