Flawed software programming appears to be at fault. Laura Zayatz, chair of the Census Bureau's disclosure-review board, says code designed to add the statistical noise to the subset of older respondents should have offset those changes with opposite adjustments made elsewhere in the data sample. This didn't happen as it should have, so that ages and other attributes were skewed.
Before the data were released in 2003, the Census Bureau's diagnostic tools flagged the problem, but it "didn't seem large enough in the judgment of our analysts to stop the release," says Dr. Groves, the Census Bureau director.
Friday, February 05, 2010
Feds: DUH. Again.
Posted by macbeach at 7:08 PM