On August 15, Venezuelans voted in a national referendum on whether to remove President Hugo Chavez. The (Chavez-run) government announced afterward that 58% had voted to keep Chavez in office. The opposition claimed fraud.
The election was held on electronic voting machines. Fortunately, the machines generated a voter-verified paper trail, so that there was some hope of recounting the ballots. Without a paper trail there could have been no recount, and Venezuelans would have had to take the result on faith, or reject it. With a paper trail, there is at least some evidence of how the votes were cast.
What evidence is there for fraud? The opposition says that the election results were inconsistent with exit polling, which they say went 58-42 in the other direction. That’s a big enough swing to raise eyebrows, but it’s hard to evaluate the accuracy of the exit polls based on the information available to me.
The opposition’s other claim is that the voting machines were programmed to cap the number of yes votes (i.e., anti-Chavez votes) recorded on each voting machine. In support of this, the opposition points to the data on machine-by-machine voting results, arguing that machines in the same polling place recorded the exact same number of yes votes too often, that is, more often than would have occurred by chance. That’s a claim that is amenable to statistical analysis. I’ll evaluate it in a future entry.
Venezuela Voting Analysis
Avi Rubin, Adam Stubblefield, and I just released a paper analyzing the reported voting data from the recent Venezuelan election. The paper is available at http://www.venezuela-referendum.com, in both English and Spanish versions. Here is the “Summary”…
below article traces the votes cap to the way, the machines worked:
http://www.venezuelanalysis.com/print.php?artno=1259
its from a party in the election, from what I understand – the wording is strong – there is something at stake though, too.
That is the part that scares me. “Yes, are system is so secure we’ll even let it stand up to the light of day, but we get to choose who shines that light of day on it.” Security through obsecurity is no security at all.
Some useful info:
* In Bush v. Gore, the U.S. Supreme Court cited evidence that 2% of votes cast are not counted in every Presidential election due to systemic error.
* I believe Princeton professor Paul Krugman is claiming that studies have shown a 6% systemic error rate in the U.S. voting systems. I have heard him say this in public, and the NYT OpEd page ran an unattributed article that cited this last week. I cannot find any such study.
* The Help America Vote Act of 2002 (HAVA 2002, Pub. L. No. 107-252, 116 Stat. 1666) sets minimum requirements for voting systems, including maintaining a paper trail of voting activity to facilitate a recount. Yet new touch screen voting machines in Florida this year can’t do that.
* Here is a link to how CNN processes exit polling information on election day. Notice that they won’t call an election unless the exit polls show an eight point margin when the polls close. http://www.cnn.com/ELECTION/2004/primaries/pages/misc/project.html
* This link points to a book that is still used as part of the literature for people learning about exit polls.
http://www.aei.org/news/newsID.12219/news_detail.asp
C. Scott: The venezuelan government is getting it’s money back, since (according to Miami Herald investigative work, sadly I can’t find a link) it owns a plurality of the company that made the voting machines. A paper trail means nothing if the software doing both the printing and the counting can’t be audited and verified, and if there is intimidation of those who might complain about or not understand the discrepancy between paper printouts and cast votes.
Too funny. Or scary. Maybe both.
Now my comments on the above:
First, I think the Venezuelan government deserves a refund.
After all, Smartmatic “guarantees 100% accuracy”, and even the most cursory audit of 82 machines found a discrepancy of .02%. This is a good number, but the sample is extremely small, not random (192 machines were intended to be audited, but only 82 actually were), and just compares the paper record to the electronic totals (why would there be any discrepancy at all?). It does not include errors due to voter confusion, registration problems, inability to cast votes, inaccurate summation, etc. [Source: http://www.cartercenter.org/doc1807.htm]
I react very strongly to snake-oil claims like “100% accuracy”. Every real election system will have some error rate, because humans are not perfect (even if the technology by some miracle is). The goal is to make this as small as possible, and to “be more careful” (recounts, etc) when the reported results are within your best estimate of the error rate of the process. Even soap only claims to be 99.44% pure. =)
Similarly, doesn’t the advertisment’s claim of ‘no null votes’ presume an error-free voter?
We also have the classic snake-oil claim, “250 layered security mechanisms”. If the security mechanisms work, why do you need 250 of them? If they don’t, will 250 insecure mechanisms yield a secure mechanism?
Don’t get me wrong — there might be real valid security underneath this claim. But the wording of the claim unsettles me.
I applaud the use of auditable systems. But it seems the real trick here is to actually audit them. We already have claims like ‘the Yes vote had been cut by more than 75 percent’ in one location (Valle de la Pascua) where the voting papers were recounted (http://www.iht.com/articles/534518.html)
There were two separate organizations watching the process (the Carter Center and Sumate), and they have both endorsed the ‘official’ results. The Carter Center conducted another small audit of 150 machines (whose quantitative results they have not yet released). They also did exit polling in a few select locations and compared these results with the results transmitted to the central server, and found agreement within 1%. Finally, they took some results from the central server and projected a country-wide result, which apparently agreed with the final result.
In my opinion, however, not enough details have been released about this process to convince me one way or another. How were locations chosen? Was there any way that the system could be gamed? What were the real error rates of the machines? What were the actual election incidents encountered (broken machines, late openings, intimidated or confused voters, etc). There are bound to be at least some. I don’t believe in perfection, and I’m suspicious of any claim to it.
And I look forward to Prof Felten’s statistical investigation of the ‘matching vote’ question soon. =)
[Text from an email by David M. Rosenberg:]
The following full-page advertisement ran on page 9 (of at least the
New England edition) of the New York Times on Sunday, 22-Aug-2004: