[EM] Does IRV elect "majority winners?"

[Abd ul-Rahman Lomax]

At 03:31 PM 1/6/2009, Dave Ketchum wrote:
>On Mon, 05 Jan 2009 00:19:29 -0500 Abd ul-Rahman Lomax wrote:
>>At 10:28 PM 1/4/2009, Dave Ketchum wrote:
>>
>>>On Sun, 04 Jan 2009 16:16:14 -0500 Abd ul-Rahman Lomax wrote:
>>>
>>>>
>>>>Perhaps. Perhaps not. That can be a *lot* of preparation, and 
>>>>people are busy, many don't already, find time for voting. Bullet 
>>>>voting is simple, it can be relatively easy to know who your favorite is.
>>>
>>>
>>>Agreed that bullet voting is often appropriate.
>>>
>>>Only occasional elections provide reason for some voters to do more ranking.
>>
>>And only certain voters. It's relatively uncommon that there are 
>>more than two frontrunners, and most voters know who they are. 
>>Under those circumstances, the only strong reason not to bullet 
>>vote is if you prefer someone other than a frontrunner, and care to 
>>express it. The argument for Plurality would be that the system 
>>shouldn't be encouraging "useless candidates" to run at all! That 
>>is, since vote-for-one usually works, and the only reason it 
>>doesn't work (usually, and even this is fairly unusual) is that 
>>some silly voters will throw away their vote on a candidate who 
>>can't win, why should we respect the unexpressed wishes of those 
>>voters? After all, they had their chance! We don't run elections as 
>>a popularity contest, i.e., so minor party candidates can brag 
>>about how many votes they got....
>
>Your argument for Plurality is empty:

It's not my argument. It is a possible argument that could be made, 
I've seen similar arguments made. I do not support Plurality, but 
Open Voting (i.e., Approval), which is Plurality with multiple votes 
allowed, is better. Plurality as a primary method is even better, in 
my opinion, and Open Voting would be even better than that, etc.

Best single-ballot method would be Open Voting with fractional votes 
allowed. I.e., Range. To do better than that requires allowing 
possible runoffs to deal with the relatively rare situations that 
Range makes a bad choice.

>Condorcet certainly costs more for the system than Plurality.  Costs 
>bullet-voters nothing - provides a service to whichever voters like 
>to do more than bullet vote.
>      Actually can be a service to candidates.  Clinton and Obama 
> had to try to kill their competitor's campaign for the Democrat 
> nomination they could not share.  A similar race in Condorcet would 
> let them both get nominated and have a more civilized fight as to 
> which should be ranked higher than the other on the ballot.

Very bad idea. It dilutes their election resources. Plurality is not 
the only reason to have a party system, and to only nominate one 
candidate from a party. It's a problem that the nomination process 
can be so divisive, but that's a different issue.


>>>Even if runoffs are possible/expected, it is wise to vote 
>>>carefully in the primary to minimize possibility of bad choices 
>>>getting to the runoff.
>>>
>>>>And when it isn't easy to know, having trouble deciding between 
>>>>two, Open Voting (Approval) allows a simple option: vote for both!
>>>
>>>
>>>What is important is that Condorcet, unlike Approval, permits 
>>>voting for both Good and Soso, while indicating that Good is preferred.
>>
>>Right. However, with American Preferential Voting (Bucklin), you 
>>*can* indicate your preferences. My point is only that equal 
>>ranking, if allowed, can be, actually, more expressive.
>
>Bucklin deserves more thought as a competitor to Condorcet.

It doesn't have to be a "competitor." Rather, condorcet analysis can 
serve as one of a number of possible runoff triggers.

Definitely, Bucklin deserves more thought. And more research, 
including better knowledge of the history. What the hell happened? We 
had an advanced voting system, in a *lot* of places, and the FairVote 
explanations of what happened are facile and self-serving and seem to 
be mostly speculation.

>>Looking over the ballots from Burlington, as I just did, I'm struck 
>>by how many voters do seem to imagine that their votes will be 
>>counted! Overvotes are more common than I'd expect if they were 
>>mere slips. It is very easy for me to imagine that voters think 
>>that if they vote for more than one candidate in one of the ranks, 
>>why, the votes will be counted, they are merely saying that, for 
>>first preference, they prefer either the Progressive or the 
>>Democrat, or some other combination. The fact is that if such votes 
>>were counted, they'd make sense, even in IRV. (Allowing equal 
>>ranking turns IRV into a much better system than without it.)
>
>How do you count equal ranking in IRV?  If I vote X>A=B>Y, A and B 
>become visible to the counters at the same time - what does this do 
>to deciding what candidate is next to mark lost?

The method doesn't change. Yes, A and B become visible at the same 
time. So? It's standard Approval voting, only in ranked rounds. The 
candidate with the lowest number of votes is eliminated. If there is 
a tie, then there are standard tie-breaking methods. (This problem 
with intermediate ties is only a problem with IRV, I think. Otherwise 
they don't matter.)

>>Those concerned about Later-No-Harm can simply avoid equal ranking!



>Approval, Plurality and IRV are distractions from need to pick a 
>live destination.  I see need to compare, more carefully, Condorcet 
>vs Range vs Bucklin.

Definitely. We need more experimentation. It's a shame that FairVote 
got stuck on one particular goal and one particular strategy for getting there.

>A "small minority" cannot win in Condorcet, except for the tortured 
>case of a zillion candidates with at most a small minority voting for any one.

Oh, come on! Obvious counterexample: everyone bullet votes. Condorcet 
winner is the Plurality winner, and this could be a small minority. 
Yes, of course, this requires more than two candidates! With three, 
the smallest it could be is one-third, etc.

>I am still trying to promote series thought as to need for a 
>majority for other than Plurality or Approval.

Serious?

It's a basic democratic principle. In standard deliberative process, 
*no decision is made without the explicit approval of a majority of 
those voting.* Anything else is a compromise.

>Go to RRONR and you see demand for a supermajority for some 
>cases.  I still question actual need for a majority for all election 
>methods (other than Plurality, which has true need).

Robert's Rules requires it; organizations can, of course, make their 
own rules, so an organization can authorize election by plurality. It 
is not recommended. Preferential voting can be used to make it more 
likely to gain a majority.


>>>Condorcet has less need, for its voters can rank all the 
>>>candidates they approve of.
>>
>>Less need, perhaps. It's possible that the Condorcet winner is a 
>>winner by a majority over all other candidates. Advanced voting 
>>systems, in general, can be used as devices to avoid unnecessary runoffs.
>
>Try:
>      77A; 76B; 75C; 74blank
>These voters like bullet voting and none ranks any competitor of 
>their first choice (though they could have ranked more if they 
>approved of any).  Why not let A win?  If you insist on a runoff, 
>who deserves to be on that ballot?

Blank ballots are considered "scrap paper" by Robert's Rules. If a 
voter wishes the ballot to be counted, the voter can add any mark to 
it. This, then, serves, effectively, as a No vote on all the candidates.

I don't like Plurality, you know that; I've merely made the point 
that it tends to work better, because of common strategic voting, 
than might otherwise be expected.

In any case, Robert's Rules simply says that the election fails. They 
do not prescribe -- or permit -- any candidate eliminations. However, 
candidates may withdraw, and new ones may be nominated. It's a *new 
election*, not a runoff.

However, runoffs are *similar.* The best runoff rules allow 
write-ins, so there aren't actually eliminations; consider the top 
two being on the ballot as a "suggestion" to the voters.

There is no good basis for allowing A to win. A has the support of a 
third of those voting, only. It's true that in this situation, there 
isn't much basis for choosing between A and B, but we don't know what 
will happen in the runoff. By the way, a three-way tie like that is 
extraordinarily rare and very unlikely. If you had presented

41 A, 30 B, 29 C, it would have been more realistic. Top two runoff 
would, of course, put A and B on the ballot. There is a possibility 
that C is the victim of center squeeze.

There is a *lot* more involved than I've seen in most analyses. 
Analysts simulating TTR have typically assumed a fixed electorate and 
fixed preferences, which is actually preposterous. Neither holds, normally.

>      77A; 77B
>Here I recommend the rules say "flip a coin" - and anyone who 
>objects to that should get to finance the runoff.

Some rules do say that. The difference between either of these 
winning and a majority is small enough that it may not be worth a runoff.

>>But Condorcet methods without some test of acceptance can fail 
>>rather badly. It could be that the majority would reject the 
>>Condorcet winner, given a choice of electing the person or running 
>>the whole election over, and the only reason it appeared that this 
>>was the Condorcet winner was that, say, voters fully ranked, and 
>>some of those who formed the majority in some of the wins were 
>>actually saying that this was the second-worst candidate, which 
>>could be pretty bad....
>
>I do not understand "fail" here - Condorcet permits ranking multiple 
>candidates by any voter who chooses to.

Look at the Burlington votes. There is no way to tell if a vote is an 
acceptance or not. If I vote that Adolf Hitler is better than Genghis 
Khan, does this mean that I have voted for Adolf Hitler? Condorcet 
methods treat the vote as if it was just that, unless there is an 
approval cutoff specified.

Range has the same problem, actually, though it's rather easy to, for 
example, set 50% rating as an approval cutoff.

>I choke on any encouragement of "fully ranked".  For example, 
>ranking the next to last is usually not worth the pain for it 
>affects only the relationship between the last two.

That's right. What's fascinating is how many voters in Burlington 
used all five ranks, often in rather strange ways.

>>If you are going to use pure ranking, with a single ballot, 
>>Cordorcet is king, so to speak. But that's an artificial limitation.
>>I do agree that if a better method is used (like Range or Score 
>>Voting), a Condorcet test should be used to ensure that the 
>>Condorcet winner isn't unconciously rejected. A runoff is needed to 
>>test this. If we have a conflict between a Range winner and a 
>>Condorcet winner, what has happened is that there may be a majority 
>>-- or a plurality -- with a weak preference, against a minority 
>>with a strong preference. Social utility theory would say that the 
>>Range winner is better; but the votes imply that a majority would 
>>reject that winner in favor of the Condorcet winner. For reasons 
>>I've described many times, I claim that a genuine Range winner will 
>>prevail in a runoff, and that this would be the best result. But 
>>occasionally, that would not happen, and the Condorcet winner would 
>>prevail; what might be happening here is that the majority is 
>>saying, "No, we do *not* want to give up our right of decision as a 
>>majority, we don't accept the Range results as valid, or, for 
>>whatever reason, we, in the majority, don't want to make the sacrifice."
>
>Ugh!  How do you get valid multiple method results without imposing 
>some kind of load on the voters?

The "load" is one voluntarily assumed. Don't want to face a runoff? 
Vote strategically! That is, make your compromise, don't bullet vote, 
be sure to give adequate support to a frontrunner.

If Range is used with a 50% approval cutoff, this would tend to 
encourage voters to vote at least an approval vote for a frontrunner. 
We get majority failure when voters for minor candidates bullet vote.


>>>BTW - cost of ballot support for complete ranking can tempt limits 
>>>on Condorcet ranking.  How bad should we complain if offered 3 as in RCV?
>Three seems like a good minimum, with cost vs voter desires 
>controlling whether more get implemented.
>>It's not the worst problem. Center squeeze is the worst problem.

Three is probably plenty for Bucklin. If multiple votes are allowed 
in all ranks instead of just the last, it is even more adequate. 
(Remember, three expressed ranks means four actual ranks. Multiple 
votes in traditional Bucklin was allowed in the last two of the four. 
All optional-ranking methods including plurality allow -- or require 
-- multiple votes in the lowest rank.)

I've taken to calling methods which allow independent ranking or 
rating of all candidates "Open Voting" methods, with Open Voting 
itself, with no qualifier, being Approval Voting. It's my observation 
or at least my opinion, that all methods improve with the allowance 
of equal ranking or rating (with some necessary details).

In Borda, for example, if a voter equal ranks some candidates, the 
total number of ranks must remain the same, and the voter must be 
able to top and bottom rank with the same score applied as if all 
ranks were occupied. This may be automatic with practical Borda 
ballots and tossing the no-overvoting rule; imagine the Burlington 
ballot, which had five ranks for five candidates on the ballot, plus 
an implied rank -- to allow for the sixth candidate, a write-in. 
There were quite a few votes which made sense, some voters seemed to 
think that equal ranking would be allowed and effective. Some equal 
ranked top preference. Did they realize that this would cause their 
entire vote to be moot? Anyway, that could have been a Borda ballot, 
assume a score of 1 for the lowest rank and 5 for the highest.