form and decide that they do not need to look at the instructions. Particularly if instructions appear long and complex, they may determine that they can figure out the questions on their own—and on their own terms.
Even for respondents who do read the instructions, there is limited space on the form to provide residence information and instructions, and not all concepts can be adequately explained in that space. If there is an easy mechanism for respondents to obtain help on the question through other means (e.g., looking at a Web site or calling a help center), some may take advantage of those options. But most will not: if the provided instructions do not address their own situation, they will answer as best they can—which may not be strictly correct, by the Census Bureau standards.
The basic consequence of difficulties with census residence rules—either in their definition or in their interpretation by census respondents—is spotty census coverage. That is, some people will be omitted from the census entirely, while others will be counted multiple times. Others may be counted only once, but in the wrong place.
Since the 1940s the Census Bureau has published evaluations of the census, showing that the census has undercounted several groups. Though census coverage was always of academic interest, undercount or overcount in the census was not perceived as a major political issue until the U.S. Supreme Court’s 1962 ruling in Baker v. Carr reinforced the “one person, one vote” principle. In the altered political landscape that followed, with its increased attention to strict mathematical equality in legislative districts and judicial invalidation of districting plans with even tiny amounts of variability, the “exactness” of the census count became ever more important and contentious.
Planning for the 1980 and 1990 censuses featured lengthy debates over the prospective undercount of certain groups, particularly in urban areas and among minority groups.15 Anticipating undercount and other coverage problems in 2000, the Census Bureau put in place an Accuracy and Coverage Evaluation (A.C.E.) Program. However, the A.C.E. analysis ultimately spotlighted an unexpected problem: compared with a separate analysis using demographic analysis, the A.C.E. suggested an overall census overcount driven by an esti-