Despite Efforts, Census Undercount of Young Children Persists

Young children ages 0 to 4, a historically undercounted population in decennial censuses, continued to be undercounted in the 2020 Census despite major efforts by the U.S. Census Bureau to mitigate this problem. 

In fact, the 2020 Census had a larger undercount of young children than every other census since 1970.

In addition to difficulties in counting young children experienced in previous decennial censuses and other Census Bureau surveys, the 2020 Census faced unique circumstances. The COVID-19 pandemic not only delayed collection but also disrupted living arrangements for many families, making the counting of young children even more challenging. 

Today the Census Bureau released estimates of net coverage error from the 2020 Post-Enumeration Survey and 2020 Demographic Analysis.

The Census Bureau has relied on two principal methods to evaluate the coverage of the decennial census.  

One method is the survey-based approach known as the Post-Enumeration Survey (PES), which creates an alternative estimate of the population totally independent of the census by surveying a sample of the population. The survey asks where these people lived on April 1, 2020 and attempts to match the information to the results of the 2020 Census.  

The other method is Demographic Analysis (DA), which uses current and historical birth and death records, data on international migration, and Medicare enrollment records to independently estimate the population. The DA net coverage errors are produced by calculating the percent difference between the 2020 Census counts and the DA population estimates as a percentage of the DA population. 

The DA population estimates for young children are produced using birth records, or birth certificates, from the National Center for Health Statistics. We also account for deaths and international migration for young children, but these components of the estimates are relatively small compared to births.  

Since the birth registration system in the United States is nearly 100% complete, the DA estimates for young children may have less error than the PES estimate for this population. When estimating the number of young children, the PES results may have more error than DA because of correlation bias, wherein some people who are missed in the census are also more likely to be missed in the survey. 

Table 1 features the net coverage errors for young children ages 0 to 4 in the 2010 and 2020 Censuses as calculated by both the PES and DA.  

The DA estimate of net coverage errors for selected age groups from the 1970, 1980, 1990, 2000, 2010, and 2020 censuses show some interesting patterns (Figure 1). In the 1970 Census, all age groups listed had undercounts. But over time, the net coverage improved for most age groups — except for young children.

The coverage rates for young children increased from 1970 to 1980 but have had a steady decrease in each census since that time. Young children had a larger undercount in the 2020 Census than in any of the censuses listed in Figure 1.  

The DA net coverage errors in the 2020 Census for children by single year of age show similar patterns in 2010 and 2020: the youngest ages are undercounted while older children are either overcounted or have a net coverage error close to zero.   

In 2020, the 0-year-old population had a net coverage error of -7.0%, which was the largest undercount among any age (Figure 2).  

In 2010, the 0-year-old population had better net coverage than other young children. However, the census counts for 0-year-olds in 2010 included some babies born after April 1, 2010, which artificially decreased the net coverage error rate. Edit improvements for the 2020 Census reduced errors caused by respondents reporting children born after April 1.

We see an undercount for children until age 10, when there is a slight overcount. This sharp increase in the coverage rate is most likely caused by “age heaping” in the 2020 Census data.  

Age heaping occurs when age is misreported so that certain ages are overreported. This is typically seen for ages ending in “0” or “5”. For example, we also see some indication of age heaping for 15-year-olds.  

Overall, the DA net coverage errors by single year of age show undercounts for most children, with the largest undercounts for the youngest ages.

The Census Bureau formed a task force in 2013 to begin work on improving the coverage of young children in the 2020 Census.  

The focus of the group was to review existing information about this issue. The team developed a set of research recommendations to understand what may contribute to the larger undercount of young children relative to other age groups.  

Next, a research team conducted several of these analyses using data from the 2010 Census, the 2010 Census Coverage Followup operation, the PES, DA, birth records, and demographic surveys. Findings from the research team were later used to inform improvements to 2020 Census operations such as revised questionnaire wording and enumerator training.  

In addition, there were numerous stakeholder groups that conducted research, organized meetings, published materials, and helped promote the need to count all young children in the 2020 Census with their specific audiences.   

In fact, some of these groups organized the “Count All Kids” Complete Count Committee to bring awareness to this issue and help provide information for community organizations that focus on children.   

The Census Bureau stressed the importance of counting young children in the 2020 Census Integrated Communications Campaign. For example, working with an advertising contractor, we developed messages involving young children that were woven throughout the media campaign. Materials displaying young children and encouraging families to count them were printed in 13 languages.

The Census Bureau hosted a national press event on February 22, 2020, that even included an appearance from popular characters the Cat in the Hat from Dr. Seuss and the Count from Sesame Street. More examples are found in 2020 Census Integrated Communications Plan v2.0.

The work from 2013 to 2020 focused largely on: 

• Understanding the characteristics of young children who are missed in the census.
• Improving the count of young children in the 2020 Census.  

This work was very important, and we want to continue the momentum and broaden the scope moving forward well in advance of 2030 Census activities.  

To that end, the Census Bureau has formed a new internal working group made up of subject matter experts focused on improving data for young children in all Census Bureau data products. 

The first objective of our new group is to continue to do research on this topic to better understand why young children have higher undercounts than other age groups.  

Second, we will continue to look for ways to improve data collection for young children in the 2030 Census and demographic surveys. For example, we are including some changes to the instructions, probes, and questions for creating the roster of people living in the household for the 2022 American Community Survey Content Test.

The goal of this project is to ensure that all household members are included in the survey, especially young children, nonrelatives, and people with tenuous ties to the household. We will also explore the use of administrative data to help reach, motivate, and enumerate this population. 

Finally, we are improving data on young children through the Population Estimates Program (PEP). The population estimates are used in many ways, such as to allocate federal funds, as population controls for demographic surveys, and as the denominators for calculating vital rates. In each of these cases, having accurate estimates of children is essential for research and policy decisions and for ensuring children get the resources they need.