CENSUS AND NONSENSE

Chicago Tribune: An effort to use statistical sampling to make up for undercounts and overcounts in the 2000 census has hit a surprising roadblock. Census officials say that a survey conducted last summer to check the quality of the door-to-door canvass was, in itself, too flawed to be reliable.
As a result, the acting director of the Census Bureau, William G. Barron Jr., says the bureau will withhold the estimated count and will use only the actual door-to-door enumeration. That's a big disappointment to those who hoped that adjusted figures based on samples of the total population could be used to make up for the tendency of head counts to miss large numbers of minorities and the urban poor.
So it comes down to this: The United States can't trust the decennial census count, and it can't trust its analysis of where the count went wrong. So it's left to rely on one of two flawed efforts to count how many people live here.
This shouldn't be shrugged off with a we'll-do-better-in-10-years attitude. Census officials have to improve their efforts to get the best actual count possible. There is a great deal at stake.
Legal reasoning: The Census Bureau showed sound legal reasoning in March when it recommended to Commerce Secretary Donald L. Evans that the states could use only an actual count, not one readjusted through statistical sampling, for political redistricting. The Supreme Court ruled in 1999 that the bureau had to use the "actual enumeration," as called for in the Constitution, to draw districts for members of the U. S. House of Representatives.
But the Constitution says nothing about how federal funds, which amount to about $185 billion a year, based on the 1990 census, should be distributed to the states for Medicaid, foster care and other social services. Now that the bureau has decided to use only enumerated, not adjusted, figures for that distribution, cities expect to be shortchanged. The Daley administration expects Chicago will lose more than $200 million in federal money over the decade as a result of the decision.
Momentous as that problem may be, the creation of virtual people through sampling -- that is, intensely scrutinizing certain geographic areas to estimate how many people were missed -- is hardly the only cure for undercounts. The bureau reports encouraging new successes in identifying those who tend most often to be missed. Barron said the bureau estimates the number of uncounted residents in the 2000 census to be about 1.6 million people, or 0.6 percent of the 281 million residents tallied last year. Officials had earlier anticipated that 3.2 million people would be missed.
Improvements like that show that the bureau, with the help of local officials and volunteers, can improve its count. Statistical sampling can be a handy tool for making sure services are provided adequately, but not if the sample has flaws of its own.