Compromise reached on census sampling.

After months of wrangling, the Clinton administration and Republican congressional leaders agreed last week on a compromise plan to prepare for the year 2000 census.

The agreement requires the Bureau of the Census to pursue a dual track. The bureau may experiment with the use of statistical sampling to supplement a traditional head count (SN: 10/11/97, p. 238), but at the same time it must test enumeration without sampling as part of its dress rehearsal, scheduled for early 1998.

Republicans, meanwhile, will pursue an expedited review by the Supreme Court of the legality and constitutionality of using statistical sampling to adjust census numbers.

The measure also establishes an eight-person, bipartisan Census Monitoring Board to oversee preparation and implementation of the 2000 census.

In effect, the compromise delays until 1999 a decision on whether nationwide sampling can be used to supplement and adjust a traditional enumeration. "We've come to an operational truce," says Martha Farnsworth Riche, Census Bureau director. The compromise lets the bureau move forward while judicial review occurs, she adds.

The bureau will test its new procedures, which include sampling, in Sacramento, Calif., and on the Menominee Reservation in Wisconsin. As a contingency plan, it will conduct a more traditional enumeration in a number of rural counties in South Carolina.

Members of Congress have expressed a variety of concerns about making statistical adjustments to census numbers, including the possibility of undermining public confidence in the decennial census and reducing participation. Several have noted the complexity of the statistical approach and the inherent uncertainties and assumptions in the chosen statistical methods.

"I'm concerned about how the sampling is actually implemented," says David W. Murray of the Statistical Assessment Service in Washington, D.C. Enumeration does not catch everyone, he concedes, but with a statistical approach, there is a danger of building in other sorts of errors because of the choice of sampling methodology.

Moreover, such errors are much more difficult to pinpoint and understand than those associated with enumeration, Murray contends. "Is one type of error being replaced by another?"

"The fact is that traditional census operations give us an undercount," says Charles L. Schultze of the Brookings Institution in Washington, D.C. Those most often missed include rural renters, migrant workers, elderly women living alone, minorities, mobile young men, and particularly children.

Sampling offers a substantial improvement in accuracy, Schultze argues. Indeed, there may no longer be a realistic alternative to sampling because of the escalating costs and diminishing returns of conducting a conventional enumeration.

"It's an issue of trust," Riche says. "We want to make the process as transparent as possible."