Census Sampling? Yes.
Everett M. Ehrlich
The Washington Post
October 13, 1998

Republicans in Congress think the Clinton administration should pull the plug on census sampling because it is inconsistent with an "actual enumeration" of the population, as required by the Constitution every 10 years.

But sampling is not a substitute for the traditional door-to-door head count that takes place in every census, including the one being planned for the year 2000. Rather, it is a supplemental strategy that is needed, according to the National Academy of Sciences, to eliminate the massive undercount of several population segments that plagued the 1990 census.

More than 8 million Americans were missed in that census, a number that is expected to grow substantially unless modern statistical methods are used as part of the 2000 census methodology. Without sampling, the 2000 census cannot produce an "actual" ö much less an accurate ö count of the population.

Moreover, sampling does not increase the level of estimation in the census so much as make it scientific. The traditional census strategy endorsed by congressional Republicans would obtain information on 10 million or so Americans through "last resort" techniques ö such as asking your neighbor what they know about you taking the average number of people in the house on the left and the house on the right. That's what happens in a traditional, non-sampling census in the final days of the count in order to include as many households as possible.

Is everybody comfortable with that? Are "last resort" estimates better than the well-developed, objective and widely endorsed sampling plan now at issue?

The truth is that the fight over the census is not a fight between Republicans and Democrats ö It's a fight between Republicans and scientists. Opponents of the sampling plan-which relies on the type of state-of-the-art techniques we would expect from any private-sector provider ö are the descendants of Galileo's church prosecutors, who found competent science politically objectionable.

Still, the Republicans have a legal right to challenge the Census Bureau's sampling plan, and on Nov. 30 the Supreme Court will hear arguments on whether their position is correct. Let's suppose they win. What is the Republican plan to eliminate the undercount that disproportionately affects minority communities? Several Republicans, including Speaker Gingrich, have suggested that we switch to a census strategy based on "administrative records" drivers license lists, Medicaid and welfare rolls, parole records and the like. Several other countries use administrative records as part of their censuses and as a means of counting the hard-to-count segments of their populations.

Leaving aside the obvious contradiction with the Republican constitutional argument ö adding up names on lists is hardly consistent with the definition of "actual enumeration" of bodies they're pushing the courts to adopt ö I believe the administrative records idea is good news. For one thing, it reflects a realization that the enumeration only" position that most sampling opponents propound hasn't worked, won't work and will never work ö not in a country where the population has grown to nearly 300 million.

The problem with administrative records is that they are inconsistent across state lines, they're routinely out of date and incomplete, and they raise serious privacy concerns. For example, while 90 percent of U.S. adults have drivers licenses with accurate names and addresses, the last 10 percent ö people who have moved, used a different name or have no license ö are usually the same people we can't find in the census, either. And records such as Medicaid not only have inconsistent names but omit such legally required information as race and age.

But that doesn't mean the federal government shouldn't work with state and local governments over the next 18 months ö before the census actually begins April 1, 2000 ö to improve the accuracy and thoroughness of administrative records. We should support any and every credible plan to include the millions of people who would otherwise go uncounted, be it sampling or administrative records. But until we can rely effectively on such records ö and the National Academy of Sciences has concluded it will take at least another 10 years to do so ö then the Census Bureau should be allowed, assuming five Supreme Court Justices agree, to proceed with its current plan to combine the traditional head count with modern statistical methods.

The alternative is another failed census at taxpayer expense.

Everett M. Ehrlich is a presidential member of the bipartisan U.S. Census Monitoring Board and served as undersecretary of commerce for economic affairs from 1993 to 1997.

[back]