2010 Post-Enumeration Estimates Using Census Coverage Measurement

Vincent Mule of the U.S. Census Bureau stressed that the purpose of Census Coverage Measurement (CCM) is not to adjust the census counts, but to estimate the net coverage of the census and the components of census coverage for housing units and people in housing units. CCM does not cover group quarters.

The design involves a sample of the population (the P sample) and an independent sample of housing units in block clusters (the enumeration, or E sample, from the census). The CCM program is under way with the independent listing of addresses in the sample areas (August – December 2009). Following the census, person interviewing is scheduled for August – October 2010, person matching is scheduled for October 2010 – April 2011, and final housing unit matching is scheduled for April – July 2011. The timing is later than 2000, when ACE results were required in time for potential use in adjusting the census counts. Release dates for the 2010 CCM results are still under consideration, but Mule indicated they would be out sometime in 2012.

As in 1980, 1990, and 2000, CCM for the 2010 census will be based on dual system estimation, but use regression modeling instead of post-stratification. The 2010 design also will introduce correlation bias adjustment for males age 18 and above.

Mule explained the basics of dual-system estimation with the familiar 2 by 2 table with cell properties as follows.

• N11 Person in P sample and in E sample
• N12 Person in P sample, but not in E sample
• N21 Person not in P sample, but in E sample
• N22 Person not in either sample

As Mule described it, CCM measures census coverage based on numbers in the first three cells, but is unable to account for persons (or housing) in the N22 cell.

The 2010 CCM will provide more on the components of census coverage. For persons in housing, CCM will identify correct enumerations, erroneous enumerations, census imputations, and omissions. For housing units, CCM will identify correct enumerations, erroneous enumerations, and omissions.

Mule went on to describe the difference between net error and components of error and elaborated on some definitions. For example, erroneous enumerations can be either duplications or persons who should not have been counted at all (e.g., those born after census day or who died before census day). There are no direct estimates of omissions, but because net error is equal to omissions minus erroneous enumerations, omissions are estimated as net error plus erroneous enumerations. The 2010 CCM will identify whether omissions were in a housing unit that was identified.

Plans call for estimates broken out for three major domains: demographic or housing characteristics, geographic areas, and census operations. For example, the intent is to produce census coverage measures (correct, erroneous, imputation, omission) specific to not only race/ethnicity or age groups, but to census operations such as mail response versus nonresponse follow up. Similarly, housing coverage will be classified by characteristics such as occupancy status; tenure; type of structure; and operational categories such as address canvassing, LUCA, or field verification.

For sub-state areas, estimates of net error would be released subject to statistical criteria not yet defined. For persons in housing who should have been counted elsewhere, an additional table has been proposed that will identify the correct residence as being in the same county, different county in same state, or a different state. Another priority is to determine ways to reduce nonsampling error in the 2010 CCM. One possibility is to reduce the CCM sample size—reducing nonsampling error at the expense of an increase in sampling error.

A lively discussion followed. Asked how big the CCM sample size was, Mule recalled it was about 300,000 housing units and is planned to be the same for 2010 (unless reduced as described above). John Thompson, who was with the U.S. Census Bureau at the time, recalled the 1990 sample as about 150,000.

Asked if the CCM results would contribute to improvements in the 2020 census, Mule described several applications related to determining the impact of operational issues and learning more about duplications.