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

[Kathy Dopp]

On Wed, Oct 8, 2008 at 7:14 PM, Jonathan Lundell <jlundell at pobox.com> wrote:
> That's true enough, though it's also true that the average citizen isn't
> going to recount (or even observe a recount of) a plurality election. I've
> participated in one myself, and it requires true dedication.

True. But citizens should have that option, and having that option is
a deterrent to fraud.

> That's the point of my suggestion, though: it's easy to audit, either 100%
> or by sampling, the ballot file, and a concerned voter could surely find an
> independent counter that she trusted, even if she couldn't manage the count
> on her own.

Well, but its virtually impossible for an average citizen to figure
out how to count all those votes in time before the election is
certified, and also introduces problems with vote buying and loss of
ballot privacy if all the choices are publicly published.  I
personally would not trust any independent counter to get it right if
I could not verify it myself - due to the complexity of the count and
the likelihood of innocent errors, even if I trusted someone, I would
not trust them to correctly write the programs to count IRV.

>
> The system could easily provide a set of test files with known results such
> that a prospective counter could have reasonable assurance that their
> counting software was counting correctly. Of course, in order to challenge a
> count, the challenger's counting software would have to be open-source, so
> it could be independently confirmed that the discrepancy wasn't due to a
> bug.

That method of trying to ensure accurate elections would do nothing
whatsoever to ensure accurate election vote counts. That is akin to
today's incompetent election officials who insist that simply because
the machines can accurately count a set of test ballots before or
after the election, that *must* mean that the election day results are
accurate. That idea is insane because anyone who has studied computer
science knows better.

> Well, we disagree on the merits of STV, but my suggestion is really
> method-independent.

Your method is not method-independent because the only way to check
machine counts is with hand counts and some methods are LOTS easier to
accurately and efficiently hand count than other methods, and checking
the results of running a small set of test ballots, or even a large
set, does zip to check election results accuracy.

Cheers,

Kathy