on narrow grounds that the secretary’s decision was not arbitrary or capricious. This ruling was vacated by a panel of the Second Circuit Court in August 1994, which required further consideration of adjustment on the merits. The states of Oklahoma and Wisconsin appealed to the U.S. Supreme Court, which in March 1996 reversed the Circuit Court decision, allowing the unadjusted census counts to remain the official 1990 census numbers (Wisconsin v. City of New York, 517 U.S. 1, 1996).
We described in Chapter 3 how the evidence of an increased net undercount in the 1990 census (which resulted despite a 33 percent increase in costs compared with 1980) fueled efforts by the Census Bureau to develop a more cost-effective design for the 2000 census. The original 2000 design called for a large 700,000-household survey and matching operation conducted on a schedule that would permit integrating DSE-based population estimates into the census counts in time to produce adjusted state totals for reapportionment of Congress by December 31, 2000. The Integrated Coverage Measurement (ICM) sample was designed to produce reliable estimates for individual states, so that each state’s coverage correction factor would be estimated from the sample for that state and not require the use of sample from other states.
The U.S. Supreme Court precluded the use of sample-based estimation for the reapportionment counts, so the Census Bureau planned the 2000 Accuracy and Coverage Evaluation Program with a 300,000-household survey and completion of DSE-based estimates in time to produce adjusted small-area census counts for legislative redistricting by the mandated deadline of April 1, 2001. The smaller sample size for the A.C.E. was made possible because there was no longer a need to produce direct state estimates; instead, estimates could borrow strength across states. In a May 1999 letter report to the Census Bureau director, our panel supported this decision, although it urged the retention of a minimum sample size in each state that could be used to evaluate the assumption that state effects on coverage correction factors are less important than the effects of such characteristics as age, race, sex, and housing tenure (see