mimic every decennial census operation—and the tests attempt to vary several major factors simultaneously. For example:

• The 2003 National Census Test involved approximately 250,000 households using only mailed questionnaires (no field follow-up was performed); it varied both the wording of the race and Hispanic origin questions as well as different cues to respond to the test by mail, Internet, or telephone.

• The 2005 National Census Test included a variety of coverage probe questions related to residence (see Chapter 6). In addition to those changes, though, the test also included experimental panels where Internet response is encouraged, and it included a panel that received a bilingual English/Spanish questionnaire.

• In addition to the Alternative Questionnaire Experiment (AQE) that was conducted alongside the 2000 census, other major experiments conducted at the same time included the C2SS (a prototype for the long-form-replacement ACS), the Response Mode and Incentive Experiment, and the Social Security Number, Privacy Attitudes, and Notification Experiment.

• The 1998 dress rehearsal that preceded the 2000 census was intended to be a true dress rehearsal, but the political circumstances that made it difficult to finalize the basic design of the 2000 census forced the 1998 “dry run” to be a particularly ambitious test. Staged in three sites, the 1998 rehearsal was actually a test of three broad design choices that varied in the degree to which nonresponse follow-up was conducted (either in full or only for a sample) and whether a postenumeration survey was used to adjust the counts for estimated nonresponse.

At the other extreme of testing in the census hierarchy are the small numbers of cognitive tests to which revised questionnaires are routinely submitted; see Box 8-3. Hunter and de la Puente (2005) tested a version of the “worksheet” approach used in the 2005 census test based on 14 cognitive interviews, conducted in the Washington, DC area in early 2005. Other cognitive tests conducted by the Bureau use similar-sized samples. Hunter (2005) reported on a 2003 test intended to see whether a proposed direct question on cohabitation (for possible inclusion on other surveys and not the census) was “direct, gender-neutral, non-offensive, and generally applicable.” Conclusions were drawn from a set of interviews with 19 people, all of whom were cohabiting at the time of the interview; the sample included both heterosexual and gay and lesbian respondents. Likewise, Hunter and DeMaio (2005) tested revisions to three separate census questions—housing tenure, age (adding a reminder to report babies as age 0 if they were less than 1 year old), and relationship—based on 18 cognitive interviews, the set of which contained