Reply #35: That is not correct

[mdhunter]

What it says is that it is almost impossible for Bush to have exceeded the MOEs based on chance alone.

That's all the binomial distribution can tell us in this case. It can tell us nothing about what the cause of the deviance actually is, only that the cause is not chance alone.

Fraud surely is one possible explanation, and I think there was fraud. But there are dozens, hundreds if we want to get creative, of others.

Consider the following, the probability of being dealt any particular hand in poker is 1/2,598,860. So, if your sitting there looking at a 3C, 7D, 4D, 6H, and 9S you would have in your hand a combination of cards that, in theory would take 2,598,860 deals to get again, a very rare and unlikely event. But, that hand is utterly unremarkable. The point here is that, even if something is unlikely and of low probability the occurance of that event is not, a priori, important in any meaningful sense.

There are parallels with this voting issue. The probability of the final tally's devating beyond the MOE of last exit poll in each state is probabably 1/20, which expresses the industry standard 95% confidence interval. Since we're concerned with only Bush "victory" outside the MOE, the probability in each state is roughly 1/40. That means one time out of 40 the polling will be wrong, due to the chance error of the sample estimate. The power of statistics is very limited here, the poll may be wrong more often than that due to other issues, be they methodological, functional, circumstances beyond the scope of the study or false assumptions.

Granted, though we may not think it terribly noteworthy if the MOE was exceeded in one state, the aggregate of 22, or however many, states is a different thing. But it is not entirely dissimilar. True, the aggregate calculations will drive down the error due to chance to infintessimal levels, but it also sums the myriad other errors or effects that could account for the wrong numbers in the first place. The question statisticians must always ask is if the numbers they're looking at are actually meaningful.

I don't think this near zero number TruthforAll calculated is particularly meaningful. We could tell, on its face - without any statistical calculation, that it would be unlikely anyone could exceed the MOE solely due to chance in some many instances. After a certain point, a point we'd reach very early on, the magnitude of the number loses any significance in the real world. It does so partly because it so large, partly because it is express only that which we already knew, and partly because it cannot, and does not pretend to, account for any other possibilities of variation beyond chance. Those other possibilities are what are meaningful now, and we don't have them. What we need is an indicator of the liklihood of fraud compared to other sources or variation. What this near zero number does tell us, positively, is that it is more likely that fraud accounts for the difference than does chance, but how much more is anyone's guess. And, whether or not that is meaningful, I leave for others to decide.