|
But one of the principles of statistical analysis is that you have much more statistical power if you start with an a priori hypothesis, and test that, rather than stare at the data and see what you can see. It's a bit like tealeaves. If you tell me that you've found the image of a giraffe in the tealeaves at the bottom of your cup, and that is amazing, and must be due to a giraffe conspiracy, I won't be very convinced. But if you say: I believe there is a giraffe conspiracy: if so, then I would expect to see a giraffe in the tealeaves at the bottom of my teacup - and you then you do, then I will start to become seriously alarmed about giraffes.
So if you think that the large exit-poll discrepancy was due to fraud, and that the fraud was on an unprecendented scale because of the huge increase in DREs - and if you then show that the discrepancy is significantly larger in DRE precincts, then I will be convinced. But if people just stare at the data, and then say - oh, look, there's a big discrepancy over there, and I've never liked the look of that county's BoE - then I will be harder to convince.
So yes, what would convince me that fraud had occurred is if any a priori hypothesis was supported by a test of that hypothesis in the data. I've thought of quite a few myself, but I am sure I have just scratched the surface. I did think that the swing-shift correlation was a good one, but that didn't show anything. If it had, I would have been VERY convinced, as it is not easy to think of a non-fraud explanation for the discrepancy being greater where Bush's improvement on 2000 was greater. But it wasn't.
Unfortunately, it won't work next time, as we already suspect fraud this time. The advantage in doing it this time was that the discrepancy in 2000 was unusually small.
|