By the census of 1990, the difficulties of physically counting all of the increasingly mobile American population had led to soaring budgetary costs and reduced accuracy. Blacks and other minorities were substantially undercounted relative to whites, and renters relative to homeowners. Armed with recommendations from three panels of the National Academy of Sciences and a blue-ribbon committee of the American Statistical Association, the Census Bureau decided, in designing the 2000 census, to augment the traditional physical counting of households with scientific sampling techniques. Carefully used, such an approach can produce both a reduction in costs and an improvement in accuracy.
Citing various objections, and fearful that a correction of the undercount among minorities and the poor would weaken their political base, the Republican majority in Congress has mounted a massive attack against the Census Bureau's plans. (Republicans are not inherently more partisan; if the electoral stakes had been reversed Democrats would be leading the charge against sampling.) The Census Bureau has, in effect, been ordered to plan two distinct types of census, one with and one without sampling. Under duress it is conducting a dual dress rehearsal, testing a traditional census in one location and a census with sampling in another. An evaluation of the results won't be available until late this year, and in any event, the Republicans have said they don't consider the dress rehearsal as indicative of whether sampling will work.
The delays and the two-track planning approach are beginning to threaten a potential fiasco. Conducting a decennial census is a massive enterprise, perhaps the biggest and most complex task the civilian arm of the federal government has to undertake. In 1990 more than 500,000 people—most of them temporary help—were engaged at one time or another in conducting the census. Moreover the decennial census is a one-time venture. It cannot be carried out on a "learn-as-you-go" basis, but must be meticulously planned in advance in all its details. Thousands of temporary supervisors have to be hired and taught how to supervise and administer training to scores of thousands of temporary enumerators. Tens of thousands of pages of training manuals have to be prepared and checked. The Census Bureau should be conducting dress rehearsals not to determine what kind of a census to undertake but to discover operational bugs in a census plan that has already been devised.
Most important, a census with sampling is not simply a traditional census with sampling added but a very different thing. To take only one example, if you are going to use sampling to complete and improve the enumerators' count, you can leave out a number of expensive operations such as the 1990 programs that tried—not always successfully—to reach such hard-to-find groups as parolees and probationers. Without the use of sampling to complete the count, a range of special programs to track down the hard-to-find must be designed and integrated into the census process.
Time is running out. The whole massive enterprise must be going full blast in less than two years. However capable, the Census Bureau cannot continue trying to prepare two different enterprises not knowing which one will be chosen.
It would be a major mistake for Congress to force the 2000 census to be carried out without the aid of sampling. According to the Census Bureau, that would add some $675 million to $800 million to budgetary costs. At best it would perpetuate the inaccuracies and inequities of the 1990 census, and more likely make them larger. There is only one thing that would be worse, and that is continuing to postpone a decision. The Census Bureau is coming perilously close to the point where it cannot carry out either kind of census in a responsible manner. And so, a word of advice to Congress and the administration: Settle this issue now.