Pencil and paper voting (was Re: Multi-winner Condorcet, vs STV)
ijackson at chiark.greenend.org.uk
Mon Dec 14 15:16:53 UTC 2009
(I have changed the subject line because this thread is no longer
about the SPI elections, but about voting in general.)
Alvaro Herrera writes ("Re: Multi-winner Condorcet, vs STV"):
> Ian Jackson wrote:
> > I wouldn't support such a thing for a public governmental election,
> > because of the need for computer involvement. Public elections should
> > be done with simple and readily verifiable processes - ie, pencil and
> > paper.
> This is nonsense. As far as I can tell, in the method Markus proposes
> ballots can be cast with just pencil and paper -- which is the important
> part on which pencil and paper should be used. The result counting part
> is going to involve computers *anyway*, but since the input data (filled
> ballots) is available and verifiable, this is not a problem because
> different parties can have their own software.
Perhaps you do things differently. Conventional STV can be
implemented entirely with pencil and paper, using the physical
ballots. It's a bit more complicated than first-past-the-post (aka
simple plurality) but the same principles apply.
You sort the ballots into piles according to the first preference.
You count each pile. Decide which candidate is eliminated. Take
their pile and split it into sub-piles according to 2nd preferences.
Repeat. The number of piles can get fairly high, but not unmanageable
for an ordinary civic hall or whatever.
Why is it important do have a system can be done without computers ?
Because having the input ballots verifiable and the counting process
checkable (eg by statistical methods) is not sufficient. You also
need a fallback method for if the verification process shows problems.
If I can demonstrate (to a court, say) that the computers are
malfunctioning, I need to be able to tell the court or the electoral
commission or whatever:
This meets the criterion in section <whatever> of the election law,
which says that there is significant doubt [or whatever the phrasing
should be]. Thus you must order a manual recount as provided in the
The requirement is that the law governing the election must provide a
workable and occasionally-tested process which can be used to
authoritatively determine the result. That method must be very
resistant to manipulation and must use only processes that a court can
Courts can't supervise software. They can in theory evaluate
competing claims about software's behaviour (although in practice
courts are far too ignorant, and anyway definitively determining the
behaviour of software is fraught with difficulty as we programmers
know). But that's not sufficient, because we must not be left in the
situation where we depend on software for the answer to the election
but there is no software which can be demonstrated to be sound. If
that were to happen then the election would either have to be
cancelled pending a rewrite of the rules (or worse, the software), or
allowed to be stolen.
It's OK for such a manual recount process to be slow, but it needs to
be feasible. It is for conventional STV but not for Shultze STV.
More information about the Spi-general