TCS Daily

The Venezuela Recall: Answering all that is Answerable

By Carroll Andrew - August 31, 2004 12:00 AM

Venezuela's national election commission (CNE) has formally ratified the 58% to 42% defeat of the attempt to recall President Hugo Chavez. Chavez's opposition continues its refusal to accept the result. Because of a number of suspicious items -- including an exit poll conducted by Penn, Schoen, and Berland Associates predicting a 59% to 41% Chavez loss, irregularities in an agreed-upon "hot audit" on election day that was only partially completed, and various reports of identical vote totals from different machines at the same voting center -- the opposition is calling for more exhaustive proof that the election was free from fraud.

The Carter Center, in the role of impartial observer, says the time has come for Chavez's opponents to concede defeat. The Carter Center does not consider the exit poll discrepancy to be of consequence. The day after the election, former President Carter was quoted as saying, "there is a high chance even in the best of circumstances that exit polls are biased". One cannot help but wonder if Carter's dismissal of exit polls is influenced by his own negative personal experience with them. Immediately following the 1980 elections, the Democratic Party worried that their election losses had been magnified by the early release of exit poll data predicting Carter's defeat. Whatever his reasoning, Carter is correct in asserting that an exit poll discrepancy, by itself, is not proof of fraud. Exit polls can only provide circumstantial evidence of electoral fraud.

Identical vote totals on different machines, in clusters large enough to be statistically significant, would provide direct evidence of fraud. On August 20, Chavez's opposition presented sets of identical voting machine totals from 1,879 voting machines (out of about 20,000) to Organization of American States' Secretary General Cesar Gaviria and the Carter Center's Jennifer McCoy. The Carter Center has analyzed the data and come to the conclusion that the number of coincident totals is within the range expected from random chance.

From a technical perspective, Chavez opponents hoping to find evidence of fraud in identical voting machine results are likely to be disappointed. They are hoping that the alleged fraudsters chose a hard-to-implement but easy-to-detect algorithm involving predetermined caps, while more effective tampering algorithms exist. Michael Barone has pointed out that programming voting machines to change one vote in six from "Yes" (in favor of recalling Chavez) to "No" (against recalling Chavez) would have changed electoral results from the exit-poll numbers of 59-41 to the reported numbers of 42-58. Executed properly, this algorithm would introduce no statistical artifacts -- like unusually high numbers of repeated totals -- into the data.

Does this mean that electronically changing one-in-x ballots is undetectable -- the perfect civic crime? Not quite. Venezuela's voting was not purely electronic. After using a touch-screen voting machine to record his or her choice, each voter received an old-fashioned paper ballot, automatically printed by the machine. The voter personally dropped the ballot into a ballot box. The Carter Center observed a CNE audit of 150 ballot boxes and confirmed that the electronic totals reported by the machines matched the paper ballot totals. This audit is a primary basis of the Carter Center's endorsement of the legitimacy of the election results.

The one-in-six scenario implies that at least 600,000 votes were changed from "Yes" to "No". 600,000+ individual Venezuelans who voted "Yes" could not all have been oblivious to the fact that their paper ballots said "No". Barone does not address the mechanics of fooling the paper ballot system in his article. We will red-team the possibilities here. Taking the integrity of the international observers as a given, any sinister activity had to have taken place before the audit. The possibilities are three...

Selective audit - Do not electronically the change the results from a certain number of polling stations. Then, rig the audit-selection process to select stations that have not been tampered with.

Switcheroo - Sometime after the end of the election but before the audit, replace the original paper ballots with new ballots matching the altered electronic totals. But is it reasonable to believe that this could be done for 8.5 million ballots in 20,000 boxes?

Switcheroo followed by a selective audit - If replacing all 8.5 million ballots is too much, set up the selective audit. Then, replace the ballots in the boxes destined to be chosen with the new ballots that match the electronic totals.

Are any of these scenarios possible? More importantly, given available data, can any of these scenarios be ruled out?

The Carter Center could help rule out the possibility of a selective audit by releasing information on exactly which polling stations were audited. If results from some polling centers were altered while results from the surrounding centers were left untouched, the distribution of results from the 150 audited stations will show systematic variations from their neighbors.

Finding evidence of a switcheroo is more involved. First, it must be shown that an opportunity to replace the paper ballots existed. This picture, available from the official website of the Venezuelan government, does not inspire confidence that the ballot boxes were tamper-proof. More importantly, both Carter Center reports give detailed accounts of ballot box custody beginning on the afternoon of August 18. They do not report on the time between August 15 (the night of the election) and August 18 in the same level of detail. The August 21 report, for example, explicitly states that "Observers remained with the cajas ("boxes") during the nights of Aug. 18, 19, and 20". There is no similar statement regarding the nights of August 15, 16, or 17. Based on the best information available at this time, an opportunity for ballot replacement appears to have existed. International observers could help dismiss this possibility by providing more detail about the amount of time alone, if any, that Venezuelan officials had with the ballot boxes.

Blogger Miguel Octavio proposes a feasible test for ballot replacement. Let the opposition pick another 150 boxes to be audited, using any criteria they prefer. In addition to tallying the "Yes" and "No" counts, find the fingerprints on the ballot. If there was no tampering, there should be a set of unique prints on every ballot. If replacement ballots were created, all of the prints on a replacement ballot will match prints from other ballots -- unless you believe that fraud involved at least 600,000 people stuffing ballot boxes over just two-and-one-half days.

The audits performed to date have been insensitive to whether each ballot reflected the choice of one and only one individual. They have assumed this fact to be true, despite the questionable provenance of the paper ballots. Fingerprint matching to test that the number of voters equals the number of votes would be direct verification of one-vote-per-voter. This type of audit would help rule out the switcheroo possibility. Given the uncertainty and the lack of trust that already exist, it is no more unreasonable an audit than the audits that have been performed to date. While this audit would be more complex than any of the previous audits, confirming that the people's input and the electronic output are one and the same is worth the manageable increase in complexity.


TCS Daily Archives