This brief article discusses why some data simply cannot be made moreaccurate. One example offered is the 2000 census, which is known to havemissed about 3.3 million people. The Census Bureau had to decidewhether to use statistical sampling to help adjust the count . Anexpert panel advised against doing so, because statistical sampling introducesits own biases and errors. Historian Edward Tenner says such quandriesusually arise when people try to change how a number is formulated. Kolata writes: "The trick, he said, is either to learn to live with knowninaccuracies or to be sure that any new number is really an improvement."
--- Allyn Jackson