Review ordered at Census to ensure accuracy of data

By Carol Morello
Washington Post Staff Writer
Friday, February 5, 2010; 8:45 PM

The head of the Census Bureau said Friday he has ordered a review of the government's efforts to protect the identity of individuals after researchers discovered months ago that years of publicly available data on the elderly contain numerous errors due to the government's use of "masking" techniques.

Robert M. Groves, director of the bureau, said that the programming mistake responsible for the bad data had been corrected going forward, and that it would not affect the 2010 Census. Groves said he is considering a rerelease of the affected statistics, dating from 2000 to 2005.

The review comes as census forms are about to be mailed to every home in the country. A House subcommittee hearing Wednesday will look at how the census is promoting itself in hard-to-count areas, usually poor, immigrant and minority locales.

Errors involving statistics on people 65 and older were found in data typically used by researchers in academia, government and marketing.

In a paper published last month by the National Bureau of Economic Research, three demographers said some of the census statistics could have been off by as much as 15 percent. In one example cited, the census had the national ratio of men to women implausibly increasing dramatically among people in their mid-60s. Census officials have already posted a warning on the data, cautioning researchers that the gender ratio is wrong.

The errors are the result of the extraordinary lengths the Census Bureau goes to maintain the confidentiality of people who fill out the forms.

Some information is made less precise, said Laura Zayatz, chair of the disclosure review board. For example, a $5 million annual income would be listed as more than $200,000. In other cases, identifying characteristics, such as age, are randomly changed, though the overall average remains the same. Both practices are widely accepted and do not affect the overall picture.

The census often will hire an outside contractor to "attack" the statistical file before it is released to see if it anyone's identity can be discerned.

"We have a very sacred burden to protect identities," Groves said. "We have given our pledge that nothing we ever do could identify who a record belongs to. We know if we ever violate that pledge, the credibility of the census would be destroyed, and it would take decades to rebuild."

J. Trent Alexander, a researcher with the Minnesota Population Center and one of the paper's authors, said most researchers look at people within an age range, rather than a specific year, so the errors are unlikely to alter many research conclusions.

View all comments that have been posted about this article.

© 2010 The Washington Post Company