The state of New York was estimated to have increased approximately 200,000. The actual increase was nearly one million. The Census Bureau's nearly 800,000 error is equal to the population of the states second through fourth largest cities (Buffalo, Rochester and Yonkers).

Pennsylvania's actual increase was more than three times the Census estimate.

States that were estimated to have lost population were found instead to have gained. For example, Connecticut's estimated 9,000 loss was instead a nearly 120,000 gain.

The Florida error of nearly 700,000 is more than the combined population of Miami and Tampa combined.

Between censuses, the Census Bureau uses a sophisticated sampling technique to produce annual estimates of state population and monthly estimates of the national total. If such methods had advanced to the point of reliability, the rather substantial errors apparent in the 2000 estimates would not have occurred.

The issue of the Census Bureau's erroneous projections is important in light of proposals to use statistical analysis and sampling to adjust the census counts. Under a US Supreme Court decision, such adjustment would be permitted for all uses except the apportionment of Congressional seats between the states. This means, for example, that the drawing of Congressional district borders within a state could be based upon adjusted figures. There are two fundamental problems with statistical adjustment of the Census figures. First; the errors described above are evidence that the state of the art is not sufficiently advanced, and particularly at the Census Bureau. Second; statistical adjustment introduces the potential of political manipulation of the Census.

The experience of another US government sampling program earlier in 2000 is instructive. The Department of Agriculture's National Resources Inventory reported large expansions of urbanization and losses of farmland in a number of states, most notably Pennsylvania, Texas, Virginia and Georgia. Administration officials, eager to make points to support their hysteria about urban sprawl, rushed to publicize the results. There was only one problem --- the data, which was based upon statistical sampling was wrong. The author and Dr. Ronald D. Utt had previously notified the Department of Agriculture of fundamental discrepancies between the statistically sampled NRI and the Agricultural Census for the same year. After some press attention had been obtained, the Department of Agriculture withdrew the numbers announcing a "computer program error."^{(2) There may have indeed been a computer program error. But it is not inconceivable that bureaucrats might have been rushed into premature press releases by officials interested in advancing a particular political agenda.}

Finally, the nation witnessed first hand the problems with statistical sampling on election night, when premature and erroneous projections were made and withdrawn. The last thing the nation needs is such a fiasco applied to its decennial census.

¹ The Census Bureau has not issued 2000 estimates for states and the District of Columbia. This paper uses the 1999 estimates and produces a 2000 estimate based upon the rate of Census Bureau estimated change from 1990 to 1999, scaled to the national estimate.

² Heritage Foundation Backgrounder #1368