Letter to the Editors of USA Today

P.B. Stark
15 January 1999

Dear Editors:

On January 12, you published an opinion piece called "Why fuss over census? Science has the answer." The author claims the 2000 Census can be fixed using sampling, but that partisan politics is muddying the waters.

The argument is not scientific. It (a) makes an analogy to TV ratings, (b) appeals to the authority of the National Academy of Science [NAS], and (c) claims that sampling is just double-checking answers.

(a) Yes, adjusting the census is like TV -- more show business than substance.

(b) Science relies on evidence, not authority. The data in the NAS reports don't support sampling. Neither the NAS nor any other scientific organization has endorsed the administration's current plan.

(c) Adjustment using a sample makes more mistakes than it catches. The Census Bureau tried to adjust the 1980 census. It didn't work. They tried to adjust the 1990 census. It didn't work. The 1990 adjustment would have added about 5 million people, but 3-4 million were errors in the adjustment, not the census. Just one computer bug added a million people.

The 2000 adjustment plan is about the same as the 1990 plan, but with less time to do the work. The sample will be larger, and the mess is likely to be larger too.

I'm a liberal, but as a statistician, I know that the adjustment scheme probably will make the census worse.

Philip B. Stark
Professor of Statistics
University of California, Berkeley