It's the sampling, stupid!

…or…Why we can be absolutely sure Jorge Rodriguez has something to hide… …or, perhaps just… Dear Greg, I thought you did a good job in your interview with...

…or…Why we can be absolutely sure Jorge Rodriguez has something to hide…


…or, perhaps just…


Dear Greg,


I thought you did a good job in your interview with JR, and I wish other Venezuelan journos would adopt that interviewing style: ask hard questions, but ask them respectfully. Like I said, I wish you’d asked him about the chavista-witness signatures on the actas, largely out of my own curiosity. I’ve never heard a convincing explanation of how you could have a megafraude and chavista witnesses willingly signing thousands of actas at the same time – but that’s a relatively minor thing, and not really central to our judgments of JR. Actually, the questions you asked – and the responses he gave – are enough to leave me 100% sure the guy is hiding something. It’s all about the sampling.


According to last night’s finally final (no, this time we mean it, really) figures, there are 1,910,965 million valid signatures and 1,192,914 we might call “in-doubt.” The opposition needs 2,436,083 million to call a referendum, meaning that 525,118 of the 1,192,914 in-doubt signatures would need to be repaired. That’s 44% of the in-doubts. What, Greg, do you think is the most accurate method for determining whether more or fewer than 44% of the in-doubt signatures are, in fact, valid? (Here I’m asking Greg the sociologist, not Greg the pundit.)


Now, JR says with great scorn that the opposition, in the context of the reparo negotia…pardon, conversations, suggested testing a random sample of 170 signatures. “Out of a universe of 3 million!”


According to this handy Margin-of-Error calculator from the American Research Group (a well respected US polling firm), a sample-size of 170 for a population size of 1.19 million gives a margin of error (at the usual 95% confidence interval) of 7.52%. Is that a lot or a little?


It depends. If 70% of the sample had been found to have signed properly, a 7.52% margin of error is perfectly okay – because the opposition only needs to show 44% of the signatures are valid. Similarly, if only 30% of the sample had been found to have signed properly, then a 7.5% margin of error would still be fine. It’s only if the sample-mean had been found to be between 36% and 52% that you have some real indeterminacy. Even then, the exercise would have been worth it, because it would’ve given you a far better sense of the scale of the problem than we now have.


But JR doesn’t like that small sample size. OK, no problem, follow Carter Center and use a sample size of 1200. The ARG Margin-of-Error calculator says at that point you’ll have a 2.8% margin of error at the 95% confidence level. So, if you did it this way, you would only have a problem if the sample mean is between 41 and 47%. A risk, sure, but one with the upside of yielding results that are a-scientifically valid, b-very difficult for either side to dismiss as partisan, c-quick, d-cheap, e-consensual, f-legal and g-fair.


Now, I’m honestly looking for your help here. I’ve been trying, hard, for months, to try to reconcile JR’s refusal to go for a sampling methodology with the multiple personal references I have vouching for his fool-proof integrity. Try as I might, I can’t do it.


I can’t think of any coherent argument to suggest that a positive reparos process, where you try to drag out all 1.19 million of the questioned signers to sign again, would yield a more accurate picture of how many people really signed than a scientific sample of questioned signatures. It’s quite obvious that there’s far greater potential for error in the reparos process, compounded by the fact that a positive reparos process can ONLY undercount the signers, and made even worse by the near-certainty of government pressure to dissuade people from repairing (cf. Roger Capella’s infamous “terrorists and conspirators” statement.)


Moreover, a reparos process has no built in mechanism to estimate its own propensity-to-error – there is no handy, internet based margin of error calculator for a reparos process. So say you did hold one, and say only 400,000 people go to repair their signatures. You have no way of knowing if those 400,000 are all the people who signed validly back in November, or if they’re 75% of the people who signed validly back in November, or half, or whatever.


You’re a sociologist, Greg, a slightly meta-post-type sociologist, but a sociologist nevertheless. Cast your mind back to your research methodology courses. Imagine your teacher faced with this problem. Imagine yourself trying to persuade him/her that a reparos process would be more accurate than a sample of the questioned signatures. Could you? Would any statistician buy JR’s argument that a reparos process clears up the “margin of doubt” more effectively than sampling?


No matter how many ways I think through this one, I keep coming back to the same conclusion, a conclusion that is deeply painful to me because people I’m very close to have assured me again and again of JR’s bona fides, and to lose faith in JR means losing faith in them as well. But there’s really no way I can avoid the conclusion that the only imaginable circumstance where you would reject using scientific sampling is if you have something to hide.


So Greg, please, please explain to me where this chain of reasoning is going wrong. For once, I want it to be wrong. It would be far easier for me to believe I’m wrong here, far more comfortable emotionally, but I can’t for the life of me see it.


Especially when I start running into stories like this:



Miami Herald:

No Chávez recall vote coming, official said


BY RICHARD BRAND

[email protected]


CARACAS – An executive of an Omaha firm that sold voting machines to Venezuela says a top electoral official asked him during a meeting last year whether his company worked for the CIA and predicted there would be no recall referendum against leftist President Hugo Chávez.


”There was a certain amount of paranoia,” recalled John Groh, president of the international division of Election Systems and Software (ES&S), of the October meeting in a Manhattan hotel room.


The meeting was with Jorge Rodríguez, considered the most pro-Chávez and influential member of the Venezuelan National Electoral Council’s five-person board of directors.


”I took an hour explaining to him that we were not part of the CIA, that I didn’t have a military background,” Groh told The Herald.


‘We said that we understand you have an election coming up where there will be a presidential referendum, and . . . Rodríguez said, `That election will never take place,’ ” Groh said.


The electoral council, with three pro-Chávez directors and two who favor the opposition, has temporarily dismissed one million of the 3 million signatures the opposition collected demanding a recall vote on Chávez.


At the time of the meeting, ES&S was seeking a $20 million contract to dust off the 7,500 ES&S Model 100 optical scanners, which read paper ballots, that the firm sold to Venezuela for the 1998 elections.


Instead, the electoral council awarded most of a $91 million contract in February to Boca Raton-based Smartmatic Corp., a tiny company whose touch-screen voting machine has never been used in an election anywhere.


”Smartmatic came out of the blue,” Groh said in a telephone interview from Omaha. “It’s a very small, very new, storefront business that should never be considered for a project of this size.”