# Iranian elections and hand counting

I haven't been closely tracking the discussions over whether the Iranian elections were subject to fraud or not, but I am able to comment on one issue I saw raised: the speed of the results. For instance, here's Karim Sadjadpour (þ "Mrs. Polly" on Obsidian Wings):

"There are a lot of signs there were major improprieties. First of all there were 40 million votes cast and just two hours after the polls had closed they announced Ahmadinejad's victory: and these votes are hand counted in Iran...

It's easy to run the numbers here. Hand counting paper ballots via the sort-and-stack method takes about 6 seconds per ballot/contest pair. This means a team (1-3 people, typically) can do about 10 ballots/min or 600 ballots/hour. If we have two hours, then each team can count 1200 ballots, so we need about 33,000 teams. This is a lot of people but isn't totally outside the realm of of possibility. If you started earlier, such as by having ballots returned in mid-day, then you could obviously get away with fewer teams.

Moreover, you don't need to count all the ballots in order to have a high level of confidence in the result—elections are routinely called with only partial counts available. Loosely speaking the more random your sampling strategy, the fewer ballots you need to count in order to have a high level of accuracy: if you just pulled random individual ballots you could probably get away with only a few thousand in an election with a reasonable margin of victory. If you're working through boxes one at a time but choosing boxes randomly, then you need more samples, and if the boxes are being counted in order then you may need to count a lot more. I have no idea what strategy Iran used, but if I wanted to run a central count system and have accurate estimates of who had won in short order it wouldn't be that difficult.

There's apparently a statistical issue with the least significant digits of the numbers
http://www.washingtonpost.com/wp-dyn/content/article/2009/06/20/AR2009062000004.html?1

(Also, I seem to remember reports that some places had a higher number of votes cast than the available voters.)

I would want to see more than these digit statistics before questioning the results. Just because something has a 1/200 probability, doesn't mean it didn't occur naturally. My new motto: Rare things happen all the time. What was the payout on the Kentucky Derby winner again? \$103 to win. Only four times as likely as that those numbers are from a fair election.

My understanding is that the ballot-counting process was conducted by the government in complete secrecy, with no observers allowed. Under those circumstances, it strikes me as pointless to discuss possible anomalies in the statistics--the results should be assumed rigged in the absence of convincing evidence to the contrary. After all, there's no reason to exclude observers from an honest count.