tum. This procedure assumes that the match rate, correct enumeration rate, and other rates involved in the dual-systems computation for a poststratum apply at all lower geographic levels or, alternatively, that the true coverage correction factor for a poststratum applies at all lower geographic levels. This assumption—known as the synthetic assumption—is strong.4

Finally, the design of the A.C.E., similar to the design of the 1990 PES and 1980 PEP, was focused on developing good estimates of net undercount. Given the clustered sampling and other features, the A.C.E. design was not well suited for estimating gross coverage errors or categorizing them by type. For example, the A.C.E. was not designed to identify duplicate census enumerations that involve a household with two or more residences in locations outside the search area for matches (e.g., a household with summer and winter residences in different states). In principle, the A.C.E. process would identify half of such duplicate enumerations as correct and half as erroneous (under the category of “other residence”). However, the balancing assumption might not hold, and, in any case, it would not be possible to identify duplicate enumerations as distinct from other kinds of erroneous enumerations in the “other residence” category. This problem became evident when the original March 2001 A.C.E. coverage estimates were subjected to additional evaluation in summer 2001 (see Section 5-D).

In the preceding discussion of dual-systems estimation, we have referred to several assumptions underlying the method. Statistical assumptions need not be precisely true for a method to be useful and, in fact, such assumptions are rarely precisely true. Determining the sensitivity of results to mild failures of the assumption is an important part of evaluation of a method and of research and development leading to improved methods.

4  

The synthetic procedure also assumes that imputed IIs are assigned correctly to poststrata (i.e., that the imputation accurately imputed their basic characteristics) and that late addition IIs resulting from the special unduplication process are correctly assigned to geographic areas (see Zhao, 2003).



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement