It is Not Whether or Not to Audit Elections, But How, as Explained in an analysis from National Election Data Archive

Bookmark and Share
How many ballots have to be counted to detect vote counting errors that are big enough to change the outcome of an election? With more and more contested elections since 2000, the question is not just academic. According to an analysis by M.I.T. mathematician Ronald Rivest, in an average US House race with a 1% margin between candidates and 440 precinct counts, a 2% audit may only have a 27% chance of uncovering vote count error, while a 20% audit may have a 97% chance of uncovering vote count error.

Read more:http://www.prweb.com/releases/2007/1/prweb497683.htm