Designing the 2010 Census: First Interim Report (2000)

The Content Reinterview Survey involves a reinterview of a sample of addresses that received a census long form. In previous censuses, the Content Reinterview Survey included a panel for which the reinterview was more intensive, so that the study could be used to measure not only response variance, but also response bias. The survey proposed for 2000 involves only a reinterview that mimics the questions used in the 2000 census, so it can only measure response variance. Measurement of response bias is arguably more important than response variance, as it will provide important information for revising questionnaire wording in future censuses.

Response bias is likely to vary by mode of response, which is important given the large number of response modes, including “Be Counted” forms, the Internet, and, possibly, greater use of the telephone. Bias, unlike variance, does not generally diminish with aggregation. Therefore, the panel proposes that the Census Bureau consider expanding or refocusing the Content Reinterview Survey to measure elements of response bias, particularly for census short-form items. To examine this, “Be Counted ” forms, telephone responses, and Internet responses—which are all short form responses—should be sampled, and at a greater than proportional rate, in the Content Reinterview Survey.

The imputation and editing routines used in previous censuses have received relatively little evaluation in preparing for the 2000 census. This is surprising, given their impact on census data quality. This issue is especially important given the recent advances in techniques for imputing missing values. Data should be collected during the 2000 census to support evaluation of the current editing and imputation procedures and, more importantly, to support research on new editing and imputation methods for the 2010 census. These data could come from administrative records or from expanding the Content Reinterview Survey to include a bias evaluation component, as described above.

Costs of the decennial census have grown in real terms for several censuses (National Research Council, 1995), and two major questions are how much the United States should spend on its decennial censuses and what additional advantages are gained through the use of more or less expensive designs. The only way that such questions can be addressed is to understand in detail the sources of the current costs. Therefore, the Census Bureau needs to collect data on census costs in as much detail as possible, to permit their comprehensive modeling for alternative design options for 2010. In addition to the use of the PAMS/ADAMS system as an input to the master trace sample, the detail should also include the costs of forming the address list, advertising, printing forms, contractors, etc.

The most error prone component of the ACE, which underlies the plans for adjustment of census counts, is probably the matching operation. For the 1990 census, two studies were used to measure the extent of matching error: the comparison of two independent matching groups, and the results of an evaluation follow-up study. As discussed by the National Research Council (1999:Ch. 4), comparison of the matching groups was not a direct measurement of final matching errors, and the evaluation follow-up study was completed so long after the census that responses were obtained for only about 40 percent of the sample. Given the importance of

measuring matching error, the Census Bureau should plan to collect sufficient information so that research on adjustment for the 2010 census can be fully informed on this issue.

In the 1990 census, ethnographers (also referred to as systematic observers) collected information for a heterogeneous collection of 29 small areas on the amount and causes of census undercoverage (Brownrigg and de la Puente, 1993). These case studies were useful in providing hypotheses as to possible ways in which the census might be modified to reduce undercoverage. The panel encourages the Census Bureau to develop similar studies for use in the 2000 census, and to make the plans for these studies available for review to various census advisory groups.

Finally, the panel is concerned that the number of households using the option of the Internet for responding to the census short-form questionnaire will exceed the capacity of the census webpage servers. The Census Bureau should be very optimistic in its assessment of the number of people likely to use this option on the days immediately after the mailing of the census questionnaires and plan accordingly.