Census undercount not as bad as expected: analysis

An analysis conducted by Washington, D.C.-based nonprofit the Urban Institute found that the 2020 census was not undercounted as severely as was initially thought, though it did determine that count accuracy varied in some states and demographics.

In order to determine the accuracy of the census, the Urban Institute created a microsimulation model of the 2020 count by projecting U.S. population as it was on Census Day. By running various “what if” scenarios through its model and comparing the results with the actual count that was reported by the census, the research group determined in its report that the U.S. census had a net undercount of 0.5 percent.

Overall, the census likely had a net average overcount of 3.6 percent and a net average undercount of 4.1 percent.

The Urban Institute noted that this was different from the 2010 census, which had nearly perfect net accuracy.

The organization also acknowledged that no census is perfectly accurate, and the 2020 census had numerous challenges affecting it, from the raging COVID-19 pandemic to the census’s politicization.

In states such as Alaska, Louisiana, Mississippi, New York and Texas, the total true populations were likely undercounted by more than 1 percent. States such as Iowa, Minnesota, New Hampshire and Wisconsin were all likely overcounted by more than 0.5 percent, according to the research group.

“For the next decade, such differences matter for these states. Mississippi and Texas residents will receive less of their fair share of federal funding for infrastructure, health care, and children’s programs. In contrast, Minnesota residents will receive more,” the Urban Institute said.

Among Black and Hispanic people, the true total population was likely undercounted by more than 2 percent in both groups, according to the Urban Institute’s model. Among major U.S. ethnic groups, white U.S. residents were the only demographic that was overcounted — roughly 0.40 percent.

The nonprofit group said innovations were needed to determine the quality of future censuses, pointing to its own microsimulation as a possible example.

“Counting the US population every decade is an extraordinary effort. Years of research and planning go into its execution on Census Day on April 1st at the start of each new decade,” the Urban Institute wrote in its report.

“Typically, the decennial census happens in predictable environments, and plans made by the US Census Bureau proceed accordingly. This was not the case in 2020. The 2020 Census was conducted during a pandemic and amidst politicization of its scientific work—threats to its execution not previously encountered.”