the advance preparation of a master list of addresses;
the mailout of questionnaires to all addresses on the list with the request that each household complete the questionnaire and return it by mail;
the follow-up by enumerators for questionnaires that are not returned and to obtain answers to missing items (some of this follow-up is done by telephone);
a series of checks to strive for more complete coverage of the population;
computerized processing of the questionnaires and tabulation of the results; and
a program to evaluate the quality of the enumeration.
With regard to content, the U.S. census has almost from the very beginning obtained added information about the population beyond the absolute minimum required for reapportionment and redistricting. Since 1940, the census has also obtained information on the nation's housing stock. In the most recent census in 1990, about 104 million questionnaires were mailed out (or delivered by census enumerators).1 Five of every six of those questionnaires ("short forms") asked households to respond to 6 questions for each person in the household and 7 questions on the housing unit. The remaining one-sixth of the questionnaires ("long forms") asked households to respond to an additional 35 questions on the household members and 23 questions on the housing unit (see Appendix A).
In other words, the 1990 census, like every census since 1940, included a sample survey as part of the complete enumeration. The 1940 census was the first to have enumerators ask questions of a sample of the population; the 1960 census was the first to have the Postal Service deliver (unaddressed) questionnaires and initiated the use of separate, printed short and long forms (the latter containing additional items asked of a sample of households).
In many respects, the 1990 census was a very successful operation, particularly considering the difficulties inherent in attempting to enumerate a large and diverse population. It was, however, expensive—in both total dollars ($2.6 billion) and costs per housing unit ($25). Moreover, these costs exceeded the costs of previous censuses, even allowing for inflation, decline in mail response rates, and growth in the number of households. Like previous censuses, the 1990 census also had errors. Most prominent and politically sensitive—because of the implications for the distribution of legislative seats and funding for government programs—were differential coverage errors, in which some population groups and geographic areas were counted less thoroughly than others. Despite the increased amount of money spent on the 1990 census, there was a somewhat higher net undercount of the total population than in 1980, as measured by the technique of demographic analysis (1.8 percent compared with 1.2 percent), and a somewhat larger difference between the net undercount rates for minorities and whites.
Following the 1990 census there was growing concern that census costs