[EM] Bucklin option wrap-up
Abd ul-Rahman Lomax
abd at lomaxdesign.com
Fri Mar 9 13:58:11 PST 2012
At 04:04 PM 3/5/2012, MIKE OSSIPOFF wrote:
>(First, when I speak of ABucklin as an Approval option, that's only
>for brevity and simplicity,
>because, starting from Approval, I'd offer AOC as an option or an
>alternative method, before ABucklin.
>AOCBucklin, for me, would be next, after AOC. Still, as Abd
>suggested, ABucklin is the natural
>1-balloting implementation of a reasonable and obvious collective
>deliberating process, and, thereby,
>might be the first alternative voting system that people would
>accept--Then AOCBucklin could be reserved
>for a later enhancement.)
Unfortunately, I don't know what "AOC" is, and I searched the subject
headers here. It was used in many posts, none of which explained what
the method was, some contained long sentences with unexplained
alphabet soup..... Sigh.
Thanks, Mike. Bucklin is a simululation of a *common* collective
deliberating process, not just a reasonable one. Bucklin/runoff is a
more sophisticated simulation that splits up the compromising process
into two ballots. The common process is unlimited-ballot, no
eliminations. Public elections will compromise with this because of
the need for election efficiency. Two ballots is probably about the
(But Asset blows all this out of the water. It reduces the election
to a deliberative process among electors, a reduced set, so it
doesn't suffer from limitations imposed by limited ballots and polls.
In theory, Asset can introduce new candidates and elect them, as
>I've said that, as a vote-management option in an Approval election,
>ABucklin can act in a way that isn't
>really in the option-chooser's interest. I was referring to the fact
>that, if a candidate whom you don't like
>gets a 1st-round majority, then you can't give a vote to anyone new,
>though it might well be in your interest
>to do so. Stepwise-to-Majority was my attempt to improve on that,
>but it fails FBC.
Yes. However, if the system checks for certain conditions on the
ballot, before awarding a first-round victory based on a simple
majority, it's possible that a runoff would be triggered. I'm looking
at an advanced implementation using a full Range ballot, with
sufficient rating detail, so that a meaningful Range winner can be
determined, and also a meaningful Condorcet winner. If they differ
from that Bucklin winner, let's suggest, a runoff is held. A majority
may be necessary to complete with the first poll, but not necessarily
sufficient, where the full ballot indicates that a better candidate
might have been missed, precisely because of failure to compromise,
as you suggest.
To me, the question is whether or not the improvement is worth
holding a special election, the runoff. I'd like to see the results
of lots of simulations. It's also possible to use the Range data in a
sophisticated way, i.e, only pairing off the Range winner (who
probably doesn't have majority approval in that round) against, say,
the Bucklin winner, if the Range margin is above some threshold. Some
very narrow Range advantage has a strong possibility of not being
true utility optimization, due to normalization error (as well as,
perhaps, poor strategic voting).
>What makes the ABucklin option sometimes not in your best interest
>is the same thing that lets it meet FBC, where
>Stepwise-to-Majority fails it.
The real issue, to my mind, is identifying, not the ideal election
method, as such, but identifying a path to reform that is open-ended
and flexible, that improves the situation with each step, and that
allows and makes way for further reform.
>I know I've said this, but the fact that ABucklin, as a
>vote-management option in Approval, doesn't look very appealing
>suggests that, in an ABucklin election, it's probably often or
>usually best to just vote as in Approval, giving top-rank
>to some candidates and not ranking anyone else. Even if it's not a
That is offensive to most voters. Mike, it's not realistic. I think
you believe it is optimal, but my study has led me to think it is not
optimal, except under certain situations that the voter will probably
know with some accuracy. Basically, with these situations, their
favorite is no-hope, and it is otherwise a very close election
between two frontunners, and thus they really want to get in their
digs immediately. Given the realities, from history, of most Bucklin
voting, if there is any doubt about the outcome, it is very unlikely
to be resolved in the first rank.
I'm assuming that a majority is required in the first round! Your
arguments apply better to single-ballot Bucklin. Still, in that
method, under the situation considered, a first-round majority is
very unlikely, therefore the voter is reasonably free to vote totally
sincerely in first rank, and adding all those other approved
candidates can be done in second rank or third rank.
I don't know "u/a" election.
>...but certainly if it is a u/a election. Sure, it could be tempting
>to use ABucklin's rank positions to distinguish between
>the merit of your various acceptable candidates. But, by not ranking
>them all at top, you could let an unacceptable win:
It's possible. But unlikely to bite at the top rank. Very unlikely, my sense.
We need a history of real elections, with much better collection and
analysis of ballot data. I want to see Range ballots used, ASAP, even
if the analysis doesn't take full advantage of that data. I.e., it
could be Range 4, with ratings of 0 and 1 as unapproved, and 2, 3,
and 4 as approved. That could be counted as simple Bucklin (the 1
rating was missing from old Bucklin ballots) but it could also be
analyzed as Range -- and for a Condorcet winner, with up to 5 ranks.
(I somewhat recommend that the 0 rating not be explicit, it's the
rating assigned to all candidates by default, but there is another
reasonable possibility that has not been studied: it could be Range
5, conceived as +/- range, with ratings of -2, -1, 0, 1, and 2,
perhaps with the rating of 0 being the default. And that zero might
be specially handled, I'm not going there today. It is possible that
for Condorcet analysis, a blank would be considered bottom rank, but
for Range analysis, it would be midrange. All kinds of possibilities
come up, once we realize that ballot form and ballot analysis can be
separate. The ballot should be easy to vote, with a natural voting
strategy being, at least, a reasonable one, not stupid.)
>Maybe an unacceptable will have a 1st place majority, and maybe your
>faction could give a certain candidate (acceptable to you) a bigger
>1st choice majority if it top-ranks hir. ...even though you'd
>sincerely rank hir below some of your other acceptables.
Most people, by the way, don't vote in factions, in non-partisan
elections. This has largely been overlooked. What I saw, studying the
performance of IRV in nonpartisan races, was that vote transfers
tended to stay the same. I.e., the supporters of A, when A was
eliminated, looked just like the rest of the voters in the election!
This is why IRV generally behaves like Plurality in nonpartisan
elections. It's not what we would think, from thinking about partisan
Poliatical activists may wish that voters would vote a certain way,
but convincing voters to do it, to vote in a way that the voters
perceive as insincere, when they have a reasonable choice that allows
them to be relatively sincere, is difficult. As it should be! It
could backfire, indeed.
>So, in a u/a election, just top-rank all the acceptables, and don't
>rank any unacceptables. Vote as if it were Approval. And when
>ABucklin is offered as an option on a u/a Approval election, don't
>use it--just vote an Approval ballot.
That's okay by me, but, historically, only about a third of voters,
in the elections I've seen data on, did not add additional
preferences. Something we often miss is that an advanced voting
system (such as top two runoff, for starters) will increase the
number of candidates. Which then puts more strain on the voting
system. However, Bucklin handled huge lists of candidates, apparently
with ease. Being able to multiply approve in lower ranks probably
helped a great deal with these massive ballots and only three approved ranks!
I highly doubt that nany voters will bullet vote as Mike suggests
would be optimal for them, and I don't believe it's truly optimal.
He's not considering the risks on the other side, just looking at a
possible benefit. But it's moot. The only people who will vote
Approval style will be those who have no strong first preference,
which means that those votes are actually quite accurate. Or voters
who support a totally no-hope candidate, and for some reason they
want to use all of their ranks. A very few might think as Mike is
suggesting, and I see no harm from it. It is not Favorite Betrayal.
It is suppression of a preference, which is obviously not terribly
strong *as to any real choice.*
I.e., perhaps I really would like to elect Jesus Christ. And I want
to make a stand for him being our leader. But I also have some shred
of realism here, and am worried that Bad Guy will win in the first
round. So, I vote for Jesus Christ, as a first rank write-in, and for
not-Bad guy in first rank also. I've "suppressed my true preference." Big deal.
>Looking at it from the other direction, the fact that it might
>sometimes be worthwhile to vote several rank positions in ABucklin
>suggests that the ABucklin option could sometimes be genuinely a
>good idea in an Approval election.
Bucklin obviously improves Approval. It answers the number one
objection that people make about Approval, that they cannot indicate
a preference for their favorite and at the same time participate in
the real election, should they happen to support a minor candidate.
>Abd pointed out something that hasn't been sufficiently emphasized:
>ABucklin is just the 1-balloting implementation of a
>familiar, natural and obvious deliberative process (well almost,
>because people aren't accustomed to Approval voting).
Right. However, Count All the Votes simply makes the vote-for-one
recursive process a bit more efficient, that's all. The Robert's
Rules editors have their head stuck up their ... okay, okay, to be
fair, Robert's Rules of Order, Newly Revised, is a manual of actual
practice, it is not normative, as such. It is not going to recommend
something that isn't in broad use, period, and some very old and very
defective arguments were swallowed about approval, long ago.
"Violates one-person, one-vote," though it's easy to show that it
doesn't. Period. Approval votes are "alternative votes" just as are
the votes in IRV. Only one of them counts toward creating a winner,
or none of them do. (in a single-ballot Approval election, it's a
little trickier for a runoff Approval election, but that's an
entirely different story, and it is the last round that counts, the
first round, when it doesn't create a winner, is really just a
Mike, by the way, thanks.
>could be a good introduction to, and argument for, ABucklin. Maybe
>people would thereby actually accept ABucklin before they'd
>accept Approval. For instance, it's known that ABucklin was used in
>the early 20th century in some U.S. cities.
It's possible. It was very, very popular. It was easy to understand.
And it obviously worked, in those implementations. It found
majorities. It busted up center squeeze and the spoiler effect. Ooops!
It was too good. The first candidate elected with Bucklin was a ...
oh, it's too horrible to contemplate, but .... I better mention it, ,,,
Of course, it was a nonpartisan election, the guy was actually pretty
popular, and didn't govern the town as a socialist. But who needs facts?
>Of course, Approval voting is a built-in option in ABucklin, without
>explicitly adding it legislatively. A person can just
>use the 1st rank position if they so choose.
Yup. IRV with Count All the Votes is an improved method. Have I
mentioned that we should Count All the Votes?
And I'm going to keep mentioning that until we are counting all the
votes. Or I'm dead, or close to it. Seriously.
>AOC, AOCBucklin and SODA
>could be options to be proposed later--Just as AOC, AOCBucklin,
>ABucklin and SODA could be added later as options for Approval.
Yup. I've mentioned, though, that I'd look to see a Range ballot be
used, and then all kinds of improved ballot analysis become possible.
The method, with a runoff, could become Condorcet compliant, which is
important. In my view, the initial Condorcet winner may not be the
optimal result, but the majority should have a chance to choose a
better outcome, explicitly. I.e., assuming sincere ballots in the
first round, to actually choose to greatly improve utility for a
minority, in exchange for a small loss for a majority. Real people do
this all the time, in real life.
So a Condocet winner, if different from, say, a Range winner in the
first round, would make it into the runoff, as would the Range winner.
I didn't mention a Bucklin winner. I'm not even sure that true
Bucklin analysis is necessary in the system I've mentioned. I'm
leaving that for later. I think that Bucklin is quite good enough to
start, and we can go into fancy ballot analysis later. One step at a time.
Count All the Votes.
>Probably best to propose Approval first, but maybe people would
>really like ABucklin because it implements, in one balloting,
>a familiar idea in collective deliberation.
Absolutely. Gradually, through repeated ballot, people effectively
lower their approval cutoff, becoming more compromising, in order to
complete the election. They may start out as highly partisan (i.e.,
voting for only one in first rank). Or they might start out more
compromising, voting for more than one in first rank. It's up to
them, and those choices are not really difficult, if you don't try to
run a full simulation in your head!
Just for for your favorite in first rank. If you have trouble
figuring out which of the candidates is your favorite, vote for more
than one, this is an advantage of allowing equal ranking in first
rank. It makes voting for some voters easier. For others, it being
allowed does no harm.
Then figure out what candidates you would prefer to a runoff being
held (again, I'm assuming majority requirement to complete in a
primary). Divide these candidates into two groups, and rank them in
second rank (the best) and third rank (barely acceptable.) If you
have a strong preference for your favorite, you can rank all those
others in third rank.
While, as Mike has pointed out, this may not be totally optimal,
because of what I'm claiming are some rare possibilities, it's quite
good, and it is terminally simple.
And you can just vote for your favorite in first rank. That is
effectively a vote against all the other candidates *in the first
poll.* It will help keep all of them -- except your favorite -- from
getting a majority. The cost: you might have to vote in a runoff, but
you can make a decision later. You might not care, it depends on who
is in the runoff.
>But, if people are to like ABucklin as a first proposal, maybe they
>wouldn't like ER-Bucklin's delay that gives it MMC compliance. That
>delay might seem a complication. I'm referring to how, in
>ER-Bucklin, if you vote for several candidates at the same rank position,
>your ballot doesn't give to the candidates at your next rank
>position any sooner than it would have if you'd ranked those several
>equally-ranked candidates in separate rank positions. That preserves
>MMC compliance for ER-Bucklin.
I'm not sure I follow. I do consider that ideally there will be
enough ranks that one could rank all candidates, but that the cost in
ballot complexity may not be worth this. Reducing the ranks will have
little effect on Bayesian Regret until the number gets pretty low.
>So maybe that delay should be left out of ABucklin as a first
>proposal. It could be added later as a later proposal. Without that delay,
>ABucklin would lose its MMC compliance, giving it about zero
>merit-improvement on Approval. But still, it could be used as Approval,
>via top-ranking. If that's the easiest and best route to Approval,
>then it's worthwhile even if it doesn't meet MMC and offers nothing
>that ordinary Approval doesn't offer. Then, the MMC-preserving delay
>could be added as a subsequent proposal.
I'm not following this, and don't have time to. However, my basic concept is to
1. Count All the Votes.
2. Further improve data collection by the voting system. Range
ballots with approval cutoff collect the maximum data, for a given
number of "ranks."
3. seek majority approval for any result. This does *not* necessarily
mean that the method terminates as soon as a majority is found. There
might be a greater majority for another candidate, in full analysis,
and there might be other conditions meriting examination. Use a
runoff election only when needed, i.e., if the first poll is inconclusive.
4. Use increasingly sophisticated analysis of ballots to optimize
outcomes, which can create multiple election criteria compliances,
made possible through the use of a runoff, which might be limited to
three candidates maximum, possibly including write-ins. Optimize the
choice of the runoff candidates.
But then there are further steps:
5. Replace all of the above with Fractional Approval Asset Voting in
a multiwinner election, using Hare quota STV, jurisdiction-wide, for
an optimally sized Assembly, that fully represents the voters, and
that handles single-winner elections deliberatively and democratically.
6. Allow electors (Asset holders) to vote directly as they choose, on
main motions before the Assembly. (To participate in deliberative
process beyond that, a seat is needed). That voting could be direct
or by internet. Those votes, my strong prediction, would rarely
influence outcomes, because, remember, the electors have created
seats that truly represent them. They won't be disagreeing all the
time! But the fact that they can will increase the sense of participation.
With a system like this, I predict, people will for the first time
have very strong reasons to think of the government as Us. Because
all voters will be able to see the specific effect of their vote,
they will know who they voted for, and they will see what was done
with that vote, if they care.
The election of the Assembly isn't a contested election. It is a
voluntary amalgamation of votes, by intelligent and free actors, to
create cooperative seats. Nobody loses. If, as an elector, you are at
the "dregs," and won't compromise, well, you might not be
represented, but in the full system, you could still vote directly if
you really want to waste your time.... and maybe it won't be a waste,
it will be up to you.
Asset creates an intermediate body, an Electoral College, to serve as
an intermediary between the voters and the Assembly. And people will,
I predict, eventually, only vote for people they personally know and
can directly communicate with. Who will then pass those votes on up a
natural hierarchy. This means a huge candidate set in the Asset
election, probably no candidate names on the ballot at all. Perhaps a
booklet at the polls with candidate names, or voting machines that
allow voters to find the candidate they want. Perhaps candidate codes
that a candidate can pass out. Whatever.
This would be a vast step in the history of democracy, the first time
that the people have truly been *represented* by people chosen
without opposition. Currently, I'm not represented in the State
Assembly, my district is.
>In case I haven't emphasized this enough, I propose ABucklin and all
>of the conditional methods, including AOCBucklin, as options in an
It all starts with Count All the Votes. And then we add ranks to
address the obvious problem with pure Approval.
>Using the MTAOC kind of conditionality implementation, all of them,
>and SODA too, could be allowed as options in the same Approval
>election. But I suggest starting by just proposing ordinary
>Approval, unless your initiative committee, or the polled public,
>insist on something
>fancier than ordinary Approval. If they do, then offer AOC, ABucklin
>&/or AOCBucklin, and SODA as possible options in Approval elections. Options
>are difficult to argue against.
UnFairVote has emitted such a huge quantity of hot air, making it
seems that some simple *option* for voters is going to ruin the
system. They are really arguing that voters are stupid and will be
fooled by what is really quite simple. Yes, you can vote for more
than one, your ballot won't be discarded if you do, and it has the
effect of supporting those candidates you vote for, equally. But you
don't have to do this, you can ignore the privilege, and if you are a
supporter of a frontrunner, it does you no harm. It helps you if you
support a minor candidate, because it means that you can vote for
your favorite and still participate in the main election.
And it happens to be quite a good method, according to political
scientists. There are better methods, quite possibly, but not that
are so completely simple and cost-free. It does no harm.
>Anyway, I propose that entire complete system, probably with those
>options added as later proposals of enhancements for Approval.
A proposal document will keep it simple. I do recommend setting up
some process that, initially, will study voting systems and recommend
the first steps, and that will continue to monitor performance and
shift recommendations as necessary.
This was completely missing from all those IRV implemtations. They
were running experiments, but forgetting to collect and analyze the
data, to see if the system was performing as promised.
>For that matter, MMT or GMAT could be an option in an Approval
>election too, but not both in the same election. And those methods don't
>allow for conditionality for some candidates but not others. I
>prefer AOC, with which you can give conditional votes to some candidates,
>and unconditional votes to others. It's the most fully-optional kind
I'm not really concerned about details at this point. I just want to
point out that once there is a sophisticated ranked ballot being
used, with equal ranking being possible, many, many forms of improved
analysis can become possible. And then we can actually study the
possible effects of changes, and we will have far better data on
which to base future reforms.
>GMATBucklin and MMTBucklin would be the same as AOCBucklin, except
>that they'd implement conditionality the GMAT way, or the MMT way,
>instead of the AOC way. Of course GMATBucklin would be compatible
>with GMAT as options in the same Approval election.
>I agree that MMPO (I prefer its MMPO2 version) and MDDTR might run
>into criticism trouble, even though I like them best, and so I don't
>consider them to be major proposals, though one might mention them
>to an initiative committee or a public that insists on something fancier
>than Approval. But I feel that it would be better, in that
>eventuality, to just offer the options of AOC, ABucklin, AOCBucklin
>and SODA, as I
>described 3 paragraphs ago. All of those could be options in the
>same Approval election, in fact. Let the initiative committee decide what
>it wants to include. SODA might sound a little too different, to the
>public, though I've gotten a favorable reaction to my brief definition of it.
The Asset concept is quite simple. SODA is a bit more complicated,
and I think that complexity adds insufficient value, but, hey, if
others disagree, you will not find me disliking giving voters a
choice, improved options. I like vote-for-one asset, or FAAV
(Fractional Approval Asset Voting) because it creates a clear
interpretation of overvotes, in a multiwinner context -- but this
also works with single-winner asset -- because I see little value in
diluting representation. You would actually be weakening your real
power, my opinion, for real power will have to do with the ability to
participate in the process. Vote-for-one in an Asset multiwinner
election can create virtual districts that push toward local
representation. You vote for your neighbor, the one you most trust....
But you can still talk to other electors....
>MTA, MCA, MTAOC and MCAOC could be offered too, but people who want
>something fancier than Approval usually want an unlimited
>ranking method, and there's a simplicity advantage in not speaking
>of too many options.
No matter how many candidates there are, I don't see more than maybe
Range 10 as adding much value. I.e, that would be 11 ranks, including
unranked, which is probably a maximal disapproval vote. (that's what
fits with traditions.)
>That all these methods/options are natural enhancements of ordinary
>Approval, and could be offered as options in an Approval election,
>makes it all more proposable than other approaches to voting system reform.
Yes. It all starts with Count All the Votes. Why didn't we think of
this years ago?
(Actually, I did.... and at one time we had a pretty good consensus
here that the first reform should be Approval. I realized, and kept
harping on, that this simply meant counting all the votes, that all
the flap about "best method," yatta yatta yatta, was distracting from
that simplicity. Uh, shouldn't we count all the votes, no matter what
system we are using? Actually using those votes, following a
reasonable interpretation of them, would almost always improve
outcomes -- or have no effect. UnFairVote then concocts completely
preposterous scenarios that make it sound like the sky will fall if
we use Approval Voting, and you can be sure they don't point out that
all that Approval means, really, is that we count all the votes.)
(UnFairVote is *against* counting all the votes, they do not want you
to cast an "illegal ballot," and they will toss out your votes if you
do. Even if your vote has a clear and obvious meaning. Think of those
Florida votes were the voter voted for Gore, and, confused by the
ballot design, and wanting to make it very clear, wrote, in the
write-in space, "Gore." Of course, that was an overvote. Ballot
discarded. Maybe it was Fred Gore of Tallahassee. If Florida had been
counting all the votes, a vote for Gore, the presidential candidate
(or, really, the pledged electors), and a vote for some unknown Gore,
a write-in. No harm at all.)
(Should anyone be so foolish as to want the very quirky method called
IRV, a misleading name concocted by UnFairVote to make it seem that
this was just a more efficient Top Two Runoff, but cheaper -- when it
performs very differently, and unfairly -- Count All the Votes works
fine with IRV, and, in fact, allows the voters to improve the method
by using overvotes in certain ways. It makes a three-rank system able
to handle many more candidates, etc. What, exactly, is not to like
about counting all the votes?)
>So then, why don't you find out what single-winner elections are
>conducted in your municipal jurisdictions, and then contact local party
>organizations and other political elections, and tell them about
>Approval. Suggest an Approval initiative.
Sure. However, I'd start by pointing out that simply counting all the
votes does no harm, and creates what is known as a quite good voting
method, Approval Voting. It fixes the spoiler effect, once people
know about it, and it costs practically nothing.
And that there is another reform, used widely in the U.S. at one
time, which was very popular, which worked, and which is even better,
which uses a ranked ballot, it's Instant Runoff Approval.
And if they are already using top two runoff, do Not try to sell
Bucklin as a runoff replacement. Instead, sell Count All the Votes
and Bucklin as a way to improve top two runoff, in ways that are easy
to show, which is likely to eliminate most runoffs. Cost savings.
Little or no cost.
Keep it very simple, unless they want more detail. For those who want
to know, a full plan for more sophisticated implementations can be
provided. Each step is an improvement, it can be done gradually, with
study at each step.
>And/Or, if you have a progressive city council or county board of
>supervisors (or whatever they call it in your area), suggest to them that they
>offer a referendum about it.
The first referendum would simply toss the section of the election
code that requires considering ballots with votes for more than one
(single-winner elections) as spoiled, and that would specify that
these votes are to be counted, and that the winner is the candidate
with the most votes. That's probably already there.
What needs to be done is to show that voting for more than one does
not violate the one-person, one-vote principle, any more than IRV
does. Only one vote has an effect in creating a winner. With IRV,
it's true, only one vote is active at a time, what approval does is
consider those multiple votes *simultaneously*. It ends up discarding
one or both of them! It cannot happen, with simple Approval, that
more than one vote influences the outcome.
(With Approval/Runoff, it can happen that multiple votes could
influence the primary outcome, as to who gets into a runoff. I'm not
dealing with that. When this happens, it is not an election, it's
really a poll of a kind, to determine, possibly, ballot position in
the runoff. Could you sign more than one petition to get a candidate
on a ballot? Of course you could!)
To my knowledge, the one-person, one-vote is the only knee-jerk
negative reaction to be expected from Count All the Votes. So
efficient and clear and effective ways of succinctly explaining this
issue should be developed and practiced.
Thanks for your work, Mike, it's been refreshing to see you back.
More information about the Election-Methods