Benford's Law and the Iranian Election | Hacker News Reader

Benford's Law and the Iranian Election

jgc.org

38 pointsjgrahamc11 comments

Thanks for writing that up in spite of the unspectacular negative finding.

Benford's law probably does not apply here. The data has to be distributed in a logarithmic fashion (or something close). This would be closer to applying Benford's law to the populations of the certain regions. In any case it just does not make sense to use it without a little justification.

IANA(M/S/EE), but is this applicable? I thought Benford's Law applied to fabricated numbers, e.g. in expense reports. But wouldn't it be too obvious to just invent vote counts like that?

If the vote counts are not fabricated, but results of a manipulated process (i.e. miscounting votes), wouldn't that mean that Benford's law still applies?

There's a professor at Ann Arbor who knows more about this than me :-) He's done work in this area and claims that Benford's Law's distribution for the second digit of vote counts can be significant. My little analysis appears to say "this data matches Benford's Law in the first and second digits"

http://www-personal.umich.edu/~wmebane/

While searching for him I see that he's written a paper very similar to my little experiment about the Iranian election: http://www.umich.edu/~wmebane/note17jun2009.pdf

It's a seven page PDF which reaches no conclusions.

Is it me, or couldn't he just say "Looked at this using Benford's Law. Can't say one way or the other."?

In other words, make it so that the conclusion can be tweeted.

You place no value on someone explaining their methods?

I wasn't looking for a lecture, only an answer.

An "Executive Summary" at the beginning would have saved time for many readers, I believe.

Exactly, I thought the allegations were not that they made up the numbers but that they swapped around one candidate's votes for another.

Very cool law though, it will come in useful in the future.

Benford's law applies to probability distributions which are (roughly) uniform on a logarithmic scale, and which span over several orders of magnitude.

If a manipulated process generated such a distribution then it would satisfy Benford's law.

However, it isn't clear to me apriori why election returns should satisfy Benford's law, so I'm unsure why this test is meaningful to begin with.

Here is another paper about applying Benford's law to the iranian election: http://arxiv.org/pdf/0906.2789v1.pdf

Reply on news.ycombinator.com