[EM] Steve Eppley's Just-In-Time Withdrawal (JITW)

[Michael Ossipoff]

>> After an election, any candidate can withdraw from the election, and
>> call for a new count of the ballots, with his name deleted from all
>> the ballots.
>>
>> I liked JITW, because it saves FBC-failing methods from their FBC failure.   .
>
> Maybe.  You could end up with a "chicken" dilemma.
>
> For example, if there are 3 candidates in a Condorcet loop.  If either
> of the 2 non-winners withdraws, then the other loser becomes the
> winner.

In a 3-candidate Condorcet cycle, for any pair of candidates, only one
of those could elect the other by withdrawing. If the other withdrew,
that would elect the 3rd candidate.

So there wouldn't be a chicken dilemma there. But, in more complicated
larger &/or compound cycles, I don't know. It might get complicated
then.

Suppose the method is IRV, with JITW. Say the candidates are a Green,
a Democrat, and a Republican.

Most or all Dem preferrers would rank Repub 2nd, because Dem & Repub
are incomparably closer to eachother than either is to Green.

The Repubs would rank the Dem 2nd, for the same reason.

Could JITW IRV make a withdrawal chicken dilemma between Dem and Repub?

Well, say both Republocrat factions rank sincerely, ranking
eachother's candidate 2nd.

Say the Dem or the Repub gets eliminated. Hir votes will go to the
other Republocrat. When two candidates' voters rank both candidates
over everyone else, and they add up to a majority, that's a mutual
majority. IRV meets the Mutual Majority Criterion.

Mutual Majority Criterion (MMC):

If the same majority of the voters all prefer the candidates in set S,
to all the other candidates, and if that majority vote sincerely, then
the winner should come from S.

[end of MMC definition]

IRV passes that criterion.

Because the Dem voters and the Repub voters rank eachother's candidate
2nd, there'd be no need for either candidate to withdraw.

When would JITW been needed in IRV?

What if the Dem gets eliminated in the first count, because s/he
doesn't have the Green voters' votes yet.

Dem is eliminated, and hir votes go to Repub, because Repub is ranked
2nd by the Dem voters. That's good for the Republocrats, but not so
good for the Green-preferring progressives (at least the ones who
think there's a difference between Dem & Repub).

So, because hir voters are very clear with hir about their wishes, and
because maybe s/he too feels that way, the Green candidate will
withdraw. Now hir voters' rankings will have Dem at the top, and Dem
will win in the 1st count, instead of getting eliminated.

There's no chicken dilemma there either.

If only the IRV advocates could be convinced to allow JITW, as part of
their IRV proposals, IRV would be an adequate method, even with out
current electorate.

That Green's withdrawal is in no way a betrayal of hir voters--S/he
lost. On the contrary, it's what they want. They'd be insisting that
s/he do so. So how could it be bad then? Whom can anyone claim is
wronged by it?


> In a L - C - R situation, if R wins, then if L or C withdraws to the
> other one becomes winner.  (assuming R is not the condorcet winner)

Not in Condorcet, because only one of those two factions (L and C)
could elect the other faction's candidate by withdrawing.

Not in IRV, because if the L and C factions both rank eachother's
candidate over R, and if they add up to a majority, then the winner
must come from {L,C}, without any need for anyone to withdraw. IRV
will automatically elect whichever of {L,C} has a larger faction.

By the way, get used to the fact that IRV appears, at the present
time, to be the only likely replacement for Plurality. Approval and
Score are better, and seem intrinsically more enactable, but IRV is
what the Greens, and maybe the Libertarians offer. It looks as if the
only way we'll get a better voting system will be if we elect a Green
(or Libertarian?) government.

A state initiative for a better voting system is prohibitively
expensive and difficult. No one's doing it. It looks like it ain't
gonna happen. It's a _lot_ easier to vote for the Green candidates, in
the already-provided elections. Then, the next voting system will be
IRV. But surely, with the Greens' improved democracy, there'd be an
initiative or referendum on voting systems, and we'd have plenty of
opportunity for the public to choose a different voting system.

And how would we choose one? By IRV, of course. Suppose that
initiative or referendum included all of the familiar proposed
methods. Here's how I'd rank them, in the IRV initiative or
referendum:

1. Approval
2. Coarse Score
3. Fine Score
4. Majority-Judgment
5.ER-Bucklin (with the MMC delay)
6. IRV

Count-labor is the standard on which that ranking is based.

The only reason why I'd rank MJ over Bucklin and IRV is because MJ has
such an easy count. It's much better to make a little extra work for
voters, in order to avoid a lot of work for counters--to minimize or
avoid count-fraud vulnerability.

A reason to rank ER-Bucklin over IRV is because Bucklin might have
less count labor, because of the job of crossing the eliminated
candidate off of each ballot when there's an elimination. But a
presidential national IRV could consists of nothing other than a
series of Plurality counts (Plus the job of crossing out eliminated
candidates on the ballots). That doesn't begin to match the complexity
and unfamiliarity of a Condorcet handcount. The Bucklin count would
probably be quicker, because it would only require one count at each
precinct, and one sending-in of precinct counts. Either would be ok.
But neither would be as easy as Approval, Score, or MJ....Especially
Approval, which has the easiest count of all.

Of course E-Bucklin's FBC-compliance is another reason to rank it over
IRV. IRV's FBC-failure is a reason to rank it below all of the
FBC-complying methods.

But, as I've said before, with an electorate that has just elected a
Green government, FBC-failure wouldn't be the problem that it is with
our current electorate. FBC would remain very desirable, though,
justifying ranking IRV below the FBC-complying methods.

No one would any longer think that the Dem is needed as a lesser-evil
compromise. But, if a rival progressive party arose, it could become
large enough to eliminate the Green candidate in an IRV election. But
surely both parties, Green and rival progressive, would rank eachother
over the Republocrats, and IRV's Mutual Majority Criterion compliance
would ensure that one of their two candidates would win.

And IRV's compliance with CD, and the stronger Later-No-Harm, would
avoid the chicken dilemmas that can spoil mutual majorities. When a
method meets MMC and CD, and when there's a mutual majority, that's
effectively almost or fully as good as meeting FBC.

ER-Bucklin's problem is that, even if it has the delay that confers
MMC-compliance, it doesn't meet CD. It can have the chicken dilemma.
And the chicken dilemma can take away the benefit of MMC compliance.

So would IRV be better than ER-Bucklin, under those conditions, with
an electorate that has just elected a Green govt? Maybe.

But Bucklin might be easier to count, and would be
favorite-burial-incentive-free under all conditions.

I didn't include JITW IRV in the ranking, above. I'd rank it above all
the other rank methods.

Of course, Approval or Score with optional delegation would bring
similar benefits, with a simpler count.


>> In group-reply e-mail, Steve and I proposed JITW IRV to IRVists. They
>> rejected it, claiming that it was completely unacceptable to let a
>> candidate withdraw. They seemed to feel that a candidate's withdrawal,
>> in JITW, would somehow be a betrayal to the people who'd voted for
>> that candidate.
>
> It would increase the complexity of IRV.  Complexity is regularly used
> as an all-purpose objection to most non-plurality voting system.
>
> Would you be intending multiple passes, or a single withdrawal round?

Just one withdrawal round. So, as you said, below, that wouldn't be a
significant complication to IRV. Unfortunately the IRV advocates
rejected JITW, without giving a good reason.
>
> So,
>
> - IRV count
> - one or more candidates withdraw
> - final IRV count
>
> or looping, with a full recount being triggered by each additional withdrawal?

The former. Just one withdrawal round.

>
> With 1 withdrawal stage, it would possibly get past the complexity
> objection (esp if IRV was supported in the first place).
>
> However, I am not entirely convinced that candidates would put their
> voters first in such a situation.

If not, then they could kiss their political career goodbye.

Voters will be quite clear about their wishes, both before the
election, and after the initial count is announced.

>Better to play chicken and probably
> lose than withdraw and definitely lose.

There won't be a chicken game.

>
> This is also a problem with Asset voting.

Well, with Asset, or (better) with Approval or Score with optional
delegation, there could, and often would, be negotiation among the
candidates, to decide the placement of delegated votes. It might work
well.

Approval or Score, with optional delegation would be better than
Asset, because voters could choose for themselves if they want to
delegate, or just cast their own vote.

Approval with delegation should let a voter cast an Approval ballot,
and still delegate some candidate (of hir own choosing) to re-vote for
hir if none of hir approved candidates wins. Score could have
something similar.

>
> Maybe the withdrawal decision could be taken by someone other than the
> candidate themselves.

Maybe, but the candidate would be responsive to hir voters' wishes,
and would, in fact, likely share them.

It would be difficult to allow voters to withdraw their candidate,
because, with  secret ballot, how can you prove that you voted for
someone?

No, let it be up to the candidates. Much simpler.
\
>
> Also, if a threshold was added then you could sometimes have only one round.
>
> IRV count (transfers below the threshold ignored)

I wouldn't want to ignore any ballots' rankings, or anyone's vote.
That would be difficult to justify, in any case.



> - if the winner gets > 50% of the ballots cast, winner wins without a
> second stage

If someone wins by 1st-count majority, then there'd be no need for
anyone to withdraw.

Mike Ossipoff