[EM] VoteFair representation ranking recommended for Czech Green Party

[Jameson Quinn]

2010/4/29 Peter Zbornik <pzbornik at gmail.com>

> Hello,
>
> thanks for the information.
> It seems a bit unusual to keep switching methods.
> I don't understand how proportionality is achieved.
>

If I may. I'm not very familiar with VoteFair, but I understand the way it
achieves proportionality. It's basically the same as RBV, and, as you say,
STV. (It's also the principle of RRV, the range voting equivalent, and it
can and probably has been extended to create many other systems.)

You use some underlying method to elect one seat. In VoteFair, this is
Kemeny-Young; in RBV, it's Bucklin; in STV, it's IRV. (For both RBV and STV,
the requirement for majority is of course relaxed to the proportional
equivalent, the Droop quota.) Then, the votes which were decisive in that
step are "used up"; if 80% of them would have been enough, then they're all
worth only 20% in succeeding rounds.

The difference between these methods, then, is in the underlying method,
before reweighting. If I understand it, VoteFair uses a Condorcet method
over the whole ballot set. This tends to make the first few members of the
council be compromise candidates. Again, if I understand it correctly, the
potential problem is that people might have their vote "used up" on a
compromise candidate, and thus unavailable to support their actual favorite
candidate. If true, this has obvious implications for strategy.

STV uses IRV. The bottom-up elimination has the opposite problem: it can
eliminate compromise candidates early in the process, making them
unavailable later on. Thus, it's a non-monotonic method. Since the Droop
quota is much less than the majority quota of pure IRV, this problem is much
less than with IRV; but it could still strike, especially if there's an
excess of candidates from one or more factions, compared to that faction's
proper share of the council. There are some good graphical representations
of this problem in the case of IRV (again, a worse case than STV) at
http://zesty.ca/lj/yee-oca-transferable-vote-3.pdf .

RBV uses Bucklin. This is essentially multi-round approval - in my
recommended form, two rounds. In the first round, when most voters will
bullet vote, it's similar to STV, so the first few members of the council
should be the same as under STV. But then, instead of eliminating
candidates, it expands the approvals on each ballot. Thus, compromise
candidates are still available to round out the council.


> I would appreciate if Votefair ranking would have some mathematical
> description and at least well described and discussed in some peer-reviewed
> paper.
> According to the description votefair ranking looks like STV.
>
> I also have some concerns about the vote-counting.
> We would need to make sure that the vote counting cannot not be manipulated
> and that the count is independently verifiable.
>

In all three of the methods discussed above (as, indeed, in any proportional
method that I know of), slight differences in ballots can be magnified as
successive candidates are elected. It's just a simple fact that
sophisticated vote manipulation is probably easier operationally (though
harder cognitively) in proportional systems than in majoritarian systems,
because smaller piles of ballots can be decisive.

So what can be done to maximize transparency? I can think of two useful
criteria. If a system is monotonic, that helps to minimize the kinds of
manipulation possible; at least you don't have to guard against a faction
sneakily lowering their own vote totals to win. And if a system is summable,
then transparency is much easier, for two reasons. First, you can delegate a
recount, with each precinct (or other subset) just responsible for
rechecking its own totals, so that local problems can be fixed locally. And
second, it's easier to publish ballot totals so that anybody can recheck
that those ballots do indeed lead to those winners.

STV satisfies neither of these criteria; VoteFair I'm not sure, but I
suspect it's summable and not monotonic; RBV is monotonic but not summable;
and SRBV is summable and not monotonic, but has a high probability of
agreeing with the monotonic RBV.


> Is the vote-counting program possible to install on a computer?
> Is it open source?
> Is the count implementable by a reasonably skilled programator?
>

I don't know the answers for VoteFair. For RBV and SRBV, the answers to all
three questions are: "it would be". I haven't programmed it yet, but it's
not too hard.

JQ

ps. Peter, you should not be shocked by the enthusiastic response to your
query. You've thrown a hunk of meat into the lion cage: the chance to see
our theories applied in a real, important case. While we can give you useful
feedback, we will never come to a consensus; in the end, you'll have to pick
a proposal yourself. Also, while we lions may present an amusing spectacle,
we still mostly respect one another underneath, and we understand that there
is no guarantee that any of our proposals will be implemented in the end.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20100429/0f963418/attachment-0004.htm>