The Devil Is in the Digits: Evidence That Iran's Election Was Rigged

http://www.washingtonpost.com/wp-dyn/content/article/2009/06/20/AR2009062000004.html

by noheartanthony • 17 years ago

65 44 17 years ago

washingtonpost.com

jgrahamc2• 17 years ago

For a counterpoint: http://www.jgc.org/blog/2009/06/last-digits-analysis-of-uk-a...

Use more detailed results from the Iranian election and you don't see this effect.

cousin_it4• 17 years ago

Thanks, that comment looks much less fishy than the original article. I wonder whether the Washington Post authors tried to run their test on past known rigged and non-rigged elections. If not, that would be a good idea.

Tichy• 17 years ago

Why does it look less fishy - because it contains charts? Or strange words like "chi-squared" that signify a high level of competence?

mbreese• 17 years ago

This is actually pretty poor statistics. They are basing their entire argument on ~100 provincial results. The deviations they are talking about with this few number of samples isn't likely to be significant.

jgrahamc2• 17 years ago

So how many results would make their result significant?

mattmanser• 17 years ago

I thought that too until I re-read the article and started trying to do the math myself.

They sum it up nicely but it's too easy to miss:

"The probability that a fair election would produce [these number distribution disparities] is less than .005. In other words, a bet that the numbers are clean is a one in two-hundred long shot."

Key thing to take away from the article if you're going to perform election fraud, use a computer to produce your random numbers.

Retric• 17 years ago

The problem with that assessment is the more tests you do the larger your chance of finding something odd.

In isolation you can look at the odds for 17+ and 4-, but two <5, or two > 16 would also look odd. As would a single >20 or <4. And, they started looking at 2 digits but the pattern is only in the last digit so there are looking at 3 groups.

PS: Also how did you calculate the odds? A single number of 17 or more increases the odds of a number of 4 or less. I am not really sure how you would calculate the odds outside of a simulation.

praptak• 17 years ago

"The probability that a fair election would produce [these number distribution disparities] is less than .005."

Careful with that reasoning. Suppose there are four candidates, and the last digits of their total vote counts are 8,7,2,8. Fewer than 1 in 9999 fair elections would produce a vote count with exactly those digits!

Does this mean that the vote was rigged?

Post-factum searching for (perceived) oddities and calculating probabilities of their occurences in uniform distributions shows some misunderstanding of statistics.

WilliamLP2• 17 years ago

> They sum it up nicely but it's too easy to miss:

> "The probability that a fair election would produce [these number distribution disparities] is less than .005. In other words, a bet that the numbers are clean is a one in two-hundred long shot."

Man, reading that makes me roll my eyes and wonder how the average person reading this site has such poor statistical intuition as to mod this up.

It's like rolling a die six times, and arguing that getting the exact numbers I got has an exceptionally low chance of occurring, so the die must be rigged.

alyssumclimbs• 17 years ago

I wonder if the first digits comform to Benford's Law. Basically, the first digits of the vote counts should be 1, 2, 3 much more often than they should be 7, 8, 9

http://en.wikipedia.org/wiki/Benfords_law

brown9-2• 17 years ago

Fivethirtyeight.com took a look at this, and found some oddities: http://www.fivethirtyeight.com/2009/06/karroubis-unlucky-7s....

bcl• 17 years ago

I think this article from 538 does a much better job covering the subject than the original one, who's methodology is questionable.

thras• 17 years ago

Indeed. It's quite non-intuitive, but the naive assumption that the last digits are randomly distributed is simply wrong. Here is a V.I. Arnold essay about Benford's Law: http://www.math.ru.nl/~mueger/arnold.pdf

riffic3• 17 years ago

At this point it doesn't matter if it was rigged or not. The regime has voided the social contract with the people of Iran and is therefore illegitimate.

cousin_it4• 17 years ago

The question whether the election was rigged matters for the same reason that the question of Iraq's WMDs mattered: this information is influencing Western public opinion and government decisions right now. Also, going "it doesn't matter 'cause you're bad anyway" whenever someone calls attention to the substance of your accusations strikes me as a bad faith tactic.

ellyagg2• 17 years ago

Ha. To be blunt, I don't think most folks on Hacker News are thinking about this in good faith. If some other country this community doesn't like a priori, say the US, killed a bunch of its citizens after an election, there'd be zero chance everyone here would be rationalizing away statistical improbabilities in the election results.

> Others have pointed to the surprisingly poor performance of Mehdi Karroubi, another reform candidate, and particularly in his home province of Lorestan, where conservative candidates fared poorly in 2005, but where Ahmadinejad allegedly captured 71 percent of the vote.

If, for example, Broward County had come up 71% for Bush in an election, that would have been considered proof of election fraud around here. In this case, it's rationalized away. Poof.

dant2• 17 years ago

I don't think the argument is that "you're bad anyway". Regardless of who you are, how you got into that situation and whether the protesters have a point, with a protest movement that big and motivated your options are same: crush the protests, re-run the election or form a unity government. If you choose to crush the protest then you give up the "government by consent of the people" type contract that you have with society. I think it's that fact that will drive western opinion more than the actual legitimacy of the result.

kevinpet• 17 years ago

You're assuming that whether the reported totals matched the votes is the only relevant point.

It is not the responsibility of critics to show that an election was fraudulent. It is the responsibility of a government to convince the populace that the election was legitimate. If a large portion of Iran thinks that the election was fraudulent, that's a good enough reason to support their efforts to replace that government.

Al Gore should have gotten his Nobel for preserving the republic by letting the 2000 election die in an orderly fashion rather than for his climate scaremongering.

philwelch4• 17 years ago

Debating the validity of Iran's election results at this juncture is kind of like worrying about whether Caesar is too ambitious after he's already crossed the Rubicon. Shooting peaceful people to death in the streets and cutting off any means for the outside world to learn about these atrocities tells us all we need to know about the Khamenei/Ahmadi regime.

ijuyhgtyu• 17 years ago

<cough>florida<cough>

ellyagg2• 17 years ago

Sorry, where were the killings after Florida again?

DannoHung• 17 years ago

Uh, forgive me if this doesn't make any sense, but why would the numbers of the actual votes be generated fradulently?

Wouldn't it be easiest to say, "Okay, Ahmadi is going to win 57% to 65% in all reporting districts," and so on and so forth and then use those fractions and the known eligible voter numbers to come up with the actual counts?

In which case, shouldn't we be looking at the last digits of the percentage of votes won by district?

chairface• 17 years ago

Well, the thing about percentages in an election is that a lot of them don't _have_ last digits. Further, if the percentages came out to nice numbers like 57% and 65%, we wouldn't even be having this discussion - it would be really, really obvious.

DavidSJ• 17 years ago

There are a number of causes for caution here, such as the possibility of publication bias. There could be 100 other statistical tests that the election passes, but who would publish those boring results?

Another problem is how reasonable the null hypotheses are. Should we really expect Iranian vote counts to fit Benford's Law in a fair election, for example? I don't know.

I have no reason to believe the election was fair, but I think any scientific results reported directly in news media without peer review a week after a very controversial election are highly suspect.

WilliamLP2• 17 years ago

This article is pure shit. Anyone with a reasonable math education should know why, and should know about data mining. That you can get away with such things in major newspapers is simply shameful.

gojomo• 17 years ago

I hate to be a credentialist, but I would have more confidence in this analysis if it came from statisticians rather than political scientists.

caffeine• 17 years ago

The statistics is completely wrong - and the conclusion is (probably) completely right...

The Devil Is in the Digits: Evidence That Iran's Election Was Rigged

Related Stories