Authorities Question Sampling's Local Accuracy
Census Bureau Director Dr. Kenneth Prewitt publicly noted the importance of local accuracy in Census 2000, describing the census as an effort to find approximately 275 million residents "and make certain we know where they are living on April1st…it’s both a huge count, of course – it’s everyone – but it’s also a count that has to identify residency as of that date, because it is residency that determines allocation of congressional seats, state legislative seats and, of course, federal funds."29
Dr. Prewitt’s emphasis on local accuracy seems out of step with the Bureau’s enthusiasm to adjust census numbers by statistical sampling. History and analysis have repeatedly questioned the ability of the Bureau’s statistical adjustment techniques to provide accurate block-level data in the census. The Secretary of Commerce did so in 1990; the Supreme Court did so in 1996; 30 top Bureau officials did so in 1992 31 and again in 1998; 32 noted statisticians and demographers did so in 1998. 33 In 1997, the National Academy of Sciences reported, "With any reasonable sampling scheme that might be used nationally, there will be some levels of aggregation (for example, census blocks) for which the census count will be less precise on average and would arguably not be an improvement over what could be obtained without the use of sampling."34
Former Bureau Director Barbara Bryant summarized the problem when she reported the findings of a group of the Bureau’s high-level statisticians and demographers, writing, "Their work [CAPE committee] suggests that no survey – either the high-quality, well controlled and interviewed 1990 PES of 170,000 households or a larger one – can be used to make a post-census fine tuning of an average undercount as small as 1.6 percent in all types of places, counties and states at a level of accuracy beyond that by which surveys are usually judged."35
The statistical adjustment proposed for Census 2000 is a version of the adjustment plan rejected for Director Bryant’s 1990 census. 36 Many of the failings of the rejected 1990 plan remain unsolved in the present plan.37 Although the Bureau has been working to reduce or eliminate those problems, critical deficiencies remain.38 To be sure, modifications as a result of the 1998 dress rehearsals may yet improve matters. However, Congressional Members of the Census Monitoring Board have seen no evidence to date that suggests a statistical adjustment in 2000 will perform demonstrably better than it would have in 1990.
Therefore, it remains an open question, requiring additional research and analysis, to determine at what geographic level, if any, adjusted counts might yield improved distributive accuracy over unadjusted counts. Before the doubts about local accuracy are resolved, it would be premature to adjust local counts using sampling.