Sampling Ruling Endangers Census Long Form
The Census Bureau said a 2000 count with sampling will cost $4 billion with a 0.1 percent undercount, while a count without it will cost $4.8 billion with a 1.9 percent undercount. Because this will cost more, experts are worried the bureau won't have enough money to send the long form to the number of households it usually targets.
TerriAnn Lowenthal of the Coalition to Preserve Census Data, Washington, said money for the long form has been allocated and the forms will be sent to the printers in April. It's unclear, however, if the bureau will have enough money to send the long form to all of the households it wants and whether it will be able to tabulate and publish all the data in a timely manner.
If the sample size shrinks, "the accuracy of the data could be in jeopardy," said Ken Hodges, director of demography at Claritas Inc., Ithaca, NY, a consumer marketing information firm that uses census data for many of its product and service offerings. "The appropriators may decide that they could save some money by sampling one household out of 12 as opposed to one out of six, which means the chances of that set of responses being accurate is just that much less."
If the bureau doesn't have enough money to deliver the long form to the planned households, "then it may not be able to tabulate the data for smaller geographic areas and may not be able to produce census-tract data or produce data only at the city level," Lowenthal said.
Sent to 118 million residential addresses, the 52-question long form is used by DMers, data users and data providers for more detailed demographic information about U.S. consumers. It also lets the Census Bureau produce demographic and housing estimates for areas as small as census tracts. The short form contains only seven questions.
"I've heard that there were some budgetary problems with the long form," said Jan Davis, vice president and general manager at PerformanceData, a list company that buys census data from resellers such as Claritas and Acxiom. "If there is less data like this available, it will be harder to create accurate models."
Lowenthal said it's important that data users continue to be vigilant in ensuring that Congress appropriates the necessary money for the long form. "The only thing that is clear about the Supreme Court ruling is that the 2000 census will cost a lot more."
The court's opinion, passed by a vote of 5-4, said law prohibits sampling to adjust figures for congressional reapportionment among the states. However, the court said sampling is permitted and even required for other purposes, such as drawing lines of federal, state and local legislative districts and distributing federal money. Whether that data will be used in these instances is still being debated -- along with how the money will be appropriated -- but Congress must make a decision before June 15.
Last year, Congress allocated $1.027 billion to pay for the census for fiscal year 1999, which runs Oct. 1 to Sept. 30. The House Appropriations Subcommittee on Commerce, Justice, State and the Judiciary said the money will be used to complete the address list, award a contract to reprint questionnaires and ensure that both the short-form and long-form questionnaires are used.
Christin Tinsowth, a spokeswoman for Rep. Dan Miller, (R-FL), chairman of the census subcommittee, said she hasn't heard anything about the long form. The sampling controversy, however, is already affecting census processes.
"The long form will be administered in 2000," Lowenthal said. "but the process of getting ready for the whole census is at risk because of the continued fight over sampling. If Congress and the administration cannot agree on the final census methods by June 15, then the Census Bureau shuts down and goes home."