> Rich Ulrich wrote: > > With 10,000 no-punches where only half that many no-votes should be > > expected (in Palm Beach County), they re-counted a 1% sample and came > > up with 47 additional votes -- about half of the 100 or so that were > > possible, and consistent with the number of no-votes that typically > > are seen. There was no report of how many no-punches had existed. > > Gore gained, as he was expected to, because Gore carried the county by > > almost 2-to-1. > > Say punched cards have a 5% reject rate, a figure I've heard several > times, then a manual count without bias always raises the totals > in proportion to the existing votes. Thus as you say, a Democratic > county manually counted picks up net Democratic votes. > > Would you say that therefore a Republican county with the reversed > numbers ought to be manually counted as well, if a Democratic one
I'm not a US citizen, and not privy to your arcane electoral practices, but as I understand it, the counting procedures vary from county to county and state to state. All counties which use the punch hole system ought to be recounted manually as the level of error (5%) is very high and non-random (because not all counties used it).
AFAIK this is unheard of in General Election in Western Europe - even in federal countries like Germany the same design of ballots and counting procedures are used in diferent states.
> > is? Otherwise it would seem a net Democratic gain is guaranteed. > The county that hand counts wins an even election. > > The benefit of the machine counts is the avoidance of bias. The > sign of the difference at the end of it all is an excellent estimate, > not biased, though the totals are biased low by machine counting.
That's simply false.
Bias in a statistical sense can be introduced to a procedure by machines or voting systems. (Bias in this sense carries no necessary implication of human intent to cause said bias). As we are in a statistical newsgroup I'm assuming the statistical sense was intended by most posters. Bias can be introduced quite easily e.g. if the machines that produce errors are no distributed across counties in a way that balances out errors between candidates. It can also be introduced if human factors (e.g., say age) are more likely to produce an error with that machine type AND those human factors (e.g., age) are correlated with voter intention. You can come up with other examples quite easily.
> A hand count improves the totals but ruins the differences. But what > you want is the differences, ie the winner, not the totals.
Possibly. But how can you calculate the correct differences without calculating the true totals in each county?
You might want to read up on the ecological fallacy and Simpson's paradox. Because the individual counties are not homogeneous and because error is not distributed in a balanced way (or even randomly) there is no way to calculate the true difference without getting the best possible counts in each county.
> > I have heard that punch cards are favored because they retain > vote privacy, and they consider the 5% drop rate acceptable.
In the UK (apart from the recent elections for London Mayor - which weren't punch card ballots) we use hand counts and get error rates far, far lower than this (usually far less than 0.5%). Voter privacy is maintained by voting behind a curtain and folding the ballot paper in half before putting it in the ballot box. The paper has a ballot number which can be matched to counterfoils with the electoral number, but only by use of a High Court order. I fail to see how the punch card improves on this (IMO it is worse because you simply can not fold it - or it won't go through the machine).