[EM] IRV vs Plurality

[Abd ul-Rahman Lomax]

At 01:44 AM 1/15/2010, robert bristow-johnson wrote:
>>And it only takes a few people to realize this to start building
>>the structures. It is *not* necessary to convince the masses. That
>>will come later, after they have examples to look at, which is what
>>most people need.
>
>all of the above resonate very closely to what i've been thinking for
>about 10 months.

Thanks, Robert. Stick around :-)

>>So my political recommendations are based on what is already known,
>>what has been already tried, with only minor variations beyond
>>that. Approval voting, one might note the critics state, can
>>default to Plurality if most voters vote for their favorite and
>>leave it at that. *That's fine.*
>
>no it ain't.  (Plurality is not fine.)

Always remember that "fine" is a relative adjective. Fine compared to what?

If the existing system is Plurality, and a change costs little or 
nothing, that the new system defaults to Plurality if most people 
vote it the same way means that the reform Does No Harm. That's what's "fine."

Further, what we are really talking about is Approval voting, right? 
In a situation where voters don't use the facility to add additional 
votes, it may indeed be fine, because those sincere Plurality votes 
(which with Approval actually will be sincere, because there is no 
reason to vote otherwise in Approval) will do the job and select, 
quite likely, the ideal winner by any reasonable standard.

We are accustomed to thinking of the situations where plurality 
fails, but those situations can be *largely* handled by simply 
allowing additional approvals. We forget that Plurality *usually* 
works, within the context where it is applied. When the context 
changes outside that functional envelope, Plurality obviously has 
defects, even serious ones.

We must, also, understand that voting systems don't exist in a 
vacuum. There is a great deal of pre-election activity that 
influences how people vote, and this activity frequently operates to 
make Plurality more functional that straight theory might indicate. 
Further, even the pathology of Plurality, such as the spoiler effect, 
also has a positive function, and if we don't understand that, we may 
indeed try to eliminate it and get rid of the baby with the 
bathwater. We need to understand that there is a baby in there, so 
that we can protect the baby and only get rid of the dirty water.

>(it's fine and good for us to have different positions.  i just
>think, and have for decades, that in a multi-candidate race, the
>problems with FPTP are too well known to revert back to that because
>IRV doesn't cut the mustard.)

So don't revert back to that. Approval, however, is almost the same 
as FPTP, there is just one very small rule change, eliminating a rule 
that was a Bad Idea in the first place, it's an historical accident 
that voting for more than one was *ever* prohibited. The rule makes 
*some kind of sense* in a repeated ballot situation, where multiple 
voting is less necessary, compromise can happen in other ways. But my 
sense and actual experience is that approval is quite efficient in 
that situation, where it is also quite possible to do an approval 
poll or a set of polls, and then actually vote Yes or No to an 
implementation motion. For some reason that I *don't* understand, 
Robert's Rules discourages polling, but my guess is that they are 
thinking of different kinds of polls than what would be used to 
choose from multiple options. They are probably thinking of polls on 
Yes/No questions, which are indeed a waste of time, just vote on the 
question! If no majority, it doesn't cause damage. If majority, there 
is a decision, and anyone voting for it, if they think that better, 
may request reconsideration, reopening the question. A poll is indeed 
a waste of time. But to quickly sort through multiple options, 
approval polling is quick, can be done by voice vote or show of 
hands, for example. If it's a written ballot, Range voting becomes 
more of a possibility and can further increase negotiation efficiency.

We need to back up and look at why we vote. What's the goal? And then 
we re-examine the various aspects of voting systems from that 
perspective, not from a perspective that incorporates a host of held 
assumptions about what's good and what's not.

>>Count All the Votes. And then, I claim, we should use the votes
>>that are counted, and political theory generally says that Approval
>>Voting, which is simply a matter of Counting All the Votes, is
>>quite a good method, superior to plain Plurality, and simply
>>defaulting to Plurality if people just vote for their favorite.
>
>i think the folks on the edges want a way to express a preference for
>their guy that will actually count against their fallback guy if the
>race were to become such that's between the two of them.

Sure. But look where this matters, with, say, Bucklin. Multiple 
majorities. If there is a multiple majority, we might back up and 
look at the higher preference votes. If there is a difference between 
the majority winner after opening up additional approvals, and the 
leader before, which defines the situation Robert is concerned about, 
this will detect it, and a runoff can be held.

Alternatively, and possibly dependent on margins, we could decide to 
follow the traditional most-votes assumption. But it's largely moot. 
Multiple approvals both with a majority is probably very rare, unless 
the candidates really are close to each other in broad perception, in 
which case it doesn't matter a great deal which one is chosen. But "a 
great deal" could be better defined by actual vote patterns.

If the method is Range Bucklin (the fractional vote equivalent of 
Bucklin, the same effective sliding down with rounds of approval 
cutoff), then it becomes more possible to detect a problematic 
multiple-majority from an insignificant one, one where it would be 
better to just go ahead and complete the election.

Consider this with ordinary Bucklin, Robert: you have a strong 
preference for your favorite so you are concerned about adding that 
second preference vote.

A. It's runoff Bucklin. Fine. Don't add any more second preferences, 
leave that for a runoff. Or follow strategy B for any candidates 
where you would prefer to finish the election rather than seeing a runoff.
B. It's deterministic. So skip the second rank and place your vote in 
third rank. Maximized LNH protection, while still, in the end, 
allowing your strategic approval(s).

It all depends on your preference strength. With Range/Bucklin, you 
could directly express the strength, and the method basically does 
the negotiating for you. My sense is that with a proper understanding 
of strategy, not difficult, you would vote effectively, and it would 
be a sincere vote. Sincerity expression is encouraged if it's 
runoff/Bucklin, and the effect of that is a somewhat increased 
likelihood of runoffs. But we must say that those runoffs happen if 
the electorate effectively prefers them. They happen, if the 
threshold is a majority, if a majority of the electorate withheld the 
compromise votes necessary to find a majority result.

The best details depend on context, there isn't necessarily a 
universal answer. When is "good enough" good enough?

>   with
>Approval, they still have to strategize "do I vote for both or do I
>vote just for my favorite?"  actually (Terry knows about this), in
>Vermont, the State Senate races are sorta weird.

Sure. And with a majority requirement, there is a sensible strategy 
behind this, easy to understand and state. In Vermont, with the 
gubernatorial election, the strategy question is "Would I prefer to 
add an additional approval, or to see this election go to the 
legislature for one of the top three to be picked?"

And the answer to that question depends on preference strength! As 
well as our trust in the legislature. Trust the legislature a lot, 
vote your favorite and you can leave it at that. Your vote is 
effectively a vote against all other candidates.

Don't trust the legislature, add an additional approval if you'd 
prefer to see that result to the legislature choosing.

It's much simpler than you think, Robert.

>unlike the
>Representatives that have legislative districts drawn (and have a
>single winner for each district), the State Senate candidates run at
>large for the whole county.  being that Burlington is the largest
>city in the state, our county is also the largest, i think.  we have
>6 state Senators and the rules are we can vote for up to 6 on a
>single ballot, and the 6 highest vote getters are elected.

Cumulative voting. Not bad, if the voters are organized. Actually 
fair if they are.

>   usually a
>party puts out 6 candidates and one might think that they could just
>plug the 6 of their party unless they like to cross over for some
>particular candidate they like.  but if there are 4 or 5 candidates
>that are "okay" with me, but one or two candidates that i
>particularly like (and i might consider an underdog), i will end up
>"bullet voting" for just that one or two candidates because i want
>them to win badly enough that i don't want to risk having another
>person of the same party displace them in the top 6.

And then the party vote is split. Rather, it would be better for the 
party to determine how many candidates it could be likely to 
collectively elect, and only put up that number of candidates. It 
would use utility-maximizing methods to pick its own candidates, so 
that most members would really be willing to vote for them with all 
their votes. It would suggest exact voting patterns to its members, 
and from experience, it would know how the members respond, so it 
could maximize effectiveness.

There are, of course, much better election methods for choosing the 
six. Cumulative voting can produce a rough proportional 
representation, but only with organized voters, otherwise it's quite 
hit-or-miss. Asset Voting would be perfect, but, hey, that's a truly 
radical reform, in spite of allowing every voter to vote sincerely, 
with a minimum number of votes being wasted, and other benefits.

So there are good PR methods. Reweighted Range Voting. Proportional 
Approval Voting. And, of course, STV, which ain't bad for PR, but it 
tends to require much more knowledge on the party of voters, or the 
use of voter information cards which gives much more power to party 
leaders and probably subtracts power from independent voters.

>   but i think a
>ranked ballot would do well for that and maybe STV is as good as we
>can do in a multi-winner case.  i dunno.

Well, it's known. We can do better. But STV is good, *in some ways.* 
It breaks down the fewer the seats being elected, the last election 
is effectively IRV, but the first seats are almost certainly optimal. 
Asset was proposed as a tweak on STV that would result in far fewer 
wasted votes, so an STV ballot and counting could be used, but my 
judgement is that Asset is so powerful a technique that a simple 
Approval ballot would be fine, which, you will note, fully empowers 
voters who simply vote for their favorite and leave it at that. I 
only would even allow additional votes because of some very slight 
theoretical improvement under some situations, it can simplify 
certain voter decisions, and counting "overvotes" reduces ballot 
spoilage with likely higher effective expression of voter intent.

>   because i can see that it's
>possible (but it doesn't always happen) for Condorcet to order
>candidates from the top (the Condorcet winner) to the bottom (the
>Condorcet loser), maybe picking the top 6 using Condorcet ordering
>would be best, but i dunno.

The Condorcet criterion doesn't really apply to multiwinner, though I 
think there is some way to extend it. Always remember an important 
thing: if we want fair representation, if a quota of votes elect a 
winner, those votes are fully represented and should be considered as 
spent. So if you are going to use the condorcet criterion, you'd need 
to look in sequence. Devalue the ballots which have been used to 
elect a C winner, then look at lower preferences, which are then 
fractionally counted. I'd say you should look at how multiwinner STV 
works, in a decent application.

>   my political licks have just been about
>IRV vs. Condorcet vs. Plurality (or the old 40%+ rule) in the 
>single- winner case.  i'll fight the multi-winner battle some other time, and
>i just don't know yet what side i'm on.  i might become STV.  sure,
>it's elegant to have the same theory for both the single-winner and
>multi-winner case (and IRV is STV for single-winner), but i think
>that IRV has enough problems that i just cannot support it over
>Condorcet, if given the choice.

I'm suggesting that if you look at Bucklin, including looking at the 
history, you would realize that it's truly a powerful method with 
much more implementation history in the U.S. than IRV. It is far 
simpler to count than IRV or even Condorcet methods.

I don't think it's been adequately studied in the simulations, which 
tend to be oversimplified in how they set up voter voting strategy. 
And if there is a majority requirement, I'm pretty sure that hasn't 
been studied yet. Bucklin, however, works, we know that, because it 
worked, and the only thing it didn't do was find majorities if too 
many voters didn't add additional preferences, which is an *intrinsic 
problem*, not avoidable unless voters are coerced, which is 
unconstitutional. It would still outperform Plurality and IRV, 
resolves the spoiler effect, is probably not subject to Center 
Squeeze, in itself, etc.

And it's really easy to vote, strategy isn't difficult.

Bucklin is really instant runoff Approval and can truly be voted that way.

With IRV, you add an additional lower ranked vote if you want your 
vote to be counted if your candidate is eliminated.

With Bucklin methods, you add an additional vote if you want your 
additional votes (additional approvals) to be counted if your higher 
preferences are not going to win with a majority. The difference 
between "eliminated" and "not going to win by a majority without more 
votes from other voters" is important.

Sure, when you allow your additional approvals to be expressed, you 
then create a possibility that your lower ranked vote will allow your 
lower preference to beat your favorite. You have not, with Bucklin, 
voted for the lower preference over your first preference and you 
have in fact abstained from that particular pairwise election (after 
it's clear that nobody is going to win with a majority). But what you 
and your candidate gain is that other voters may push your favorite 
over the victory margin. You lose a little, and you might gain a lot.

As to losing a little, suppose it was a situation where you would, in 
fact, be losing a lot. I.e., if it happens that your second rank vote 
causes your second favorite to win, would you regret it? *How much* 
would you regret it. Do remember that if your vote has that effect, 
the regret has to be considered at half value, because we are talking 
about the difference between a tie and a decision. Perhaps I should 
do the game theory analysis....

If, considering all this, you believe you would regret your second 
rank vote, that your preference strength is too strong to allow you 
to feel comfortable with this, then not adding the lower ranked vote 
is quite an appropriate strategy. With a majority required, it's quite safe.

With Range/Bucklin, which I don't see as a politically practical 
first step, you'd essentially control your approvals using 
specification of preference strength. That's what Range/Bucklin does, 
and the maximimally effective strategy there is quite likely the most 
accurate expression of your actual preference strengths among the 
major candidates. But more study is needed, for sure. Nobody has 
studied this method, to my knowledge.

Then, under some circumstances, the Range information could be used 
to avoid runoffs, or to resolve multiple majorities, which, if you 
think about it, is the situation where the voting problem you mention 
actually becomes important. (If a majority is required.)

>Abd ul, my position has always been consistent in the last 10
>months.  i fully support the *goals* of IRV because i think they are
>the same *goals* that we have with Condorcet or any of the other
>ranked-ballot methods.  those goals were, for me, boiled down to 4
>salient principles that i outlined in my paper that i have plugged
>here at least a few times.  those principles are (i'm repeating 3,
>but hey, bits are cheap):

Thanks. Specifying the basis for your opinions is very helpful.


>___________________________
>
>1. If a majority (not just a mere plurality) of voters agree that
>candidate A is
>better than candidate B, then candidate B should not be elected.

And I've shown that this is not a basic principle, there are 
situations where it's obviously a suboptimal result. Even seriously 
suboptimal. It sounds good because we don't ordinarily have the 
information in elections to notice the problem with it, we would if 
we used Range methods or did good Range polling.

>2. The relative merit of candidates A and B is not affected by the
>presence of a
>third candidate C. If a majority (not just a mere plurality) of
>voters agree that
>candidate A is better than B, whether candidate C enters the race or
>not,
>indeed whether candidate C is better (in the minds of voters) than
>either
>candidates A or B (or both or neither), it does not reverse the
>preference of
>candidate A over candidate B. If that relative preference of
>candidate is not
>affected among voters, then the relative outcome of the election
>should not
>be affected (candidate B winning over candidate A). In the converse,
>this
>means that by removing any loser from the race and from all ballots,
>that
>this should not alter who the winner is.

This is a wordy version of IIA, and the vulnerability of a system to 
IIA violation is a matter of great controversy. Range isn't 
vulnerable to IIA if we just remove a candidate from the votes, nor 
is Approval, for example. But voters might choose a different 
strategy, making it vulnerable in that sense. Basically if the 
removal of a candidate causes *any* change in how voters make their 
decisions, it will cause an effect from the removal of what is called 
an irrelevant candidate. In practical reality, we must recognize that 
no method is completely invulnerable to IIA, because of the effects 
of voter perception. Let me provide the reductio ad absurdem:

The voters all very strongly prefer A, except for a few, so even if 
the voters have the option of ranking or rating other preferences, 
they mostly don't bother. But a few voters for whatever reason don't 
like A, or like A little enough that they add additional approvals. 
(Let's say the method is approval). Easily, if A is eliminated, the A 
voters will take a stronger look at the remaining alternatives, and 
how they will pick from among them cannot be predicted as a general 
rule. So it can *easily* happen that the elimination of A can shift 
the result from B to C, for example.

IIA is considered by many experts to be the weakest of the Arrovian 
criteria. But I'll agree that a good method won't be vulnerable to 
IIA if all you do is remove the A votes, and don't do a separate 
election. And Range, for example, satisfies that.

>3. Voters should not be called upon to do "strategic voting". Voters
>should feel
>free to simply vote their conscience and vote for the candidates they
>like
>best, without worrying about whom that they think is most electable.
>Voters
>should be able to vote for the candidate of their choosing (e.g.
>Perot in 1992
>or Nader in 2000) without risk of contributing to the election of the
>candidate
>they least prefer (perhaps Clinton in 1992 or Bush in 2000). They should
>not have to sacrifice their vote for their ideal choice because they are
>concerned about "wasting" their vote and helping elect the candidate
>they
>dislike the most. As an ancillary principle, a candidate should not
>have to
>worry about electing his/her least desirable opponent by choosing to run
>against another opponent that may be more desirable.

Now, consider this from a basic perspective. Voters don't have votes 
in their head as some kind of immediate "sincere" vote, voters don't 
even have relative opinions about candidates, except usually they 
will have a favorite. What you want, I'd think, is a system that will 
allow a sincere expression of a first preference, or maybe even a 
full expression of preferences, all the way down. There are such 
systems. But this is the problem: say, with Range, you accurately 
express your relevant preferences, normalizing your vote so that it 
has maximum impact. This means that you vote at at least one 
candidate with maximum vote (100%) and at least one with minimum vote 
(0%). Suppose, however, there is an irrelevant candidate! Somehow, 
Satan got on the ballot. Satan doesn't have a prayer of winning. For 
sure -- he's Satan!

So, do you vote Satan 0% and everyone else 99% or 100% by comparison? 
Only if you think there is a good chance of Satan winning! There is 
nothing really wrong with your sincere vote, but it doesn't allow you 
much voting power between the real candidates. Okay, so we use a 
condorcet system. But this system can't express preference strength, 
so it can really screw up.

But people are quite accustomed to this problem. They don't alter 
their votes in Plurality through the presence of irrelevant 
alternatives, not in the meaningful sense. And they will vote in 
Range and Approval quite like that. They will disapprove of Satan and 
vote in the rest of the election quite as if Satan weren't on the 
ballot. They are exerting full voting strength against Satan, and in 
a runoff system, they couldn't do more to prevent Satan from being 
elected. The issue is how they vote for other candidates. Quite by 
the same argument, they won't waste positive voting strength in 
irrelevant pairwise elections. So, with Range, they might prefer 
Nader to Gore, but a vote of Nader 100% Gore 50%, even if that is 
"sincere" would have wasted half their vote in 2000. So they would be 
more likely to vote, say, Nader 100%, Gore 99%. Because Gore is a 
*relevant* candidate, a frontrunner, and the real race is likely to 
be between Gore and Nader.

Bucklin allows full strength voting in the relevant race while still 
allowing clear preference expression. Range/Bucklin would work in a 
similar way. You wouldn't waste any of your vote by adding a lower 
relevant preference and a higher no-hope preference.

>4. Election policy that decreases convenience for voters will
>decrease voter
>participation. Having to vote once for your preferred candidate, and
>then
>being called on to return to the polls at a later date and vote again
>for your
>preferred candidate (if he/she makes it to the run-off) is decidedly
>less
>convenient and we must expect that significantly fewer voters will
>show up
>for the run-off. Or, if your most-preferred candidate did not make it
>to the
>run-off, the motivation to return to the polls to vote for a somewhat
>less
>preferred candidate (or to vote against a much disliked candidate) is
>reduced and fewer voters show up. Electing candidates with decreased
>legitimate voter participation cannot be considered as democratic or as
>indicative of the will of the people, as electing candidates with
>higher voter
>participation.

Is "fewer voters" a bad thing. Doesn't it depend on what kind of 
sample it is? It is standard practice in democracy that decisions are 
made by the members of an organization or society who show up to 
vote. Why? Repeated balloting in a standard direct democracy can be a 
pain in the neck, very inconvenient. But that very inconvenience has 
a function.

Here, though, I'm addressing the idea that runoff elections are 
harmful. Obviously they have a cost and obviously either they or the 
primary are likely inconvenient. Is that a good thing or a bad thing? 
I've mentioned that some have repeated arguments here that are 
essentially repeats of the familiar FairVote arguments used to sell a 
fish bicycle to eskimos. (How's that for a hybrid, eh?)

Very obviously as well, many jurisdictions were willing to put up 
with the cost and inconvenience in order to find majorities, and 
there really isn't any way to guarantee a majority, there are merely 
ways to make a majority more likely. (The guaranteed majority by 
preventing any more than two candidates in a runoff election, no 
write-ins allowed, is a faux majority, essentially coerced. A real 
majority is a voluntary approval by a majority of voters of an 
outcome, when they were free to express themselves otherwise.)

But here is the new information that should be digested, it seems to 
be new because I haven't come across this elsewhere, even though I 
consider it obvious. Like a lot of the stuff that I've come up with 
that takes a few years of mention before people start saying ... 
hmmm.... maybe there is something to this after all.

Robert doesn't like the consideration of preference strength, he 
really hasn't said why. But the only reasonably objective standard 
that anyone has come up with for judging voting system quality has 
been to postulate absolute voter utilities and predict behavior from 
them as simulated voters encounter a voting system in a series of 
simulated elections. Voting system criteria address the limits of a 
system's behavior but not its normal behavior. Lots of assertions and 
claims are made about strategic behavior that aren't based on solid 
ground, only on speculation and possibility.

In any case, if we postulate absolute utilities, never mind that 
there is no clear way to extract them from voters, and then we make 
reasonable predictions about how voters will behave, and we vary this 
behavior according to, in one approach, varying degrees of "sincere" 
and "strategic" voting behavior, we can see how a voting system 
behaves under varying conditions, across many elections. In order to 
compare voting systems, we need to be able to compare outcomes, and 
when we take the "criterion satisfaction" approach, except for a 
criterion that doesn't have an accepted name yet, we really are 
reduced to saying "My satisfied criteria are more important than 
yours! No, mine are more important!"

In order to consider voting system quality we have to back up and 
consider the purpose of elections. If we define different purposes, 
we may come up with different methods that most satisfy the purpose we choose!

For example, consider a possible application for voting. A society 
exists, say, the members of which can be classified into factions 
defined by a strong leader. Without some means of making decisions 
that would reflect what will happen if it boils down to which faction 
has the strongest army, the factions will in fact resort to this 
approach, and the faction with the strongest army will prevail, in 
general. So what if a vote is taken where every citizen can vote, and 
the faction with the most members can be expected, on average, to be 
able to field the strongest army, other things being equal. Presto. 
Plurality voting. Add coalitions of factions that decide to 
cooperate, and those factions make decisions internally by the same 
method, generally raising the quality and stability of the plurality 
votes. That leads to a 2-party system, with which, of course, 
anything more complicated than Plurality isn't necessary.

But there are better models and better goals. Long ago, we came to 
the concept of a majority faction, and we decided to respect the will 
of the majority, when it was expressed. We also allowed the subset of 
people who actually voted to represent the rest of society. Note 
something: people who don't vote, on average, don't care about the 
outcome as much as people who vote. Low voter turnout is, more than 
anything, a rational response by voters to a situation where they 
don't believe that their vote will make an important difference. That 
view is commonly derided, but it's probably more true in many 
situations than not. This disinterest in the outcome may be a sign of 
serious discontent coupled with a belief in the impossibility of 
changing the system, but that has its greatest impact in what 
candidates arise and how the public supports them. Or doesn't. But, 
alternatively, and this is a reality in many places, and quite clear, 
it reflects a contentment with the decisions made by the more highly 
informed and interested voters who do actually vote. As long as the 
group of interested voters doesn't stray too far from the interests 
of the entire electorate, this situation can continue. The voters 
aren't a sample of the voters, exactly, because they are, in general, 
more informed than the disinterested voters. They are not necessarily 
more informed than some of those who don't vote because they dislike 
all the options.

If voting is totally convenient and quick, no pain at all, will this 
improve results? Quite likely not! My sense is that, absent certain 
precautions, the results would get worse. Large uninformed and 
disinterested voters are easily manipulated by media masters, all 
that has to be done is to understand voter psychology and press the 
right buttons with enough skill. It is much more difficult to pull 
the wool over the eyes of people who put in much more time and have 
more interest.

The parliamentarians who edit Robert's Rules of Order are quite aware 
of the inconvenience of repeated balloting. Yet they make a majority 
result an absolute requirement, and recommend that it *never* be 
waived. Not even for mail elections (FairVote has issued some 
propaganda about this that could have initially been excused as a 
simple error. Yet, when the error was pointed out, Rob Richie and 
others took one of two approaches: in some cases they modified their 
claims so that the claims weren't exactly false, but remained just as 
misleading; on other cases they simply asserted I was being 
deceptive, implying that what I showed as the plain meaning of the 
words in the manual was preposterous, because "so many organizations 
don't do that." But many organizations abandon democratic process 
because the leaders decide that democracy is too much work, too 
expensive, and, besides, we know better than the general membership.

The situation where the active voters decide that they deserve to 
make decisions better than the general membership, which can be true 
at a point, and is reflected by the lack of participation of less 
interested members, turns into something else when the refraining 
from participation becomes a fixed barrier, or something maintained 
by differential obstacles. For example, I've seen ostensibly 
democratic organizations, any member can vote at the annual meeting, 
but ... the annual meeting is held in Podunk, Idaho, where the 
founder and the members of the board live. And, of course, proxy 
voting is not allowed. That's differential access, enforcing and 
maintaining oligarchical control. And it's self-reinforcing, because 
the oligarchy can see itself as being those who really care. Sure. But.

Where access to voting is equitable, differential turnout functions 
as a kind of range voting, by excluding votes based on low preference 
strength, while including votes where the voter cares about the 
outcome. My sense is that runoffs improve the outcome of about one 
out of three runoff elections, where top two runoff, with all of its 
flaws, is used. (This is with nopartisan elections, I have no 
information matching this about partisan elections). In the other two 
elections, accepting the plurality winner is the same result as you 
get from the runoff.

Now, a good voting system is likely to reduce the need for runoffs. 
But there is no voting system that can match the power of 
deliberative election, in terms of the intelligence of the result. 
The trade-off is with efficiency, for deliberative methods have been 
considered impossible as the scale gets large. And indeed they are, 
if more advanced deliberative methods are not introduces.

Bottom line: we need to understand that repeated balloting, vote for 
one, with a majority required is such a powerful method that 
improvements can only be in the direction of reducing the number of 
ballots necessary to gain a majority. Then, we might come to a point 
where the improvement from holding an indefinite series of runoffs is 
negligible, assuming we use good methods for primary and runoff, say. 
Some Range advocates believe that Range is quite good enough for a 
deterministic single-ballot system. I disagree, and for reasons that 
I believe can be well-explained. But I don't have quantitative data: 
the basic difficulty with single ballot, irreducible and unavoidable, 
is that voters have less focus in single-ballot and are not as likely 
to understand the issues and the necessary compromises if they don't 
have the information from the first poll. And this is precisely what 
Robert's Rules of Order points out with respect to Preferential 
Voting in general (not just IRV). With IRV, it also points out center 
squeeze, which, of course, afflicts ordinary top-two runoff as well.

However, using an advanced method with top two runoff, both in 
primary and runoff, would, my opinion, so well simulate the process 
of repeated election, majority required that, (1) fewer runoffs would 
be needed, maybe many fewer, and (2) the improvement in quality of 
result from extending the series would be negligible, and would be 
outweighed by the damage from delay in resolution.

But we still need a method of assessing the quality of an outcome. 
And my contention is that there is really only one way to do that: 
test it with sets of postulated internal utilities, on a presumed 
absolute scale, and, with two caveats, the best result is the one 
which maximizes overall utility. "Absolute utilities" means utilities 
on a scale that is commensurable between voters and is summable 
across them linearly.

So if I encounter some election scenario and it is alleged that 
system A performs better than system B, given such and such a set of 
preferences, I always want to know what pattern of absolute utilities 
underlies the voting pattern.

When we are talking about motivated voters, we may be able to use 
normalized utilities. This is equivalent, in a sense, to one-person, 
one-vote, it assumes that we will attempt to serve voters equally. 
Maximum satisfaction for one voter is equivalent to maximum 
satisfaction for another, and the same for the other side of the 
scale, maximum dissatisfaction. However, that assumption clearly 
isn't accurate. So for a deeper understanding that includes 
differential turnout, we need to know what motivates voters to vote, 
and the normal assumption I follow is that absolute preference 
strength motivates it. If voters with low absolute preference 
strength were to "sincerely vote," it could be argued, they might 
vote like this in Range 100:
A 51, B, 50, C, 49. Weak votes. If they did so, the overall utility 
of the result is enhanced over the same voter normalizing and voting 
A 100, B 50, C 0, even though both scenarios are quite accurately 
expressed by A>B>C and each of those ranked preferences is of equal 
strength. (If everyone was like this with respect to their 
preferences, Borda would work fine! -- but we still don't understand 
the whole picture until we factor in turnout.)

There are voting system proposals that test or measure absolute 
preference strength, most notably a Clarke tax. But inconvenience of 
voting is, in fact, a kind of test of sincere preference strength. 
And this isn't of little consequence. If not for it, my sense, 
top-two runoff wouldn't have so many comeback elections!

>so Abd ul, Plurality can and has violated all 4 of those principles
>(in a multi-candidate context) and we've known that for a long time.
>for the average politically-savvy voter in a multi-party context
>(which i consider myself one of), those were the main reasons we
>supported IRV in the first place (over Plurality).

Except that, of course, Plurality and a two-party system are married. 
Rather, in a two-party system, the accommodation for divergent ideas 
is within the existing parties. The party system functions as part of 
the voting system, and by creating separate parties that don't 
cooperate with one of the majors (as a faction within it), one is 
actually bucking the political system and can cause it to break down 
and produce an extreme result.

A great deal could be written about how a healthy two-party system 
works; the parties appropriate the center and overlap across it. If 
they were strictly divided into left of center and right of center, 
the system would be dangerous and, indeed, we have seen what happens 
when a strong faction within a party, more extreme than the center of 
the part rather than more moderate, is able to dominate for a time. 
It produces stronger polarization and therefore a greater swing 
between election results, about which my general comment is that it's 
highly inefficient and can cause crashes, like any oversteering.

Notice this characteristic of Robert's approach: multiparty context. 
But where has IRV been most often implemented? Where is top-two 
runoff most often used? With nonpartisan elections. And, please get 
this: with nonpartisan elections, IRV is astonishingly faithful to 
plurality results. If we are going to understand voting systems, we 
need to understand this fact. And if we are going to design and 
recommend systems for specific applications, we'd better understand 
that the optimal system may vary with the context.

If people just vote for their favorite, in a nonpartisan election, we 
can predict that (unless it's very close), IRV will produce the same 
result and Plurality. Now, if you think that IRV is a decent method, 
and if you research and discover that this assertion is true about 
nonpartisan elections, surely the conclusion is inescapable: 
plurality is a decent method! Where does it break down? With partisan 
elections!

When the systems in use in the U.S. were first designed, the 
political party system had not arise. All elections were nonpartisan! 
So plurality isn't as stupid as we might think; but it was unable to 
handle the rise of political parties, just as those parties corrupted 
the electoral college which was a beautiful conception that turned 
into something quite different because the Constitutional Convention 
punted and did not address the election of the electors, but left it 
to the state legislatures, which easily became tools of the political 
party that happened to dominate there.

Well, done is done. What can we do? I have some suggestions:

1. Preserve top two runoff where it exists and improve it, don't dump 
it. Reduce the need for runoffs by using better methods of finding a 
majority; possibly use algorithms shown to confidently predict runoff 
results from voting patterns in the primary which could again reduce 
runoffs. (Is a majority requirement too tight? The simple-minded 40% 
threshold neglects the possibility of, say, 40%, 39%, 21%. That's not 
predictable, generally. (It might be predictable from ballot data in 
Range/Bucklin, because better preference strength information could 
become available. But the proof is in the pudding. Start collecting 
the data, start with a majority requirement, and accumulate 
experience and see if it's possible to lower the threshold. what's 
really important is lead. 40%, 21%, 20%, 19% is probably quite 
predictable from plurality ballots, more so with better data.)

2. If you can't do anything else, at least stop tossing overvotes. 
Count All the Votes. This, all by itself, turns almost any voting 
system into a better one, and, at the very least, it causes fewer 
spoiled ballots. If you vote for more than one, well, you have voted 
for more than one. If it was an error, it should be counted just as 
your vote is counted if you vote for the wrong candidate! But this 
*allows* additional approvals, which turns Plurality into Approval 
Voting, which is one of the top contenders among experts for best 
voting system! Indeed, with repeated balloting, that might be true. 
(Range/Bucklin amounts to the same thing with a more thorough 
disclosure on the ballot, but the complexity would probably be 
overkill, the really cool thing about repeated balloting with 
approval is that you can accumulate person understanding of the 
necessary compromises though the series of ballots. That happens with 
repeated plurality voting, but it can simply become more efficient, 
requiring fewer ballots in order to find a majority result.)

3. Don't even think about using IRV for nonpartisan elections. It's 
an expensive waste of time and money. It's quite likely that most of 
the votes cast aren't ever counted (except through systems that 
report ballot images). Think about it: normally there are two major 
candidates that lead the others, and those candidates aren't 
eliminated until one of them finally is in the last round. Together, 
the top two are normally a majority of voters. Now, have those voters 
added extra ranked votes? If they did, that justifies the comment 
and, realize, any lower preference votes for eliminated candidates, 
candidates eliminated where a higher preference candidate was still 
viable, are also not counted. This is why some people think that IRV 
violates equal voting principles, since two people may cast a second 
preference vote, but only one of these votes counts, the other didn't 
because the candidate was eliminated first. Not all votes are equal in IRV.

>   and i had hoped
>that it would be very rare indeed that IRV would be consistent with
>those goals (as it would if it agreed with Condorcet, as best as i
>can tell).  it succeeded in 2006 in Burlington and *failed* in 2009
>(regarding Principles 1, 2, & 3, it succeeds with Principle 4).
>that's 1 for 2.  not great odds.
>
>but that is why it might sound like i'm with the IRV proponents,
>because those principles are important to me and *sorta* consistent
>with what IRVers like FairVote.org want.

Basically, you've understood half the problem....

>anyway, it is because IRV so clearly *failed* to accomplish the very
>goals that we had for it when we adopted it is what has motivated me
>to learn a little bit about the whole election theory thing.

What a concept! Here we are trying to get jurisdictions to spend 
millions of dollars, but without actually studying all the reasonable 
alternatives. Nor with any study of the history and what's known 
about voting systems. Without inviting experts to testify and then 
making sure that what they have stated is critically examined and understood.

>>But my concern is the deceptive arguments that have been advanced
>>by FairVote, including their arguments against other voting
>>systems, and it's very important to expose these.
>
>me too.  but i still value those 4 principles and Plurality does
>worse (than IRV which, evidently, does worse than Condorcet).

Plurality obviously fails in certain contexts. But the method isn't 
necessarily to blame for that. Plurality works quite well with 
nonpartisan elections, and it's very simple to vote. It can be 
improved, but, quite simply, it's not as bad as it has been made out 
to be. It fails badly in certain partisan conditions. It also fails 
in nonpartisan elections with very many candidates, but, then again, 
IRV fails spectacularly there as well. No voting system performs 
really well presented, all in one step, with a hundred candidates.

>[large section of my prior comment I have removed.]
>i really agree with all that above, Abd ul.

Please understand how much work it was to discover all that. Nobody 
was challenging FairVote on the Robert's Rules of Order claim. Nobody 
noticed that the voter information pamphlet in San Francisco was 
deceptive on "the winner will still be required to get a majority of votes."

(That is not a "requirement" on a winner under the IRV method used in 
San Francisco, the proposition actually removed the majority 
requirement from the election code. The faux IRV "majority" is not a 
majority of votes, but a majority of votes for the top two, so it's a 
tautology, not some standard that the winner must meet. What was said 
is true for the Robert's Rules of Order process, not what FairVote promotes.)

And, now, I'll point out again that nobody, to my knowledge, has 
noticed the likely effect of runoff "inconvenience" on election 
quality. Except for me and those who read what I've written on this. 
What I'm trying to do is encourage people to step back and try to 
understand the foundations of voting, as well as to look at how 
voting systems actually function. This requires much more study and 
thought than a knee-jerk identification of problems. One of the 
dangers of reform is that sometimes reformers don't understand what 
they are tearing down and they lose the baby with the bathwater, they 
lose the positivefunction of some process or system that they don't 
understand, only seeing the down side.

Communists missed the function of speculators in regulating and 
buffering markets, and only saw greed. While in theory, a scientific 
study of supply and demand combined with the best predictive sciences 
could outperform the ad-hoc and chaotic function of speculation, but, 
it turns out, it ain't so easy to do that and avoid corruption as 
well as simple incompetence in bureaucracies. A speculator makes a 
mistake, he loses his shirt. It's self-regulatory, within limits. 
(Hence the best general process so far is to regulate and tax the 
profits of speculation in ways that still allow the positive 
function, while protecting the public against the possible extreme 
negative impact, under some conditions. And I'm not saying that it 
can't get even better than that, I'm saying that we should be careful 
when monkeying with long-standing traditions and processes. They 
might exist for reasons we don't yet understand.)


>>Focus on pure winning makes sense in the heat of a gladiatorial
>>contest, but, note, the gladiators served a very unhealthy system,
>>at the expense of themselves, they were pawns, sacrificed for
>>entertainment, fighting each other to the death, which, rather
>>obviously, wasn't good for gladiators. Sooner or later someone else
>>is faster or stronger or one slips.
>>
>>> > about this, Kathy, i don't believe your veracity at all.  since
>>>March of
>>> > 2009 (when Burlington IRV failed to elect the Condorcet winner
>>>and all sorts
>>
>>Kathy may make mistakes, but I'd be astonished to find her lying.
>
>she's pretty partisan (as am i), now i don't even remember what she
>said that i found so hard to believe.

Probably a good idea to completely forget about it, then, because, 
while Kathy makes mistakes, so do you, and perhaps what you found 
hard to believe, then, was actually true. She's not partisan as to 
political party, I think. She's a voting integrity expert and is 
really concerned about that, and is only starting to look at voting 
systems themselves in terms of performance.

>>>In my own imagination, I **do** support the Condorcet method,
>>>although
>>>I don't know how to solve the Condorcet cycles or how often, if ever,
>>>they might occur.
>>
>>There are Condorcet-compliant methods, and the first-order
>>intuition of most of us who start studying voting systems is that a
>>Condorcet winner should always win the election. Turns out, no. Not
>>necessarily.
>
>i haven't yet (despite Terry trying) been persuaded of that.

Well, Terry is leading you largely down the wrong path, and he didn't 
propose a sound method of assessing election quality. Basically, if 
you don't  understand how the Condorcet winner can be a *lousy* 
choice, and clearly so, uncontroversially so, you've missed the whole 
topic of social utility as a method of considering election quality. 
But it can be done with simple examples. Ask.

>   i
>*still* believe that electing the non-CW (assuming there is a CW) is
>fundamentally less democratic (reflective of the will of the
>electorate) than electing the CW.  if the CW exists and your
>candidate is not that person, the CW beat that candidate when the
>electorate is asked to choose between the two.  it's my Principle #1.

Okay, let me suggest an approach. Should a good election method work 
regardless of the number of voters? We don't use voting, generally, 
in very small groups, because we don't see it as necessary.

I won't describe the "pizza election," I've done that so many times, 
maybe someone else will. But if we consider three people trying to 
make a common choice, and they use a condorcet method, and two of 
them favor one choice, that will be the condorcet winner, but the 
situation can be such that, once the preferences and preference 
strengths of all the voters is known, *the majority will revise its 
position and consider that *for the group* the best choice is 
actually a different one.*

Majority rule, based on information, Robert. It's how we really make 
decisions in small groups when the relationships are functional. We 
can't understand large-scale decision-making if we don't understand 
small-scale decision-making!

When the scale becomes large, certain approaches become impractical, 
or may require more sophisticated techology or procedures, but that 
does not make them undesirable.

And it's possible to come up with many examples where the condorcet 
winner is clearly suboptimal. Consider the 
placement-of-the-state-capitol election used as an example on 
Wikipedia. How can we judge the quality of the result, aside from the 
method? I.e., what is the optimal placement of the capitol. Most 
methods only consider numbers of voters in each category and their 
preferences, but it's pretty obvious that the ideal result, other 
things being equal, would be the placement that minimizes average 
distance to the capitol. And this isn't necessarily the Condorcet 
winner. If the voters were to vote sincere votes based on mileage, 
they would actually, with a voting method, pick the optimal placement.

But we can transcend voting systems almost entirely. Consensus is 
powerful, and it's much more possible than most people think. 
Absolute and complete consensus on a large scale is probably 
unreachable, there are practical limits, but that doesn't make 
consensus undesirable! And when the goal is consensus, the 
differences between voting systems become far less important, but 
what becomes of interest is collecting the best information, at a 
point in time, as to voter preferences, and it's obvious that this 
should include preference strengths, because otherwise we have a 
mouse looking like a monster and vice-versa.

>>But the exceptions are probably relatively rare, and, in order to
>>understand it, you need to have a deeper understanding of the
>>science of public choice than is possible with only consideration
>>of pure ranking.
>
>but the problem with considering *more* than pure ranking (Range) is
>that it requires too much information from the voter.  and the
>problem with *less* (Approval or FPTP) is that it obtains too little
>information from the voter.

And this is an argument for runoff voting! Use Bucklin in the 
primary. The voter can vote Bucklin with bullet votes for the 
favorite (safe and easy). Look, you are assuming that the system will 
*require* complicated expression. Rather, the point is to *allow* 
expression. Each voter still has and will normally exercise one full vote.

This is something that should be realized: if voters just vote 
sincere normalized preferences, they can vote Range quite the same as 
a ranked ballot. Basically, vote Range as Borda count. And then if it 
looks wrong, *nudge it*! That's a sincere Range vote. Now, can you 
vote a more powerful vote?

Sure. But that requires strategic consideration. What if you don't do 
that? You won't be harmed, actually, though you may not get totally 
*maximized* results. If you want to optimize your vote fully, you 
have to do some work! TANSTAAFL.

Any voter can vote Range as Approval, and it's a powerful vote. My 
own study showed that the expected utility for a voter in a 
simplified Range 2 election with three candidates and a midrange 
candidate for the voter was the same if the voter voted pure approval 
or if the voter voted the midrange. But the idea that voting Range is 
difficult is based on some idea that you have to get it exactly 
right. You don't.

I would start in voting Range by ranking the candidates. That's 
because ranking is easy. But if I found it difficult to rank two 
candidates, I'd lump them together. If I don't want to do strategy at 
all, note, I may not maximize my own personal outcome, but I will 
maximize that of the overall election result! All I have to do is 
sincerely express what the election of each candidate means to me.

I'd put my favorite at the top, and the worst at the bottom. Now, 
what do I do with the rest? Some of them I won't know, perhaps. Where 
do I put them? I can simply not rank them, different range methods 
handle that differently. I wouldn't try to do this, if there were 
many candidates, in the voting booth! I'd do it at home. I'd spread 
out the names of the candidates on a range scale, favorite top, worst 
bottom, then I'd try to arrange them in a way that made sense in 
terms of their value to me. I find it more helpful to think of a 
positive-negative scale rather than the zero to positive number scale 
normally used with Range, because with the top end, I might think of 
how much I'd be willing to pay for the election of a candidate, to 
try to get a quantity. And on the negative half, how much I'd pay to 
avoid the election of this candidate.

Looking at each election pair, candidates which are more like each 
other would be placed together. Candidates which are more unalike (in 
terms of my like/dislike of them, I'd put futher apart. There is a 
formal process for this which would truly maximize the accuracy. But, 
really, it's only an election and I'm only casting one vote! How much 
trouble is it worth?

Once you realize that with a majority requirement, a bullet vote for 
the favorite is just fine, and the rest is just a method of being 
more expressive if one wants to be so, because with a majority 
requirement, a bullet vote is a vote against all other candidates.

Now, is that what you really want. If there is another decent 
candidate you'd be happy to see election, do you really want to vote 
against this one?

Bucklin makes it really easy. You rank them. The basic rule for 
adding lower preference votes in Bucklin with runoff would be that if 
you would prefer to see the election of a candidate than a runoff, 
add a vote for the candidate. Where you add that vote, in what rank, 
depends on your preference strengths. Bucklin essentially votes for 
your in a series of approval elections where you lower your approval 
threshold, which is exactly what happens in repeated balloting. You 
make compromises in order to find a majority for some result.

Honestly, Robert, when this sinks in, you'll wonder why you didn't 
notice it all. It is not rocket science. But we aren't used to 
thinking in these ways, we've accepted certain conditions as normal 
and proper without really looking at the foundations.

I'll quite here, it's late and this is already insanely long. But I 
intend to come back with the rest.