In order to benefit historically undercounted groups, the census must reduce racial and ethnic differential undercounts among local geographic areas. The census must answer two questions: "How many people are there?" and, "Where are they?" Answering the first without answering the second is insufficient.
This is why. Political power and money are distributed according to geographic or political areas Ė not demographic groups. If the census determines how many minorities live in a state, but not how many live on a city block, the people in that block still wonít receive a fair share of representation or revenue. Since administrative records can add specific people to specific blocks, they have a greater potential to reduce the racial, ethnic and geographic population coverage differentials than any statistical adjustment. Statistical adjustment might answer, "How many?" However, it canít accurately answer, "Where?"
One reason is that undercounts are spread unevenly over geographic areas, while statistical adjustments tend to be spread evenly over geographic areas. That is, for many small geographic areas the undercount is minimal or nonexistent, or there may even be an overcount. Other areas have large undercount errors, say on the order of 20 percent. Between these two extremes are some areas with moderate undercounts.
The Census Bureauís plan to use statistical adjustment does not adequately focus on specific areas. Instead of locating people, statistical adjustments relocate people. Adjusting the census by a post-enumeration survey takes a concentrated population and dilutes it throughout a state, until a specific community of common race, income and location is spread across several communities that share similar characteristics.
By way of illustration, consider the following simplified situation. Assume there are ten geographic areas of equal population comprised of members of one single post-strata, such as black males in rental housing. Further assume that nine of these areas have zero (or close to zero) undercount and the tenth area has a 20 percent undercount. Obviously, the tenth area is of greatest concern to the census.
Now assume the statistical survey estimates the undercount for these combined areas is 2 percent. If the undercount is distributed by the Bureauís statistical method, each of the nine accurately-counted areas is increased by 2 percent, resulting in 102 percent of actual population. The tenth area is also increased by 2 percent, reaching only about 82 percent of its actual count.
Little is done to minimize coverage differentials among geographic areas. While the population total for the ten areas is accurate, the people in the undercounted tenth area remain unidentified, underserved and underrepresented.