Thursday, November 04, 2004
Lies, Damn Lies, and Discrepancies
Hmm. This is interesting, and worth keeping an eye on. With a good stout grain of salt, as several of the comments point out, but nonetheless statistically suggestive.
In brief, what is being asserted (without sources quoted, but with the firmness of solid data and one not afraid to be checked back on), is this:
- There is a discrepancy between the exit poll results, and the actual results, in several states, but not all of them.
- In a number of states [list provided in the link], swing states with e-voting machines, the discrepancy is invariably ~5% in Bush's favour. It is asserted without proof that there are no examples of a discrepancy in Kerry's favour.
- In other states, swing states without e-voting (or with a paper trail), the discrepancy is much, much smaller - 0.1 to 0.5%. Unfortunately, this is quoted from a quite small sample of states (I can't tell if it's two or three, but not more.)
- The major networks have revised their exit poll data since the election, claiming "reweighting" of the data or giving no explanation at all for the change. Their exit polls now more closely reflect the 'results' of the election.
I want to see a more thorough look at this data, with the complete data set; this interests me. Purely in the name of justice, goddammit, if nothing else. (Does Kerry's concession render the whole thing moot?) It should be possible, given the known sizes of poll error bars and so on, to calculate the probability of these discrepancies (always in favour of Bush, etc) occurring randomly. Those error bars are, indeed, largish... but not so large that the systematic nature of the data doesn't beg that the calculation be performed.
The article also points out that exit polls are frequently used by election observers as one check on the integrity of an electoral process. Neither side of the argument cites sources for how large the discrepancies need to be, or what guidelines exist for this, or moreover what other indicators are used as well... but I am inclined to at least support the assertion that if a third world country showed this kind of discrepancy, election observers would be correct in asking some very tough questions.
The game may not be over yet. This flap may die down, as many of them will... but it may also be the first faint cry of something that becomes very much more.
In brief, what is being asserted (without sources quoted, but with the firmness of solid data and one not afraid to be checked back on), is this:
- There is a discrepancy between the exit poll results, and the actual results, in several states, but not all of them.
- In a number of states [list provided in the link], swing states with e-voting machines, the discrepancy is invariably ~5% in Bush's favour. It is asserted without proof that there are no examples of a discrepancy in Kerry's favour.
- In other states, swing states without e-voting (or with a paper trail), the discrepancy is much, much smaller - 0.1 to 0.5%. Unfortunately, this is quoted from a quite small sample of states (I can't tell if it's two or three, but not more.)
- The major networks have revised their exit poll data since the election, claiming "reweighting" of the data or giving no explanation at all for the change. Their exit polls now more closely reflect the 'results' of the election.
I want to see a more thorough look at this data, with the complete data set; this interests me. Purely in the name of justice, goddammit, if nothing else. (Does Kerry's concession render the whole thing moot?) It should be possible, given the known sizes of poll error bars and so on, to calculate the probability of these discrepancies (always in favour of Bush, etc) occurring randomly. Those error bars are, indeed, largish... but not so large that the systematic nature of the data doesn't beg that the calculation be performed.
The article also points out that exit polls are frequently used by election observers as one check on the integrity of an electoral process. Neither side of the argument cites sources for how large the discrepancies need to be, or what guidelines exist for this, or moreover what other indicators are used as well... but I am inclined to at least support the assertion that if a third world country showed this kind of discrepancy, election observers would be correct in asking some very tough questions.
The game may not be over yet. This flap may die down, as many of them will... but it may also be the first faint cry of something that becomes very much more.
<< Home