[EM] Reply to Lomax

[Abd ul-Rahman Lomax]

At 03:44 PM 3/3/2012, MIKE OSSIPOFF wrote:


>Abd:
>
>I might post this when it's only partially finished, and then 
>continue it Monday.

I didn't see it when this first went up. I did notice a later post, 
and agreed with most of what you wrote there, Mike. So let's see 
where you were wrong here. Isn't that the goal, to show that the 
other person is wrong?

Seriously, I hope not.


>You wrote:
>>
>>At 04:55 PM 3/1/2012, MIKE OSSIPOFF wrote:
>>
>> >If you rank your favorite, F,  in 1st place, s/he gets a majority,
>> >even though s/he doesn't win, because someone else has a higher
>> >majority.
>>
>>That's apparently quite unusual. Even if multple votes in first rank
>>are allowed -- they certainly should be -- most voters will not use them.
>
>
>You don't have sufficient information to make that prediction.

Sure I do. There are some scenarios that can be asserted that can 
lead to a conclusion that if overvoting is allowed in first rank, 
voters will use it strategically. Otherwise, from what we know about 
Approval Voting, and from the history of Bucklin in certian 
elections, I *predict* that most voters won't use them. Mike, do you 
have sufficient information to show that this is unlikely to be true?

>Yes, the IRVists point out that, when Bucklin was used in the early
>20th century, few if any voters would even rank a 2nd choice.

That was misleading on FairVote's part. Bucklin was used initially in 
public elections, with lots of candidates. Overvoting was prohibited 
in first rank, but maybe one-third of voters, as I recalled, did not 
add additional preferences. What we do know from those elections, 
though, is that a fair number of voters who did add additional 
preferences skipped the second rank and only added a vote in the third rank.
These were obviously voters who had a relatively strong preference. 
It makes sense as a "sincere strategy."

>  My answer
>to that is that plumping is a valid good strategy if no one but your favorite
>is acceptable to you, or if you're sure that s/he will win if you don't rank
>anyone else.

"Plumping" here means? I get two possible meanings. It means bullet 
voting, entirely, or it means only voting for one in first rank.

Many voters only care about voting for their favorite, no matter what 
system you give them, unless you *force* them to add additional 
preferences. Indeed, that was the thinking behind Carroll's invention 
of Asset Voting.

>Those were only municipal elections, of course. You can't use them to predict
>voting in national or state elections. In important elections, 
>people would soon
>learn what voting strategy is in their best interest.

Let's start with small scale elections, eh? First of all, there are 
as yet no national elections in the U.S., the largest jurisdiction to 
hold an election is a state. What we think of as presidential 
elections are actually local elections of pledged electors.

There is a lot of crap out there on what strategy is in the voter's 
best interest, and there is a large class of voters who vote for what 
is in the society's best interest (in their opinion, of course, but 
these voters will value *consensus* and will recognize that getting 
their own preference is not necessarily best for the society).

The importance of an election is subjective, and where it has an 
impact is in the mind of the voter. The influence of what is casually 
and incautiously called "strategic voting" is very poorly 
established. Some kinds of "strategy" are quite obvious: with Range, 
vote at least one candidate maximum, and one minimum. The rest is 
mostly speculation.

I did do a limited study comparing approval-style voting with the 
voting of an intermediate vote, when the intermediate vote was 
assumed to be fully sincere, i.e., this was a three candidate 
election, Range 2, and the voter had zero knowledge and sincere 
normalized ratings of 0, 1, and 2 for the candidates. It turned out 
that the expected value of the election was the same whether the 
voter voted approval style (either choice!, i.e., plumping or ) or 
full range. However, *that the intermediate vote was allowed*, 
improved the overall expected utility for the voter *even if the 
voter did not use it.* (Obviously, as long as someone else did, at 
least one voter). My speculation was that it did this through dithering.

But real elections are not zero-knowledge, and so some kind of voting 
strategy is necessary if the voter wants to optimize the power of 
their vote. That means shifting Range votes to von 
Neumann-Morganstern utilities. We instinctively do this, even though 
it's a big word. We don't waste voting power on irrelevant options.

I'll assert that voters want a system that is easy to understand, and 
that reasonably takes into account their vote(s). They would rather 
not break their head trying to figure out some totally-optimal 
strategy. Favorite Betrayal, as to what it means in fully ranked 
systems, is offensive to most voters, but what can be called FB in 
Approval isn't offensive, it is normal compromise.

And you won't do it, with Bucklin-ER, if you have a significant 
preference. Unless there is a very weird situation.

>It seems to me that, in ABucklin, most people's best strategy would 
>often or usually
>be to just vote certain candidates in 1st place, and not rank anyone else.

Well, whether it is "best" or not, that is how many voters will vote, 
because it is what they know to do. They will only add additional 
preferences in first rank if they really have little preference 
between those candidates.

Now, as to the FairVote deception. Most people in the full-public 
Bucklin elections did add additional preferences, I think it may have 
been about two-thirds. FairVote cites the remaining Bucklin 
implementations, the last to go, which were party primaries. Again, 
many voters will only know one of the candidates, and political 
activists, more highly represented in a party primary, will be more 
likely to bullet vote, they already have a strong favorite, and are 
working for that election!

Further, FairVote's core support criterion makes some sense when we 
are talking about a party election. Which candidate has put together 
the biggest election machine?

But this has practically nothing to do with public elections. From 
the history, we can expect to see, on ballots with significant 
numbers of candidates, voters who approve more than one, on the 
ballot. Only with very large candidate lists, with high uncertainty 
as to who is leading, would there ever be a substantial motive to add 
additional approvals in first rank, unless -- very important 
exception -- the voter has some difficulty figuring out which of a 
number of approved candidates is the best. Then voting for more than 
one in first rank *makes complete and total sense.*


>You continued:
>
>Sequential approval voting, i.e., a series of polls where voters
>start out with "insisting on their favorite," and then gradually
>lower their approval cutoff until a majority is found, is simply a
>more efficient version of what is standard deliberative process,
>i.e., vote-for-one majority-required, repeated until a majority is found
>
>[endquote]
>
>I have nothing against that, but it would be expensive for large 
>public elections.
>ABucklin, would be a perfectly good substitute. AOCBucklin would be better.

You are making assumptions about "public elections." Voting systems 
are general. Full sequential approval voting is used for elections at 
meetings, and could be practical with on-line elections, i.e., 
wherever it is practical and inexpensive to rapidly hold multiple ballots.

Public elections, two ballots is probably the tolerable limit. 
Obviously, finding a majority has been important enough to many that 
they cost has been considered worthwhile.

What is possible with Bucklin/Runoff is that each poll can simulate a 
series of Approval elections with declining approval cutoff. Let me 
hasten to note that the name "Bucklin/Runoff" is quite imprecise 
here. It should not be assumed that the primary result is simply the 
top two -- if no majority is found. It could be *way* more 
sophisticated than that.

And the ballot could be a full-on Range ballot. And I would add 
additional information in any case: the ballots should, by some means 
or other, specify an approval cutoff. Approval has an easy meaning in 
game theory: it means "I prefer the election of this candidate to the 
expense and possible risk of a runoff election."

>You continued:
>
>In any case, to me, if the number of ballots were not to be limited,
>I'd want to see Range polling, with explicit approval cutoff, plus a
>ratification vote that explicitly approves the result.
>
>[endquote]
>
>More expensive still. A good proposal would propose only one balloting.

Again, I'm talking about the ideal, expense aside. This is a 
practical suggestion, though.

But here I'm going to take very strong exception. Limiting an 
election method to a single ballot, making it deterministic on the 
ballots in a single poll, is probably the most damaging single aspect 
of present public elections. You would never do this in person. You 
would insist, and organizations do insist -- it's Robert's Rules -- 
on a majority approval of a result, however that is obtained. 
Robert's Rules disapproves of even mail elections by a plurality, 
they would have the organization repeat the election, and that is a 
repeat with new nominations, the whole nine yards. *And that is even 
with the use of a preferential ballot.*

However, it is possible to design a ballot and system that will, with 
a modest number of candidates, complete *almost always* in a single 
poll. Bucklin did this with some significant candidate numbers! It 
only failed later, in party primaries. And the simple reform there 
should have been to go to top two runoff, with a Bucklin primary. 
Instead they just went to top two runoff. Live and learn.

>You continued:
>
>  In some
>organizations, a mere majority margin, thin, really isn't desirable,
>it should be better than that. Popes were elected by repeated
>approval polling, two-thirds majority required. But I'd prefer to
>leave it to the majority to decide what margin is needed. Otherwise
>it is the *rules* which are in charge. I.e., the past is ruling the
>present, which I'm learning is not a great idea, for many reasons.
>Informing and suggesting, yes, but ruling, no.
>
>[endquote]
>
>???
>
>Ok, I and those who agree with me have a 51% majority. So you ask us 
>how large a
>majority should be required, and we reply (guess what) "51%".

No, the standard understanding is "majority," i.e., more than half. 
Under Robert's Rules, that is more than half of all non-blank 
ballots. That's not a mere detail. A non-blank ballot has been 
"voted," and RR wants to see a majority of votes for a winner.

Look, I'm here talking about social structure. I know of 
organizations that require full consensus for any decision. Bad idea, 
long term. (But it can work spectacularly short-term, that's why 
people do it.) Others have been very successful with *advisory 
structures* that require "substantial consensus," typically 
considered a two-thirds majority as a minimum, and they work hard to 
compromise and make that higher.

My own conclusion has been that the majority properly has -- always 
-- the right of decision, and the majority can decide whether to 
continue negotiating or accept a result as "good enough." Without 
that, requiring a higher margin for the right of decision, leads to 
minority rule whenever the status quo favors a sufficient minority. I 
think I said it, did you notice, Mike. "I'd prefer to leave it to the 
majority to decide what margin is needed."


>I'm not sure how or why you'd implement the flexible 
>majority-magnitude requirement
>that you suggest.

I didn't suggest it. But there are organizations and situations where 
it might be useful. This was dicta, Mike, not the central argument.


> >A number of people rank F, and, if you help F get a majority, then
> >they won't give a vote to their next choice.
> >
> >That's regrettable, because their next choice could win with those
> >votes, while F can't win. And when their next choice doesn't win,
> >someone worse than s/he (as judged by you) wins.
> >
> >You got a worse result because you didn't favorite-bury.
>
>You reply:
>
>Mike, I'm not sure I'm following you here, but the situation,
>multiple majorities in the first round, would be indicative of a
>highly unusual context.
>
>[endquote]
>
>Maybe. FBC-failure won't be common in Beatpath either. But it's 
>possible. It's likewise
>possible in Stepwise-to-Majority. In any method where there could be 
>a situation where
>your best outcome can only be gotten by favorite-burial, you can't 
>assure people that they
>have no need for favorite-burial. I believe that voters have shown 
>us, in elections and in
>straw-polls, that it's absolutely necessary to assure voters that 
>it's entirely impossible
>for there be to be a situation where they can get their best outcome 
>only by favorite-burial.

There is no wide public experience with this. Sure, we should always 
design systems to avoid undesirable outcomes, if possible. But you 
can also, just about always, design a scenario that makes a method 
look bad, i.e., there is, we will assert, a *possible* outcome. Even 
if the probablity of that outcome is such that it is, effectively, 
impossible, won't happen in a million years. Probably!

I'm sorry, most people will definitely not be thinking of a "need for 
Favorite Betrayal," unless they have seen an actual election where 
they got burned and are presented with a possible repeat. But 
multiple approvals is not a sensible meaning for Favorite Betrayal, 
and yes, Mike, I'm aware that you invented the criterion.

And most people will be highly suspicious of anyone who suggests that 
they not vote sincerely.

Approval improves election process over Plurality, and Bucklin takes 
approval and allows multiple approvals without the need to equally 
rank. People loved the method, Mike, that's obvious from the history. 
They did not think like FairVote activists would have them think.



>You said:
>
>Let's see if I understand.
>
>[endquote]
>
>I meant what I said, nothing more or less.

Aw, Mike, that's trolling for a putdown. Stop it! If you do what you 
just said you do, you are phenomenally unique. Nobody *ever* does 
that. You can mean what you say, but what does that mean? Well, what 
I said. But you already said it. The statement adds no information, 
but "meaning" is complex for humans, there are always aspects to it 
that we haven't expressed, and it would take, shall we say, a very 
long time to complete. And might still be impossible.

But you might mean this in some special way that makes it true. But 
why did you say it? I was commenting that I was taking a look at what 
what you had said *meant to me*, and you then can say, "Yes, you 
understand," or "No, you don't," and we can then move on.


>But the first thing you should understand is that I've already said that my
>FBC failiure scenario doesn't work for ABucklin. ABucklin apparently 
>passes FBC.
>I now am convinced that it does, and that there's good reason to 
>believe that a
>failure example cannot be found for ABucklin.
>
>The failure we're talking about, therefore, is only that of 
>Stepwise-to-Majority.

I want something to be understood. If we have Bucklin/Runoff, or 
Range/runoff, or IdealPrimaryMethod/runoff, there will be people who 
will suppress voting for a candidate until the last round. It's 
simplest to start with Approval/Runoff. The existence of a majority 
requirement, or something like that, in the primary, and the 
possibility of a runoff will encourage voters to vote more 
"selfishly" in the primary. If they have to compromise, they will do 
it in the runoff. But runoffs have a cost. The voter must go to the 
trouble of voting again. And about one-third of runoffs reverse the 
position shown in the primary (with TTR). So there is a cost and 
risks of postponing adding additional approvals. That is helpful! It 
causes the voting to have some pressure toward compromise.
Now, say that the election is majority required in the primary, to 
complete, but some kind of plurality approval result is possible in 
the runoff. Suppose the primary and the runoff use, otherwise, the 
same method. A Bucklin method, if stepwise toward majority, in this 
system, will have split the ranks into two elections, effectively. In 
the first election, one would be unlikely to approve a candidate if 
the disapproval of that candidate was serious enough that one would 
prefer to see the election go into runoff.

By definition, a majority-requited in the primary with a runoff is 
"stepwise to majority," though it might not find a majority in the 
runoff. (And methods that pretend to find a majority by severely 
restricting candidacies in the runoff are doing just that. 
Pretending. The real democratic election process is step-wise to 
majority through an unlimited number of steps, with no restrictions 
on candidacy (other than raw eligibility) at each step. It is an 
iterative, intelligent process, and the voters communicate with each 
other between each poll. Candidates withdraw, new candidates are added.

And public elections should emulate this to the degree possible, 
within practical limitations.

The only method that *almost completely* emulates it is Asset, with 
decent post-poll structure (which the holders of the Assets can 
create, the structure for actually determining that a majority has 
been found could be very simple.)

>
>You continued:
>
>If you vote for your Favorite in first
>place, someone else has a higher majority, call him or her A. In the
>first round? There is a third candidate who has a lesser majority, B,
>whom you prefer to A.
>
>[endquote]
>
>B might not have hir majority yet. S/he might get it when the other voters I
>spoke of give votes to their next choice, in the next round.

I postulated it. Or, altenrnatively, I expect it, from the next 
round, as you say.


>You continued:
>
>If you vote for B in first rank, they might tie
>the other majority candidate.
>
>[endquote]
>
>Or, then again, they might not. In general, "might" isn't good 
>enough for timid
>overcompromisers.

Of course they might not. And a tornado might blow away all the 
ballots, or not.


>You continued:
>
>I tend to think of my own votesg as
>being representative of a class of voters, i.e., what I do, others
>may do
>
>[endquote]
>
>Of course.
>
>You continued:
>
>, so this might flip the result to B, an improvement from my perspective.
>
>[endquote]
>
>"Might" won' do.

Actually, you are quite correct, but I don't think you realize the 
implications. I'm describing the possible strategic motivation that 
makes Bucklin fail FBC. You may be able to improve the outcome by 
upranking B, to equally rank B with F. You are calling that a 
betrayal, it's more what I'd call a "stand-aside," i.e., allowing a 
compromise result. It's voting Approval-Style.

But there is a risk to the strategy. Sitting with election results, 
you can say, well, if I'd have done this, I'd have had a better 
result. But, in fact, I did not get a bad result, and the public 
obviously more preferred B than F. I might have guessed wrong, but 
that's always the case, i.e., outcomes are not certain, and we can 
always look back and find ways we could have acted to improve things, 
and we can also find things we did, in which we made investments, 
that were a waste.
Most people, I'll content, don't make it all this complicated. 
Bucklin ER allows them to equal rank. That is, in my view, and 
improvement over *not allowing them to equal rank.* Most people, I'll 
contend, won't use it even if there are rare circumstances where it 
could improve the outcome.

What I saw in my study of Approval vs. Range style voting was that, 
sure, one could look back and think that the result might have been 
improved by, say, a bullet vote. But the harm to the result from 
voting intermediate rating was never full-harm, it was only 
half-harm. Bullet voting can result in full regret. The vote was 
wasted, when it might have prevented a maximally bad outcome.

The situation you are postulating is one where there is a candidate 
you really don't like, who is reasonably likely to get a majority. 
That's a very difficult situation! If this is truly important to you, 
you have a great strategy: approve every candidate but Bad Guy in 
first rank. The most effective thing you can do. However, you 
probably have some idea of who the real candidates are.

Just remember, the "problem scenario" is one where you think Bad Guy 
may get a majority, in the first round of a Bucklin runoff system. 
Bad Guy would have to be really popular to get that result. You are 
trying to swim upstream. It's probably too late.

In fact, when serious Bad Guys get into runoffs, voters turn out in 
droves, the runoff turnout can easily exceed the voting in the primary.


>You continued:
>
>But if I really fear this, I can vote for B in first place in
>addition to F. That's not burying, that is normal Approval/Range
>strategy. It's equal ranking, not preference reversal.
>
>[endquote]
>
>...and there's no assurance that it will keep A from winning. Maybe A
>_needs_ those votes that those other voters will give to hir if you
>don't give F a majority.

Eh? There is never any assurance of anything. What are you talking 
about, Mike. I'm not getting it. How the other voters vote is not 
dependent on how I vote. You must be thinking of votes already cast. 
But this doesn't make sense. If I give F a majority with my vote, the 
election is either over, or there is a multiple majority (which may 
have implications for a runoff, it depends on details.)

Suppose this leads to a runoff between F and A, the Bad Guy. If you 
feat that A will win this election, you are fearing an outcome that 
is desirable socially. You are in trouble. If you are serious, work 
hard in the campaign, because that is where you can be much more 
effective than with your single vote, which really doesn't count for 
much, usually.


>You continued:
>
>The optimal number of ranks in a Bucklin ballot would be such that
>nearly all voters bullet vote in the first rank.


>Nonsense. If there are unacceptable candidates who could win, your best
>strategy is to rank all the acceptables in 1st place, and to not rank anyone
>else.

Mike, I'm disappointed, I'd expect better from you. This is a common 
assertion about Approval strategy. It's way overgeneralized. While 
you are technically correct, with proper specification of conditions, 
it's misleading.

The conditions for what you have said to be correct: the 
"unacceptable candidates" are truly Bad. Like genocidally bad. Or, at 
least, that they are so bad that you don't care about who among the 
other candidates wins, you will be *so* relieved that one of the Bad 
Guys did not win.
But what you are fearing here is that a majority of the electorate is 
ready to approve one or more of the Bad Guys. You don't trust the 
electorate, a difficult position to be fixed in. Perhaps you should 
buy a ticket out of the country for the day when the results of the 
election will be known, before the black helicopters can be sent. 
Cast your vote and tie your camel. Seriously, remember, a contested 
Bucklin election with more than two viable candidates will almost 
never complete in the first round, for obvious reasons. It is *highly 
unlikely*, in most scenarios, that you would *reasonably fear* that 
Bad Guy would win in the first round, unless your position is really 
no-hope. Obama is your Bad Guy.

>Further, even if it isn't a u/a election, in my posting about 
>voting-options in Approval elections, I said that the
>seeming (often) suboptimality of the ABucklin option in an Approval election
>suggests that voting only at 1st rank is often the best ABucklin strategy,
>even if it isn't an unacceptables/acceptables election. You might 
>not agree with
>that conclusion. I admit that it's only a subjective impression.

I don't really know the name ABucklin, so I'm assuming that what you 
wrote applies to simple Bucklin-ER. The traditional three-rank 
Bucklin method, but with overvotes allowed in first rank. Perhaps 
ABucklin means Approval Bucklin. Traditional Bucklin was Approval in 
all ranks but the first. Traditions die hard. They really wanted to 
get those Favorite votes. Makes the method uncontestably Majority 
Criterion compliant.

But at the cost of collecting garbage information, indicating strong 
preference when the preference might be weak or a coin toss.

If allowin that overvote allows voters, in some circumstances, to 
improve outcomes by equal ranking a candidate with their favorite, I 
see absolutely *no harm* in that. They are either correct, or they 
are stupid. In any case, they act as they decide, and making a 
compromise is not a bad thing, in itself. And that is all they are 
doing. They are accepting, in the scenario above, b, because they 
prefer B to A.

But I think that very few voters will do this unless there is a 
really unusual situation, an election with three viable candidates 
*and* a reasonable possibility of the worst one winning in the first 
round. 3 viable candidates, and one, A, is in range of getting over 
50% first-preference votes? That must mean that the other two are 
*together* at 50% or less. If one of them is at 30%, the other is at 
20%. We don't know about second-preference, but presumably polls 
would be available.

What kind of election is this? If it is nonpartisan, it is *highly 
unlikely*, with those first preference numbers, that anyone else 
could win this election but A.

Now, what could demolish this analysis is a lot of voters who prefer 
A second, but who add votes for A in the first round. But why would 
they do that? These voters, if they vote that way, must think that 
their favorite can't win, and they must strongly dislike the runner-up.

No, almost all voters, in Bucklin, even if allowed to overvote, will 
not. Exceptions would be strong supporters of no-hope candidates who 
don't want to waste *any* voting power. By definition, this will be 
very few -- or the candidate would not be "no-hope."

Runoffs will depress additional approvals in the primary. I know that 
Mike will point to turkey-raising strategy, I'll look at that later. 
As it applies to Bucklin, turkey-raising strategy would probably be 
sincere, in a way. Whether or not this is viable at all may depend on 
rule details, and most importantly, the common assumption that if it 
is runoff, it must be some kind of top-two runoff, is just that, an 
assumption. We will look, eventually, at the evolution of the Voting 
System of the Future, and my own personal question is when we will 
get wise to Asset, which makes complex voting systems *obsolete,* 
Range polls plus efficient deliberative process blows them all away.


>You continlued:
>
>When there are more than two viable candidates, I'd expect majority
>failure to occur in the first rank, routinely. The scenario presented
>won't occur, at all
>
>[endquote]
>
>Wrong. Of course it could. If you want to actually _demostrate_ that it
>won't often happen, then I invite you to do so.

Demonstrate that it will happen, if you want to raise this bete noir. 
I'll warn you, I'll want to see some realistic utility profiles. Many 
"nightmare scenarios" involve preposterous profiles, or something 
that *sounds bad* but if you look at Bayesian regret, it is either an 
improvement, not a harm, or it's relatively harmless.


>But of course it could happen.
>
>You continued:
>
>, so a voter worrying about it is worrying about
>something quite unlikely.
>
>[endquote]
>
>As I've said, I've watched someone favorite-bury in a Condorcet-counted
>rank-balloting election. If you can't firmly assure voters that they can't
>possibly benefit from favorite-burial, then timid overcompromisers are going
>to favorite-bury.

Favorite burial is a rational strategy with many voting systems. 
Equal ranking amost always makes it unnecessary to acdtually "bury."

Mike, you have this idea of elections ruined, right and left, by 
"timid overcompromisers." People compromise because they don't see it 
as a high cost. "Timid" is your projection. If they care a great 
deal, strong preference, they won't compromise so easily. That's all. 
What, exactly, is the problem with  "timid overcompromisers"? What is 
"overcompromising"? Each voter sees benefits and risks. If they see 
the risk of failing to compromise as greater than the benefit of 
refusing to compromise, they will compromise. It indicates preference 
strength. It works.
I would guess that you would define "overcompromising" as 
compromising when you might be able to get a personally improved 
benefit by refusing to compromise. But missing from this analysis is 
the *amount* of the benefit. If it is small, there is little damage 
to social utility, if any. It might be an improvement.


>(The below-quoted question has been answered: ABucklin can't fail FBC)
>
> >Does anyone know if there's actually a proof that ER-Bucklin meets FBC?
>
>You answered:
>
>It's an Approval method, so this depends on how you define "Favorite
>Betrayal." If equal ranking is betrayal, yes. But that's weird.
>
>[endquote]
>
>Favorite-betrayal is favorite-burial. Voting someone over your favorite.

Is equal ranking the favorite with someone less preferred, "favorite-betrayal"?

However, Mike, you should know that I place little weight in most 
voting system criteria, as absolutes. I define an ideal voting system 
as one that efficiently finds maximized absolute utility (which can 
probably only be known in simulations). An ideal system *must 
violate* certain criteria as applied to single-ballot systems. Few of 
the criteria have been carefully applied to repeated ballot systems, 
except with unrealistic -- or just plain false -- assumptions.

Failing FBC is not necessarily a problem. Obviously, it's, in itself, 
undesirable. Whether it's a problem or not depends on the effect on 
SU maximization. *How much of a problem*?


>And no, not all Approval-related methods pass FBC. For instance,
>Stepwise-to-Majority and Stepwise-When-Needed fail FBC.
>
>Those are stepwise Approval methods.

Mike, you know a great deal more than I about the details of many 
different systems. You were discussing and working on these issues 
when I was only working on delegable proxy, and knew little about the 
great variety of polling systems.

> >Can it be shown that the verbal FBC-Failure scenario described above
> >couldn't really happen?
> >
> >Might ABucklin fail FBC?
>
>You continue:
>
>I don't see how what you described is Favorite Betrayal
>
>Voting someone over your favorite is what I mean by favorite betrayal
>or favorite-burial.

Fine. So equal ranking does not cause failure. Approval passes. 
Bucklin passes. We may then look at Bucklin/runoff.

The benefits of runoff are not only from raw SU optimization, but the 
*process* brings focused collective intelligence to bear on the 
election problem. That second campaign matters, and the differential 
turnout matters. Little of this has been carefully considered, though 
Robert's Rules is quite aware of the benefits of repeated ballot, 
i.e., voters are now informed by the results of the earlier round, 
and will make more informed choices.


>(I don't capitalize it, except as part of the name of the Favorite
>Betrayal Criterion)
>
>You continued:
>
>, but I
>probably don't realize details of the method you are considering.
>
>[endquote]
>
>1. I'm no longer considering Stepwise-to-Majority
>2. I posted all of its details.
>
>I'd be glad to answer any questions about what I meant when I defined it,
>but remember that I don't propose it, due to its FBC failure.

I don't know what Stepwise-to-Majority is, specifically, but Bucklin 
certainly looks like that name. Range/Bucklin, i.e., a Range ballot 
analyzed stepwise per Bucklin, could use that name.
Once we are getting sophisticated with the canvassing method, 
however, there are many tests that can be applied to a ballot. Once 
there is a runoff possibility, there can be Condorcet Criterion 
satisfaction, for example, due to the runoff behavior. There are a 
host of details that can make the difference. Are write-ins allowed? 
If so, there aren't actually any eliminations, there might just be 
ballot position. It's possible that a runoff has more than two 
candidates on the ballot, if an advanced system is used in the runoff. Etc.

Getting a majority is desirable. It's essential in direct process, 
but we are talking about compromised public process. Some 
jurisdictions do allow a write-in in the runoff, and that is clearly 
more "democratic." (Robert's Rules standard secret-ballot elections 
use blank pieces of paper. Voters can vote for anyone eligible for 
the office.) Write-ins can and have won. But my real point here is 
that by using an advanced method that is relatively spoiler-proof, 
bullet voting may cause failure to reach a majority, but there is no 
need to bullet vote, say, with a Bucklin runoff. You can vote for 
your favorite (who might be write-in) in first rank. You are then 
free to vote for a frontrunner. If you have strong preference, you 
might vote for a frontrunner in third place, leaving the second rank 
blank. Or you might vote for a frontrunner not at all. It's up to 
you, and there is no single strategy that optimizes the outcome for 
you in all situations, but the method, properly done, will amalgamate 
sincere preferences properly.

Bottom line, the search for the Holy Grail of voting systems is 
probably looking for something that does not exist. But we can get 
quite close. Let's suppose we have an ideal system. Fine. Be sure to 
use it for the final round in a runoff system that will guarantee 
that an "ideal winner" will get into the runoff. The voters will have 
a vastly improved level of information about the overall position of 
the electorate.

Just remember, in any decision-making system, utilities will be 
combined with probability information. The latter is the basis of 
what is called "strategic voting." It is inevitable.



>You continued:
>
>I
>haven't been reading the list, but "Stepwise Bucklin"
>
>[endquote]
>
>All Bucklin is stepwise. Bucklin is a stepwise Approval.

Right. Bucklin can use a Range ballot, stepping down through the 
ratings, seeking majority approval. If the ballot has an explicit 
approval cutoff, with the meaning being as I've described (i.e., 
majority approval is required for election in a primary, or else the 
natural consequence is a runoff, a nuisance for most voters), then it 
really can find that majority, if it exists.

This is not necessarily ideal, but it's not been simulated, to my 
knowledge. Runoff simulations have assumed that the voters are the 
same in the first round and in the runoff, which is totally 
unrealistic. A more sophisticated simulation will start with a voting 
population and an algorithm to determine who votes, i.e., the 
probability that the voter cares enough to vote. What has been missed 
by many is that turnout is a factor that incorporates *absolute range 
data* into the process.

>  Not the only one, but the
>only kind I know of that passes FBC. I propose the method described 
>in the electowiki
>as "ER-Bucklin". I call it ABucklin.

Yeah, by the end of this reply, I'd figured that out.
Raw Bucklin, I do not see as ideal. Just damn good. Far better than 
UnFairVote would have us think. UnFairVote has been going around 
killing Top Two Runoff, the most widely adopted voting system reform, 
which is much better than IRV, in terms of performance. Sure, it can 
be more costly. But that is fairly easy to fix, and it's amazing to 
me that this was apparently never done.

Bucklin was, like IRV, oversold as a method of finding majorities in 
multi-candidate elections. It often succeeded, and they were genuine 
majorities, unlike the faux majority that IRV promises. But it could 
fail and, later, under party primary conditions, did fail.

Because of this, those states went to Top Two Runoff. Which is 
vulnerable to Center Squeeze, and which can have higher cost.

So use a runoff system with a Bucklin primary, at least. Still not 
ideal, but a tremendous step forward. The two most widely-implemented 
election reforms in the U.S., combined as a hybrid.

But it all starts with Count All the Votes.