[EM] Gustafsson: Complete-ranking criterion

[Michael Ossipoff]

(I meant to use the original subject-title, but I forgot to copy it. I
reply to posts by copying posts and pasting them into e-mail to the
mailing list)

Gustafsson writes:

There are lots of voting system criteria that have been described, but
I have not seen this one - or any one like it - described before.

Bullet-voting prohibition Criterion:
"A voting system should not be constructed in such a way so that it is
both legal and rational for a voter to fill in a ballot with only one
party or candidate name, so that the voter refuses to order by
preference all candidates that are not his first preference."

In every method it's legal to "bullet-vote" or "plump". You want
balloting rules to _require_ that voters not bullet-vote, and you want
there to be a strategy-need to vote for all the candidates.

I disagree with your first goal. Often there will only be one
acceptable candidate. It would be undemocratic to force people to vote
for additional ones, by a balloting regulation.
And what if your favorite has a clear win? Why should have to support others?

Requiring complete rankings, so that voters will inform themselves
about all the candidates isn't necessary. The Approval voter who
approves only one candidate might, and probably does, do so because
that's the only candidate s/he likes.

Should there be a strategic need to vote lots of candidates? What if
there are 20 or 25 candidates in the election, and most of them are
unacceptable. Do you really want the voter to have to rank almost 20
unacceptable candidates?

That's why I proposed Symmetrical ICT,  as a replacement for ordinary
ICT. Symmetrical ICT meets 0-info Later-No-Help, which is effectively
as good as complying with the actual Later-No-Help criterion. I didn't
want voters to have to rank any unacceptable candidates.

But, for some purposes, I might agree about wanting people to vote a
complete ranking, of all the candidates. Say there are only a few
candidates, in an informational poll, and you want to find the CW. It
could be desirable to encourage complete rankings. In ordinary ICT, a
voter has incentive to rank the unacceptable candidates in sincere
order of preference.
(unless the voter has improbable predictive information, which could
make it more complicated).

You probably would like a method to meet Later-No-Harm (LNHa), and
fail Later-No-Help (LNHe). Actually, I know of two methods that do
that. Both of them have been criticized here, because of criticism
vulnerability. MinMax-Pairwise-Opposition (MMPO) was criticized for
failing Plurality, in a spectacular and embarrassing way described by
Kevin Venzke. Majority-Defeat-Disqualification-Top-Rating (MDDTR)
fails Mono-Add-Plump (making it just a little worse than the Condorcet
versions in terms of monotonicity criteria).

Those criticism-vulnerabilities would be problematic enough, for a
public proposal. But what I don't like about those two methods is that
there's a strategic need to rank unacceptable candidates in reverse
order of  (at least some unknown measure or function of) winnability.
I don't like giving the voter that strategy-need.

IRV passes LNHa. It meets LNHe, meaning that there's no need to rank
candidates below set S, in order to increase the probability of
electing from set S. However, IRV does provide a mild incentive to
rank all the candidates, because, if all of set S get eliminated, then
you might as well transfer your vote to the least-bad of the non-S
candidates.

Maybe IRV is the closest actually-proposed method to your requirements.


You wrote:

Why do I consider the BVP criterion necessary? Two reasons:
1. It makes impossible some strategies that could otherwise be used by
voters who are in favor of one of the top-2 parties, and want to
ensure a two-party duopoly, even after the FPTP system is scrapped. I
hold it for necessary that any change of the voting laws that scrap
FPTP should not only mandate a better voting system, but also make it
impossible for hidebound big-party voters to continue in their old
mindset of "vote up my party, loathe the other big party, and pay no
attention to any other party."

[endquote]

They can vote however they want, but if another set of voters have a
majority and prefer someone else, then that someone else will win. You
don't need a vote-for-many. Such a rule would be undemocratic anyway.
Voters shouldn't be forced to vote for an opponent of the candidate(s)
whom they want to win.

ICT probably gives voters reason to rank all the candidates. IRV gives
some mild incentive to do so, and gives no strategic disincentive to
do so.

Mike Ossipoff