Thursday, June 25, 2009

Statistics reveal high likelihood of fraud in Iran election

At New Scientist, an article on a statistical analysis of the Iranian presidential election. It turns out that initial reports of fraud -- i.e. that Ahmadinejad's numbers remained almost constant as the vote was counted -- don't stand up to statistical muster. In fact, such consistency is the statistically likely result given the way the votes were counted.

However, there is much more interesting and convincing statistical evidence of fraud in the election. Opposition candidate Karroubi's numbers violate Benford's Law, which states that for numbers in real-life data, the leading digit follows a specific pattern. For example, the leading digit is 1 about one third of the time. Numbers from three of the other candidates follow Benford's law, but Karroubi has an unexpectedly large number of vote counts beginning with 7. The likelihood of this happening in fair data is only 0.7%, indicating that Karroubi's numbers were tampered with.

No comments:

Post a Comment