This report includes a high-level review of the issue of the undercount of children age 0-4 in censuses and surveys. It summarizes possible causes for that undercount and highlights areas that the task force was able to investigate to assess the validity of some of our hypotheses. The primary product is a list of recommended research (pages 17 - 19). This executive summary includes a broad set of observations and suggestions for Census Bureau managers; suggestions that we feel could move us in the right direction in addressing this problem in the future.
1. The undercount of children under age five in the decennial census, and in surveys like the American Community Survey (ACS), is real and growing. The task force believes that Demographic Analysis (DA) provides the best measure of this undercount in the 2010 Census at 4.6 percent, nationally. This is not a new problem and has been present in decennial censuses for many decades. The differential undercount of this population across geography and demographics makes this a larger problem for some racial and ethnic groups and some parts of the country.
2. Both DA and Census Coverage Measurement (CCM) are valuable tools to measure coverage and managers should use both sets of results to understand the areas warranting improvement in future censuses. In 2000, DA found that children age 0-4 had higher levels of undercoverage than most populations with the exception of black males age 20-59, a group historically recognized as having high levels of coverage error. These results should have led to efforts to address coverage of young children in the same spirit as efforts to address coverage of black men.
3. The task force found that many of the managers working on the development of methods and the design of experiments and evaluations in 2010 were largely unaware of this undercount problem and especially the degree to which the problem existed in 2000. This may be due in part to a reliance on CCM to identify coverage concerns. As a result, the methods employed in 2010 did not address the issue in ways that might have been possible and the 2010 research and evaluation program provides no formal or even informal assessments of the likely causes. Staff responsible for designing the 2020 evaluation program should embed appropriate evaluations and experiments specific to the quality of the enumeration of young children. After 2020, we should not find ourselves in the position we are today, with limited knowledge of what happened.
4. Census Bureau managers need to understand and communicate the reality of this problem with staff responsible for data collection operations in both the census and in surveys such as the ACS. Staff working on 2020 planning need to ensure that development work this decade includes a more conscious effort to address this problem. Testing in the next few years should reflect a greater understanding of how to reduce this undercount. Ideally, 2020 managers should establish a planning group with this as their focus and possibly with an external advisory group to support this effort. At the very least, there should be someone in the 2020 Research and Planning Office designated to be the point person for this issue.
5. The task force is convinced that there is no single cause for this undercount, so there will be no single solution. Planners should explore multiple avenues to be confident that we can reduce the undercount of young children in 2020 from the levels found in 2010. Demographic and decennial surveys should be a part of the discussion and work with 2020 to research the problem and develop improved methods.
6. There is a strong relationship between observed differences in population estimates for young children and census counts of young children in the largest and most densely populated areas. Minority children in these areas are most at risk of being undercounted. Additional analysis of these correlates of undercoverage can help us to understand possible sources and solutions.
7. The census requires substitution methods to account for households that enumerators cannot interview. Research suggests that areas with the lowest levels of cooperation have higher levels of coverage and nonresponse error. The growing number of hard-to-count households, as evidenced by increases in household substitutions, can contribute to the risk of miscounting young children. Without a plan to reduce the noninterviews and proxies and collect better data for these hardest-toenumerate populations, the 2020 Census will include nonresponse error that will add to the undercount of young children. In particular, research that will document, profile, and target the growing number of “complex households” can set a strong foundation for new methods to improve their enumeration.
8. Additional research using existing 2010 datasets, such as DA, population estimates, the planning database, census control and response files, and CCM, holds promise to provide greater insights into causes and possible solutions. Staff involved in 2020 planning should be mining these data to understand this problem. This work must look below the national level to determine if certain areas, populations, or census operations were more likely to have these errors. Our report makes several specific suggestions. It also notes areas unlikely to be a significant cause (e.g., missed housing units).
9. Administrative records matching with 2010 census data and ethnographic research are other valuable tools that we believe could shed light on the characteristics of these missed young children and their households.
10. The task force believes there could be value in directing outreach and promotion for the 2020 Census to agencies working with parents and young children, especially minority children. It is possible that advertising that highlights the importance of all children being included in the census could have a positive impact.