[Election-Methods] RELEASE: Instant Runoff Voting (Chris Benham)

[Abd ul-Rahman Lomax]

At 01:51 PM 6/24/2008, Chris Benham wrote:


>>----- Original Message ----
>>From: Abd ul-Rahman Lomax <abd at lomaxdesign.com>
>>No. That fight is over the Democratic Party nomination and
>>endorsement. It means that the whole apparatus of the Democratic
>>Party is devoted to one candidate, which is, of course, strongly in
>>the interest of the Democratic Party.
>
>You know that that is somewhat beside the point. But I get the impression
>that most of the money goes directly to the campaigns of the individual
>candidates and that the media attention is mainly focused on the individual
>candidates, rather than say  "the policies of the Democratic Party" 
>irrespective
>of who is their endorsed candidate.

Actually not beside the point. Media attention is, of course, focused 
on the candidates. *In the primary.* And then on one candidate from 
both major parties *in the general election.*

If we were to try to combine the primaries with the general election, 
which is what happened in Louisiana, we'd have similar problems with 
vote-splitting. Primaries aren't an indispensable part of the U.S. 
political system; used to be that state parties elected delegates, 
and delegates made the decision. It was actually a better system, in 
my opinion. Decided whom to run in a general election is a terribly 
complicated decision, the question isn't only "who is best," but also 
"who is electable." Instead of expensive primary races, I'd focus on 
much cheaper methods of making sure that the party convention is very 
representative and trustworthy. Guess what method(s) I'd use, Chris?

Then, the drama of the convention brings free publicity, much better 
than happens now with rubber-stamp conventions where nothing really 
exciting happens, they are just big celebrations. And the real money 
is then saved for the real election. While it's true that the primary 
system allows a candidate to "show his or her stuff" under difficult 
conditions, one single race isn't really enough to test that well.

> >Why not simply endorse both candidates?  After all, one cannot
> >possibly spoil the election for the other because Approval has
> >no spoiler problem. Voters simply approve candidates or not
> >completely regardless of what other candidates are on the ballot,
> >right?
>>
>>Sure. If we imagine that somehow the parties have decided not to
>>nominate candidates, snowballs in hell nevermind, running both Obama
>>and Clinton against a single McCain would probaby result in very
>>common double-voting. Now, if Obama and Clinton heavily campaign
>>against each other, slinging mud, etc, trying to convince the voters
>>that the other one is practically the devil, nobody would benefit
>>from this except McCain. Which is quite why we don't do things this
>>way. Parties in Australia don't run multiple candidates for the same
>>single-winner office, do they?

>No, but very closely allied candidates sometimes run against each other,
>such as a candidate each from both Coalition partners (the Liberals and
>the Nationals).

I consider the Coalition partners to be the same party, effectively, 
and usually they don't run two candidates in the same race. It would 
be interesting to see what happened where they do. It could be that 
the pact broke down, or it could be that it was safe. With IRV, a 
certain amount of this could be done safely. But if there was a real 
race with the other leg of the tripod, the splitting of campaign 
effort could result in a bad result. I think that's important to 
realize. It's not just what's on the ballot and how the votes are 
counted. That may be less than half of what's involved.

>I don't see how the split-vote problem in Approval is a "very different
>animal than the split vote problem in Plurality".  To me it is just much
>less severe. The "split-vote problem in IRV"  is much less and normally
>unnoticable.

You have to understand how Plurality works: it works through 
processes outside the ballot, it works through party nominations. 
Split vote happens when a third party throws a monkey wrench in the 
process. Under those conditions, Approval fixes the problem quite as 
well as IRV, without the fuss. Bucklin *clearly* fixes it. If a 
supporter of a third party isn't going to add an additional vote in 
Bucklin, they are going to truncate in IRV. Pretty much the same with Approval.

However, what happens if that third party gets on the order of 
one-third the first-preference vote? This is where IRV can clearly 
break down. Approval as a plurality method can do poorly as well, 
though not generally as poorly. And, of course, if a majority is 
required, Approval should come out just fine.

>I think in the US scenario with voluntary voting, if  both Clinton and
>Obama ran McCain would have less chance of winning with IRV
>than with Approval or Range or Bucklin or any other reasonable
>method that springs to mind. This is because both Clinton and
>Obama have their enthusiastic supporters some of whom wouldn't
>bother voting if their favourite wasn't running, but if their favourite
>was running they would show up and (at the urging of their favourite)
>rank both Clinton and Obama  above McCain.

Nah. Most Clinton supporters will vote for Obama and most Obama 
supporters would have voted for Clinton. People get fired up about 
the Primaries and say all kinds of wild things. But, in fact, it's a 
Bad Idea to have Obama and Clinton in a race against each other 
outside of a party primary, and I think that even inside the party, 
it wasn't such a great idea, but that's our present system.

Chris, I predict high turnout in the next November election, unless 
it has become so much of a landslide that some don't bother voting. I 
think low turnout unlikely even if it looks like a landslide. People 
have had enough here, I'm pretty sure. Either Clinton or Obama would 
have won, but, I think, the fight would have been a *little* tougher 
with Clinton. She has extraordinary negatives, people seem to either 
love her or hate her. But faced with Clinton vs McCain? Obama 
supporters, even if they disliked Clinton, and some clearly do, would 
still have voted for her. Look at the Lizard vs. the Wizard, it's a 
case in point. Very high turnout (more than in the primary). The 
Wizard got hardly more votes than he got in the primary, whereas the 
Lizard (truly reptilian in his public image) swept it with almost 
two-thirds of the vote.

>IRV, meeting both Majority for Solid Coalitions and  Later-no-Harm
>has no  "defection  incentive" like other methods.

And the practical effect of these obscure strategic incentives? I 
think that with IRV people will almost all vote the same as they 
would vote in Bucklin. It's the same question on the ballot, really, 
though Bucklin adds the possibility of using multiple votes in third 
rank (Duluth) or in all ranks (what I'd propose, why not? If voters 
want to do it....)

> >What I actually wrote in my initial post on the 5  "fairness
> >principles in your paper (regarding IIA):
> >
> >In practical effect  *no* method meets this.Approval and Range can
> >be said to meet
> >Independence of Irrelevant Alternatives (IIA) only if the votes are
> >interpreted as the voters giving
> >ratings on some fixed scale that is independent of the actual candidates.
>
>>No, that's not correct. Perhaps it would be useful if you actually
>>state the version of IIA you are using. Usually, it refers to adding
>>or subtracting a candidate without changing the "preference order" of
>>the other candidates, but if you are going to use it with Range and
>>Approval, you have to modify it; the basic modification is that the
>>Range Votes or Approval Votes don't change, and all that happens is
>>that a new candidate is added to the ballot or taken off the ballot.
>
>If  the voters rate the candidates on some fixed scale that is independent
>of the candidates, then by definition the Range or Approval votes would
>be unchanged by adding (or removing) a candidate. What's "not correct"
>about it?

What's not correct is that this isn't the only possibility. The 
ratings could be for the complete candidate set (including the IA 
candidate). That isn't a scale "independent of the actual 
candidates." True, it's not fixed solely on the first candidate set. 
Alternatively, the scale can be the ratings with the irrelevant 
candidate not included, and then, again, the ratings don't change 
when the new candidate is added.

The only way that the irrelevant alternative affects the outcome is 
if the voters change their votes based on the candidate's entry. If 
they do, it becomes a tad misleading to all the candidate "irrelevant."

>>If voters are allowed to actually change their votes, *no method
>>meets IIA.* Simple proof: there is a candidate whose name is a
>>trigger for a long-hidden internal program that causes human beings
>>to fall into a trance when they contemplate whether or not to vote
>>for a candidate, and they leave the booth with false memories of what
>>happened (really happens with trance, sometimes, i.e, false memory).
>>The voters see this new name on the ballot, and regardless of how
>>they would have voted, they become incapable of voting, so all
>>candidates tie with no votes. And thus the winner could change.
>>
>> >On this perverse interpretation Approval and Range do not reduce
>> >to  FPP in the 2 candidate election,
>> >in violation of  Dopp's "fairness principle 4":
>> >
>> >"Any candidate who is the favorite [first] choice of a majority of
>> >voters should win."
>> >
>> >(approval or non-approval counts as "rating" on a 2-point  scale).
>>
>>Chris, you should look at Dhillon and Mertens, "Relative
>>Utilitarianism," where they purport to prove that Range Voting is a
>>unique solution to a version of Arrow's voting axioms that
>>accommodate Range Voting. Relative Utilitarianism refers to "votes"
>>which are "normalized von Neuman-Morgenstern utilities in the range
>>of 0-1. I.e., Range Voting. Warren Smith is actually not in outer
>>space on this (their work preceded his).
>>
>>Because of the normalization, in the two candidate case, Majority is
>>satisfied. Because vN-M utilities are modified by probabilities, it
>>gets complicated in the three-candidate case, where RU is considered
>>the unique solution. If I remember correctly. I'm hoping to help get
>>a popularization of Dhillon and Mertens prepared, it's needed. Smith
>>calls their use of symbols "Notation from Hell." And he's familiar
>>with the conventions!
>
>It might I suppose be of some interest as an intellectual curiosity, but
>Warren Smith is a mathematician who says he doesn't understand it,
>and my understanding of  Mathese is nil.

That's correct. You could help write an interpretation of it. 
However, the basics are pretty clear. They are describing the 
aggregation of preferences by simple addition of "von 
Neuman-Morganstern" utilities normalized to the Range of 0-1. And 
they show that this is a unique solution to the problem of Arrow's 
theorem (as they modify the "axioms." While I agree that there is 
lots of room for error there -- does anyone know how this paper was 
received? Was it a big yawn? (That can happen with something 
brilliant, but also with a dud.) Or did someone refute it?

My *sense* of the proof is that they got it right, their approach was 
correct. But that is far short of proof. That's why I'm interested in 
translating it to English. I don't think the math is that complex, it 
is the *notation* of that field that makes it just about unreadable in detail.

>Range Voting isn't normalised.  Normalised Range Voting doesn't
>meet  Kathy Dopp's first  "fairness condition" (IIA).

Range Voting, in practice, will be normalized unless voters choose to 
abstain (fully or partially). That's the point. Early Range 
implementations will probably be Range 2 or 3: something like:

3 Yes.
2 Acceptable.
1 Poor, I'd rather see a runoff.
0 No.

(If you do not vote Yes for one candidate and No for another, your 
vote will be weak, effectively a half-vote or less)

or something like that.

Is Approval normalized? Range is approval with fractional votes 
allowed, that's all. Thus it does become *possible* that voters will 
cast weak votes, but this is harmless.

Early on in the discussion of the Center for Range Voting, there were 
proposals to normalize ballots. I.e, if, on the ballot described 
above, the voter had voted 1 and 2 only, the ballot would be 
normalized to 0 and 3. But it simply adds complexity while reducing 
voter options. If the voter wants to vote 0 and 3, the voter can!

When you consider that the candidate set, with write-ins, is broader 
than what appears on the ballot, we can consider zero to be the 
rating of a host of candidates not shown. As I'd have Range work (sum 
of votes), it's No to all write-ins, for example. At the top, it 
would represent an idea of the voter that there is a better possible 
candidate, but the voter doesn't bother writing that name in because 
the voter knows about snowballs and hell.

The point is that if voters vote accurately their vN-M utilities 
(which are absolute utilities modified by election probabilities) and 
they are normalized, Range voting is, quite arguably, the ideal 
election method. Whether they will vote accurately is another story. 
We don't say that a Condorcet method fails Condorcet because the 
voters might not vote sincere preferences!