Reply #225: Thanks for the nice comments, TIA

[Febble]

(though I could have done without the bit about the knickers - leave a menopausal woman a bit of dignity ;-).

And I do understand, that if you do believe that I am, as I believe you to be, essentially good and intelligent, why my stance must seem incomprehensible to you.

I think the evidence that the 2004 was corrupt is compelling, both as regards fraud and voter suppression.

I think it is possible that the corruption contributed to the exit poll discrepancy. Where we differ - and it is a huge difference - is in whether we believe that the magnitude of that discrepancy is commensurate with the scale of the fraud.

The reason I wrote, the in post you quoted earlier, that the evidence against massive fraud was mounting, was because of some specific pieces of evidence, neither of which you are likely to accept (I know you do not) but which I, for reasons which probably stem simply from my background in behavioural science, found at that time persuasive.

These were simply the evidence regarding the WPE data in the E-M report. It was an inadequate report in many ways, some of the ineadequacies being ones you yourself have cited: that hypotheses regarding fraud seem to have been given cursory attention; that no statistical details were given of the tests used; that no multiple regession results were reported; and that, IMO, the dependent variable, the WPE, is a flawed measure.

However, I do not agree that what was presented was "no evidence" for the differential response hypothesis. Sampling bias is a survey design problem, and will tend to be greater where sampling is non-random. The report clearly showed that the the "red shift" was indeed greater where sampling protocol was most likely to be compromised. As a statistician, this is strongly suggestive that sampling bias accounted for at least part of the "red shift", although it does not in itself explain why the shift itself should have been red. It is nonetheless a well-known phenomenon in behavioural research that where an underlying response bias is present, that it will be ampligied by poor random sampling protocols.

However, a further piece of evidence that for me, heaped the scale on the side of response bias, rather than fraud, as the likely more prevalent contributor the the "red shift" was the lack of a linear relationship between vote-count margin and bias. At first, and at the time when I wrote my paper, I simply regarded the postulated (at that time) lack of linear correlation as evidence against the hypothesis that fraud was concentrated in "Bush strongholds", and considered that all it would show (and in the event did show) was that the problem, whether fraud or sampling bias, was randomly distributed across the whole range of precincts. However, my own modelling exercises have since convinced me that any massive fraud, i.e. fraud on a scale large enough to account for even a substantial part of the "red shift", would produce a linear slope between bias and vote-count margin, essentially because of the Gaussian distribution of the precincts. Fraud would shift the precincts right-wards, into Bush territory, and upwards, into bias territory. This would tend to induce a slope in the correlation that is absent in the data. However, one remaining hypothesis, to my mind, that might account for massive fraud, but nonetheless be consistent with the observed non-significant slope, is that fraud may have been concentrated, not in Bush strongholds, but in Kerry strongholds - thus counteracting the slope. But at this point, I think the more parsimonious explanation (and Occam's razor applies here) is simply that fraud, if it occurred, was not on a scale to tip the slope. I may be wrong. I am simply saying that that was why I said the evidence is mounting.

I don't think the exit polls will ever provide conclusive evidence that either fraud played no part in the discrepancy, or that it did. I think they are simply too crude a tool for auditing elections. It is why I so strongly support your campaign, and the campaign of USCV to fight for fair, secure, and auditable elections. But I do think the exit poll data needs to be subjected to re-analysis, and I hope this will happen. The E-M report was simply not convincing.

Anyway, I wish you well. I am not denying my innermost beliefs when i say these things. I am simply trying to be as honest as I can. No, I don't sleep well at night. I take what you, and others on DU say both about me and about my arguments, seriously. But I cannot say what I believe to be either false, or unsupported. And at present, my conclusions regarding the exit polls simply differ from yours.

Peace? :)

(edited after posting as I clicked post instead of preview)