by Daniel on November 3, 2004

Academics from Harvard University have conducted statistical tests on the output of voting machines and concluded that there is less than a 1% chance that their results could have been generated in any way other than massive voting fraud.

Of course, we shouldn’t get too excited about this right now. For one thing, their work has been called into question. And for another thing, they were talking about the Venezuelan referendum. But I’m certainly not above using cheap shock tactics to draw attention to a really interesting piece of statistical argumentation.

It’s one of those cases where there are points on both sides, and both sides have, extraordinarily, managed to treat each other with respect and not throw around “Devastating Critiques” of each other. I think that Mark Weisbrot has the best of it; his analysis of the audited “clean” sample shows the same result as the allegedly “dirty” sample. The only way that one could have got that result fraudulently would rely on a level of compromise of the Carter Centre’s random selection apparatus that seems very implausible.

But Rigobon and Hausman are correct to say that it is troubling, to say the least, that the audit sample shows a very different relationship between polling station size and the “Si” vote; I’ve been thinking about it off and on for a month and I can’t come up with any reason why that might be. If you’ve got some spare time on your hands (and if you’re a Democrat, you do), both papers are worth a read, if only to clear your palate after the crap that’s been thrown around at the Lancet study (btw, on that subject, I will post more, but in the meantime, nice one, Chris Lightfoot).

(PS: I misspell Rigobon’s name out of ignorance of how to create accents in HTML rather than any other kind)



Stick Candor 11.03.04 at 8:31 pm

It doesn’t help that much to advance the case for fraud that the “academic” of Harvard is no other than Ricardo Hausmann, former Minister of Planning of Venezuela and member of the Board of the Central Bank of Venezuela during the years of the corrupt presidency of Carlos Andres Perez, tried and sentenced and presently a fugitive of justice (although I don’t know that Hausmann himself has anything to do with the corruption). This is the same guy who calls Chavez anti-democratic every few minutes while at the same time supporting every coup d’etat attempt that appears in front of his eyes. I heard him a few times and he seems pathologically consumed with hatred for Chavez.


novalis 11.03.04 at 9:14 pm

I’m not too impressed with either the paper or the critique.
1. It seems to me that the government’s refusal to do a full recount is damning.

2. HR propose the following model for fraud:

a. Polling stations are divided at random into two categories: altered and unaltered.

b. Altered polling stations’ results are changed such that, when they are summed with the unaltered polling stations, No wins.

c. At audit time, only ballots from unaltered polling stations are used.

WRT correctly criticize this model on the grounds that results from the unaltered polling stations match the declared national result; if the declared result were in fact nearly 20 points off the actual result, we would not expect a random sample of unaltered polling stations to so closely match the declared result (it is much more likely that it would match the actual result).

But what if HR was sloppy? What if unaltered polling stations were not chosen at random? Instead, when polling stations reported in, a sample summing to the declared national result was chosen to be unaltered. This is possible if the distribution of Yes and No votes was not uniform (likely).

The US analogy (assuming the Republicans wanted to rig this election to produce a 59-41 result, and that we used electronic voting more-or-less everywhere) would be to choose unaltered results from mostly red areas.

Did this happen in Venezuela? I don’t know if politics is as divided by geography as it is in the US. So, I can’t evaluate this claim. But for HM to suggest that the unaltered stations were chosen at random was sloppy; for WRT to ignore this was uncharitable.

Unless I misunderstand both papers, which is always possible.


novalis 11.03.04 at 9:18 pm

Oops, I’m an idiot. WRT do consider and reject that.


novalis 11.03.04 at 9:19 pm

Oops, I’m an idiot. WRT do consider and reject that.


novalis 11.03.04 at 9:20 pm

Oops, I’m an idiot. WRT do consider and reject that.


dsquared 11.03.04 at 9:21 pm

But what if HR was sloppy? What if unaltered polling stations were not chosen at random? Instead, when polling stations reported in, a sample summing to the declared national result was chosen to be unaltered.

Yeh, H&R suggested this in comments on the CEPR paper. But in order to do this, Chavez would have needed to know that the “correct” sample of clean stations would be produced by the Carter Centre’s random number generator. Even if they could have controlled the “seed” of the PRNG, selecting a seed to give a particular desired output from one of those things is a task about as difficult as cracking a triple-DES encryption (mainly because it’s a similar algorithm at work)


novalis 11.03.04 at 9:29 pm

I’m not convinced that it’s as difficult as cracking 3DES. For one, the input to the PRNG is a 1-8 digit number, so at most, 30 bits. 3DES is, of course, equivalent to 112 bits.

And HR suggest that the government did choose the seed.

But I think the statistical argument of WRT suggests that unless those particular precincts were chosen extremely carefully, among populations which had had massive political shifts over the last four years, there’s no way that could have happened.

So, WRT probably have the best of it.


Ron 11.04.04 at 12:53 am

Pardon me, I haven’t seen the paper but from a statistical point of view it is hard to imagine a design that would allow you to conclude that the results reported by the machines had to be fraud. It’s just not that kind of a problem. Sounds like snake oil to me

Also, Chavez achieved what we didn’t: a landslide of popular support for his presidency in which the venezuelan people reaffirmed its democracy in the face of big media, the banks and the US government


Chris Lightfoot 11.04.04 at 9:59 am

I don’t understand the point about random-number generators. Note,

“In this sense, it has to be pointed out that the National Electoral Council refused to make use of the random numbers-generating program proposed by the Carter Center and insisted on the use of their own program installed in their own computer.”

If this is true, there’s no question of trying to find a seed which will produce a given sequence. If the RNG is under the control of the bad guys, they can design it to produce a sequence that they want.

The fact that the Carter center audited source code and an executable for a RNG doesn’t tell us anything about the program which was actually run to pick the sample (see this Ken Thompson paper if you don’t see the problem, though a much simpler deception could have served in this case).

Note, of course, that this is exactly the same problem as direct-recording electronic voting has itself.

I haven’t read the Hausmann/Rigabón paper in enough detail to give an opinion on whether they’re right or not, but if the question of fraud hinges on the particular configuration and design of the RNG used to pick the audited sample (as it would in the H&R model) and the program was under the control of the government, then the audit is not trustworthy.


dsquared 11.04.04 at 10:42 am

Good point Chris; I had previously assumed that Chavez only had control over the RNG output through choice of seed.


Marie 11.04.04 at 6:51 pm

Does anyone know if anyone is analyzing the election data along these lines, i.e., seeing whether counties with electronic voting have different patterns (oh, say, higher proportion of votes for B/Ch) than other similar counties? Does anyone else suspect that perhaps vote counts were adjusted? Or am I just a victim of blue-sate paranoia?


just me 11.04.04 at 9:01 pm

It may not be massive, but the cumulative effect of things like this: are pretty wearing on prospective voters. Along with a new infrastructure for conveying a message to voters, we need to work on improving our election system itself.


dsquared 11.05.04 at 12:32 am

Greg Palast has had a go, treating a perennial and known serious problem (the racial profile of spoiled ballots in the USA) with his usual combination of tact and nuance.

Comments on this entry are closed.