Yesterday, Sumate held a press conference which I did not mention because I simply did not understand what they had done in terms of proving or not the existence of fraud, this was based on a study by Ricardo Hausmann of Harvard University and Roberto Rigobon of MIT (HR) which was made available (at least to me!) today. The study they did is quite technical, so I will not try to explain it in detail but only give you an idea of what they did. The report is entitled: Searching for a black swan: Analysis of the statistical evidence about electoral fraud in Venezuela.
Besides the technical details, the report repeats a number of issues which are important in understanding the possibility of fraud; it also adds some information which should be here for the record. According to HR these are the elements that lead to a presumption of fraud:
1-The opposition wanted a manual count from the beginning, but it was electronic.
2-There was no manual counting of the printed ballots, instead there was to be a “hot audit” of 1% of the ballot boxes which never took place. Only 78 boxes were counted, with the opposition present in the counting of 28.
3-International Observers were not allowed in the totalization room, neither was the opposition.
4-The voting machines had bidirectional communications. (My note: The Head of Smartmatic said this was not the case in his press conference the week after the vote).
5-Contrary to what was agreed on, the voting machines communicated with the servers before printing the results.
6-Exit polls disagree with the results. (I add: And those whose details have been made available agree with each other)
7-In the second audit the random selection of boxes was made using the software provided by the CNE.
The report provides interesting data on them. For example:
Percentage of SI votes in Sumate’s exit poll 59.5%
Percentage of SI’s according to CNE in centers where Sumate did exit polls 42.9%
Percentage of SI votes in PJ’s exit poll 62.6%
Percentage of Si’s according to CNE in centers where PJ did exit polls 42.9%
The idea is that people always dismiss exit polls suggesting they are not done in the right place. Well, given that the SI received at the national level 40.63%, the difference is not that significant in the final results where the exit polls were performed.
Caps or coincidences
They test statistically for the caps and conclude that if there was fraud it was not via imposing caps on the maximum number of votes per machine. For quite a while I have referred to the caps as coincidences, believing they may be the consequence rather than how a fraud may be perpetrated. This would agree with that.
Detection of statistical fraud
What HR did was to look how to measure the intention of vote. To do that, they looked at two independent measures of the intention of vote: The exit poll results and the signatures from the Reafirmazo process. The idea is that each of these two represents real data, distinct from the actual vote, on vote intention. They then do a regression between signatures per center and the actual vote at those centers and the same regression between exit polls and the actual vote at the centers. In a regression you calculate the line or equation that best fits all of the points of the data you have, i. e. in the case of the exit polls what is the line that best fits the results announced by the CNE at those same centers.
When the above two regressions are done, there are errors, that is differences between the line and the points. But the sources of these errors are independents for the two processes. The only way in which they could be correlated (similar) is if the error has a common source, in this case fraud is the only possibility of “correlation” between the two. Well, the mathematical comparison of the errors of the regression yielded that there was a correlation of 0.24, where two things that do exactly the same have correlation 1 and two that have nothing to do with each other have correlation zero. To put it in a simple way: In voting centers where the signatures predicted a higher number of Si’s than the actual vote recorded, the exit polls also predicted a higher number of final votes.
Given that this correlation is simply too high and can not be explained away, they concluded that the only thing in common the two processes may have, is fraud.
Using statistical theory HR calculated the possibility that some of the voting centers had their votes manipulated and other did not. What they did was to compare those centers that were audited with those that were not audited. If those audited came from the “same” sample there should be no difference as the sample should be random. The result is quite remarkable: the results for the centers that were audited generated 10% more SI votes than those that were not audited. The probability that this was coincidental is less than 1%. Thus, the rather strong conclusion is that the centers were not chosen at random!
HR conclude by saying that in statistics it is impossible to confirm a hypothesis, but you can reject one. They then quote Popper who said that observing 1000 white swans did not prove all swans were white, but if you see a black one, you can reject that hypothesis. To HR their results are that they found a black swan, therefore, the hypothesis that there was a fraud is consistent with their results and thus, they can not reject it.
Well, my feeling is that with Elio’s work, Bruno’s and Raquel’s and some more that are soon to be revealed, what we have is a wedge of black swans getting together and forming! Someone should be getting worried, both here and abroad!