[EM] Classifying 3-cand scenarios. LNHarm methods again.

[Abd ul-Rahman Lomax]

At 01:30 AM 4/20/2010, Dave Ketchum wrote:
>On Apr 19, 2010, at 11:48 AM, Abd ul-Rahman Lomax wrote:
>>Or other advanced method. What is often overlooked in the discussion
>>of voting methods, due to the emphasis on deterministic methods that
>>always find a winner with one ballot, is that runoff voting provides
>>the voters with an opportunity to gain information and use it in the
>>runoff. If runoffs are special elections, the voters will tend to be
>>more informed. Runoffs make FPTP work much better, because,
>>obviously, if a majority, just voting for one, vote for a single
>>candidate, that's a deserving winner! (Sure, there are possible
>>exceptions, but that doesn't break the general rule. It just means
>>that there might be room for improvement.)
>
>Plurality has been the major method in the US.

Yes. And the reasons for this should be thoroughly understood if we 
want to improve the system. The second most common method used is top 
two runoff, with or without write-in votes being allowed in the 
runoff. If they are not allowed in the runoff, the runoff becomes a 
kind of appendage of the first election instead of an independent 
election. In theory, a runoff with write-ins allowed is really 
repeated ballot, limited to two ballots, not pure runoff voting. 
I.e., it's closer to what Robert's Rules of Order actually 
recommends. The difference is the "suggested votes" in the runoff. 
Note that with Robert's Rules standard voting, there are no names 
printed on ballots.... It's all write-in.

>It gets in trouble with multiple candidates and none getting a
>majority - so do TTR runoffs.  TTR gets in trouble if the best
>candidate did not get enough votes to go to TTR (two others got more,
>even though not truly better).

Yup. In this way, TTR (neglecting the runoff option, which does 
sometimes fix this) is like IRV.

>It gets in trouble with clones and almost-clones.  This problem gets
>reduced a bit, though not cured, by parties doing primary elections.

Which then makes the general election a "runoff" between the nominees 
of two major parties. The problem arises when there is a third party 
that preferentially draws off votes from one party; this is typically 
a more extreme party, so it will suck votes from the closer party. 
There is another effect if a centrist minor party arises (or a major 
party becomes minor or close to minor because of a major extremist 
party). This leads to the "center squeeze" effect, which is similar 
to but different from vote-splitting with a major and minor party.

>Anyway, the aim was to do the major campaigning before the "main"
>election.
>
>Abd would do more runoffs, leaning toward doing the major campaigning
>then.  I choke because I still see need for campaigning before the
>main election to get the "right" leaders to the runoff.

Well, no, I would not do "more runoffs," not in places that already 
have runoffs. I'd do fewer runoffs, by using a method more likely to 
detect when a runoff is not necessary in order to gain majority 
approval of a result. Bucklin. IRV as a method of reducing the need 
for repeated process (which is the only way that Robert's Rules 
recommends it) basically doesn't work in real elections, it avoids 
very few runoffs. Bucklin would probably eliminate more than half of 
them. Real Bucklin elections did, in fact, find majorities, even 
"multiple majorities," at least initially, I suspect that this effect 
would fade and majority failure would become more common, and that it 
would settle down at about half.

At some point, possibly with cooperating researchers looking up local 
election records, we should have a thorough study of the rise, use, 
and fall of Bucklin voting in the U.S. At one point it was, according 
to the sources, in use in about 90 jurisdictions. What the hell 
happened? This is far, far more penetration than IRV has gotten, 
ever, in the U.S.

>I would go to Condorcet:
>      Forget primaries - Condorcet can tolerate clones and voters
>should be able to learn related voting.
>      I would do less runoffs - voters can more completely express
>their desires.

The big problem is going to be truncation. You can take the IRV 
ballots where there is data and analyze them to find Condorcet 
winners, you will find some, but frequently with no majority voting 
for them. Robert's Rules explains the value of repeated ballot, and 
it is about far more than simple amalgamation.


>Bucklin?  I see its way of handling ranking as not worth voters having
>to learn and use those rules.  Note that Condorcet only asks for
>ordering of ranks per voter desires.

Eh? Bucklin uses a ranked ballot. Same as RCV. Sure, it interprets 
the votes differently than RCV (IRV?) but it is very simple to 
understand. Bucklin was very, very popular, which makes its 
disappearance quite mysterious. It worked. Now, yes, with time, 
voters will "realize" that if they add a second preference vote, they 
can cause this candidate to win, it's possible that they will 
contribute by to their candidate losing. But, in reality, if the 
election comes down to their favorite vs someone else, adding a vote 
for someone else only ends up abstaining in that pairwise race. They 
don't actually *cause their favorite to lose*. They merely allow the 
rest of the voters to make that choice, in that case. This is normal 
human compromise process. We say, "okay, if the rest of you feel that 
way, I'll accept it."

If you wouldn't accept it, don't add the extra vote! This gets very 
clear if runoffs are required when there is majority failure, but 
that was never tried. Yes. With runoffs in the case of majority 
failure, more voters will bullet vote. If they don't mind runoffs! If 
they want to avoid a runoff, then they will look to see if there is a 
compromise candidate possible, and cast a vote, which can be in 
second or third rank. (Note that a full ranked ballot can be used for 
Bucklin, the method is sequential approval, going down the ranks, 
seeking a majority. It really is Approval voting with a tweak that 
fixes the Majority Criterion failure of Approval. If voters set the 
optimal approval cutoff, the majority failure isn't an actual 
failure, but there's the rub: there is no way for voters to know 
where to set their approval cutoff optimally; with Bucklin, there are 
natural forces that will do it.

To my mind, ordinary Bucklin strategy is very, very simple. First of 
all, my favorite goes in first rank, period, I don't need to worry 
about that. I can also add, if I want, any other candidate who is so 
closely approved by me to my favorite that I don't really care which 
is elected, I'll be quite happy with both. Original Bucklin did not 
allow equal ranking in first and second rank. But Duluth Bucklin did 
allow it in third rank. I'd allow it in all ranks, because voters 
should be *allowed* to equal rank, it shouldn't be prohibited, and it 
can be a more accurate expression of preference than required ranking 
distinguishing the candidates.

First rank is all I really need to consider deeply. If I have no 
preference among the rest of the candidates, I'm done, unless it 
happens that I do prefer seeing one of them elected over the regret 
of later recognizing that I didn't cast my vote in a way that would 
have improved the outcome.

That's the standard for ranking lower than first rank, to me: first 
of all, have I ranked a frontrunner? If not, I'm probably wasting my 
vote. So I want to be sure to rank at least one frontrunner. It is 
the same as Approval strategy and Range strategy, and, in fact, my 
operating hypothesis is that a Bucklin ballot is a Range 4 ballot, 
with ratings of 4, 3, 2, and 0. 2 represents approval cutoff, bare 
minimum approval. If it is in a runoff environment, i.e., if a runoff 
is held if a majority is not found, then a rank/rating of 2 means "I 
prefer electing this candidate to a runoff being held." That is very 
simple. We can expect that many or most voters will, in fact, not 
have such a preference, and that's why majority failure will take place.

Where to rank? Well, if there is only one alternative to my favorite 
that I'm willing to rank, I'd suggest ranking in third rank, if it is 
three-rank Bucklin. Minimum Approval. That gives my vote maximum 
power, under the assumption that multiple majorities are unlikely 
(which it will be in stable usage). I'd only use the middle rank 
(rating of 3) if I wanted to approve more than two candidates, and I 
had a preference between them worth expressing.

If I'm correct about optimal strategy, A Bucklin-ER ballot is a Range 
ballot. Which points to a possible improvement, adding a rating of 1, 
which means "I don't approve of this candidate, but this candidate is 
better than the worst, I'm willing to express that." What would this 
be used for? Probably to make better runoff choices. Probably not to 
determine a winner in the primary, but it might cause, sometimes, a 
winner *not* to be chosen. How? Well, what if there is a Condorcet 
winner, but Bucklin is about to declare someone else the winner? 
That's a situation calling for a runoff. What if there is a Range 
winner, based on the full counting of the votes at fractional values, 
i.e., 1.00, 0.75, 0.50, 0.25, 0, who is likewise different from the 
straight Bucklin winner? If Bucklin is used in the runoff, in fact, 
it can accomodate three or more candidates, i.e., in the extreme, the 
first-round Bucklin winner, a Condorcet winner, a Range winner, and a 
write-in. That's four candidates, which can be fully ranked with a 
three-rank Bucklin ballot. And this would be very rare, because the 
norm, by far, would be that all three of the named candidates would 
be the same candidate.

Bucklin-ER voting involves categorizing candidates into as many as 
four categories: Favorite, Good, Barely Acceptable, Disapproved. A 
fifth category can be added: Disapproved, but better than the worst.

That is not a difficult task, particularly if voters realize that 
approving someone they have no knowledge of is probably a bad idea. 
But if a voter really wants to vote "Anybody but Adolf," they can do 
it with Bucklin, it's very simple: Vote the favorite in first rank, 
vote for someone considered acceptable, and then vote for everyone 
else in third rank, but Adolf. If a majority is required, though, 
simply not voting for a candidate exercises maximum power against the 
election of that candidate, so one would only cast a ballot like this 
in the runoff round.

There is no more new content below the next quoted material, but 
since Dave thought it interesting, I'll leave it in. Yes, Asset is 
*very* interesting, it would radically reform the system while 
allowing, if that's what was wanted, the use of vote-for-one ballots. 
It would truly empower voters, far more than a system which tries to 
extract from them a complex set of largely uninformed preferences, an 
effort that is doomed to failure and that probably would not make 
particularly good choices even if it succeeds. I can be fooled by 
media appearances. Hey, what's Obama really like? I like his media 
image, very much. Is it real or is it fake? How would I know? Tell 
you what, in an Asset election for President, even though I like 
Obama, I'll vote for an elector who can actually meet Obama, see him 
face-to-face at the meeting of the Electoral College. Sure, I'll vote 
for an elector who thinks more or less like me, or whom I otherwise 
trust, so I'm effectively voting for Obama, but with a difference. My 
elector can change his or her mind.

Electoral College? Isn't that what we have already? Yup! And this is 
a bit more like it was conceived to be, but the Constitutional 
Convention punted and left the method of election of the College to 
the state legislatures, and the process was then co-opted by the 
political party system which did not exist, particularly, when the 
Constitution was put together. Asset Voting is much closer to that 
system than any later method. All candidates in Asset become Electors 
who can then determine the winner. It is so simple that it is amazing 
to me that I only know of one actual asset election. Possibly the 
reason is that with repeated ballot, where democracy is really 
practiced, someone who gets a lot of votes in the first round may 
decide to endorse another candidate in the second, exercising 
influence, which ends up having a similar effect. Asset collapses 
that to one ballot, and, if Asset were used for single-winner 
elections, I'd certainly require a majority. No majority, I'd require 
a real runoff. The voters screwed up in the first ballot and didn't 
vote for people who would properly compromise, and that will be quite 
visible in what happened, since the Electors vote publically. The 
voters can then fix this!

Asset can be used to avoid runoff elections with any primary method, 
though. All voting systems are likely to produce initial majority 
failure, because the information needed to compromise can depend on 
the outcome of the first ballot. Asset with IRV would be powerful 
(and that was the original design by Dodgson). Asset with Bucklin 
would likewise work quite well (the Asset provisions would be used if 
there was no majority winner, and I'd have the votes revert back to 
first rank only, if equal rating, the vote is split.) Asset is thus a 
general principle, not so much a specific voting method. It is a 
device for finding a "virtual majority," possibly avoiding a runoff.

>Below Abd talks of Asset - interesting.
>>
>>However, Dodgson realized, and published in 1883 or so, a very
>>simple fact: most ordinary voters, working people, busy people with
>>families, etc., would often not know much more than a single
>>favorite candidate. With STV, which he was working on, these voters
>>are effectively disenfranchised, to a degree, unless they can
>>identify candidates by party or use voting guides or the like, which
>>then gives special power to political parties and leads away from
>>electing independent representatives who might be closer to the
>>people. So he invented Asset Voting, with an idea that is so simple
>>that when I first heard about Single Transferable Vote, I thought it
>>would be this: an exhausted ballot (or excess votes) would become
>>"the property" of the candidate receiving them. (I presume that if
>>the ballot wasn't a bullet vote, that it would go to the candidate
>>in first position, since that candidate would clearly best represent
>>the voter.)
>>
>>To my knowledge, only one Asset election has ever been held, the
>>election of the steering committee for the Election Science
>>Foundation. It was, to me, quite interesting, and confirmed, more
>>than I expected, that Asset is a powerful techique for obtaining
>>full representation. There were 17 voters electing a three-member
>>committee, and the election settled in about a week. The rules
>>weren't well nailed down, but the power of Asset was such that this
>>didn't matter. In the end, there was unanimity on the result, i.e.,
>>all agreed it was fair (except for one person whose objections were
>>a little unclear, at least to me, and it has to be said that he did
>>vote for someone who did produce the result.)
>>
>>We should produce a standard set of rules for Asset election for 
>>on- line use, as through a mailing list with some provision for voting.
>>The election was secret ballot; this is possible with Asset and is
>>perhaps more difficult with delegable proxy. DP, of course, can be
>>used to create an Asset Assembly with any desired number of
>>representatives, the principles are similar. Asset is generally
>>designed to create a peer assembly, where every member has the same
>>voting power. I don't recommend DP for actual election, but for
>>negotiation of election, if you can see the difference.... Asset,
>>though, could be used immediately and raises no particular security
>>issues beyond what are already issues with secret ballot.