[EM] Why We Shouldn't Count Votes with Machines

[Raph Frank]

On Sat, Oct 4, 2008 at 5:13 PM, James Gilmour <jgilmour at globalnet.co.uk> wrote:
> Correct  -  the images were not made available.  Those that were subject to adjudication by a Returning Officer were seen, by all
> who want to look, on a computer screen, large or small, during the count.  The images (and the adjudication decisions on them) are
> stored on the hard drives used at each of the 32 counting centres.  These disks have to be kept securely for four years  -  no
> access to anyone except with a Court Order.

What is the basis for granting access?

> Yes, BUT adjudication on "doubtful" images can be critical.  So if you had access to the images and ran them through your own OCR
> software you may well come up with different vote files.  Interestingly, in one pre-election "validation test" of the electronic
> processing versus manual counting for the STV-PR elections, the results were different  -  due only to a difference in the
> adjudication decision the ROs made when they look at the on-screen image and when they looked at the actual ballot paper.

Abd's proposal is to apply a serial number of each ballot (after it
comes out of the ballot box).

There would probably be consensus on 99% of the ballots and then the
returning officer can check the last 1%.

A judge might be called in for 0.1%, if there still was a dispute
after the RO gave a decision on the disputed ballots.

If there was a large number of disputed ballots, it could be done on a
sampling basis.

The petitioner would submit their opinion on what the N ballots are
and the judge might check 10 of them at random.  If the petitioner
only got 1 right, 1 reasonable error and 8 clearly wrong, then his
dispute could be dismissed as he is just chancing his arm.

>> A blank ballot is one that has no writing on it, or one that is not used?
>
> "Blank ballot paper" here means one that came out of a sealed ballot box at the counting centre and had no vote recorded on it.

Ahh, it is a check that all ballot papers are accounted for?  I
wouldn't see an issue with imaging them too.

> Those who want access to real ballot data from real elections (STV-PR) will be interested to know that the full ballot data for each
> of the 21 wards (= local government electoral districts) within the City of Glasgow were published on the City Council's website at
> the conclusion of the count on 4 May 2007.

Cool, there is also data from Ireland for the 3 constituencies that
ran the pilot e-voting test.

> It is arguable that in publishing the full
> ballot data, the Glasgow Returning Officer broke the current law, but no-one has demanded that he remove the data, and the Scottish
> Government is proposing to make this a requirement for all local government elections.  This MAY be applied retrospectively so that
> we get all the data from the 2007 STV-PR elections.  Meanwhile, the Glasgow data are invaluable resources for research, as they show
> what real voters do in real elections.

Sounds good.  This allows theories to be tested based on real data.
The can give examples where voters vote in the same order as the
ballot for lower ranked candidates.