Given the recently held presidential elections in Iran, and the claims from each candidate that they won the election, I am wondering if it is possible to reliably estimate the amount of voter fraud in favor of one candidate. At the very least, can we reliably estimate the amount of fraud perpetrated by the incumbent, who naturally has the advantage in terms of infrastructure and resources at their disposal? I honestly don’t know and have never thought deeply about such a problem. However, my instincts tell me that there ought to be a way, depending on the quality and volume of data one has at their disposal to determine when outcomes are highly improbable and what the size of the anomalous effect is. The problem is, for those elections that we most care about–in this case, Iran–the polling data is not as detailed and representative as is necessary.

Over at, Renard Sexton took a look at the data from Iran before the polls opened. He notes that while there as been extensive polling in the run up to the elections, the data is skewed due to a number of factors, including geography (Tehran) and issues with the questions themselves. While polls showed a much closer outcome between the top two candidates, those polls focused on Tehran, where support for Mousavi is highest (Ahmadinejad’s base is rural). While I certainly don’t trust the numbers coming from Iran’s state media (62.6% to 34%), if the original polls where highly skewed towards urban centers like Tehran the difference between the expected results and the actual results may simply be a function of sampling-error. The one issue, as Renard points out, is that historically as voter turnout increases in Iran the share of the winning vote has decreased. It is hard to believe that if there was in fact record turnout, Ahmadinejad would have actually earned more votes this time out than during the previous election (where there was a run-off).

Here’s hoping that Nate and the folks at fivethirtyeight do some work on the question of estimating election fraud abroad.