At 02:19 PM 12/20/2008, James Gilmour wrote:
Abd ul-Rahman Lomax > Sent: Wednesday, December 17, 2008 12:42 AM
> LNH, has, I think, been pretty widely misunderstood. I don't consider
> it desirable *at all*. That is, it interferes with the very desirable
> process of compromise that public elections should simulate.
I don't have time to read any of the extended essays that now
feature on this list, but these two remarks in a recent post caught
my eye and I could not let them pass.
I don't expect anyone in particular to read all that I write. If I
write something of note, worth responding to, someone may respond....
that's as it should be.
LNH may well be pretty widely misunderstood, but Abd's view that it
is not desirable at all conflicts with my experience of the reaction
of ordinary electors.
LNH as an absolute principle, which, as an election criterion, it is,
is harmful. It prevents the system acting as a negotiator seeking
compromise, because it prevents compromise until and unless the
favorite is eliminated. Frankly, I doubt that anyone who fully
understands the implications would prefer an LNH system to one which
more appropriately negotiates on behalf of the voter, seeking the
best compromise. LNH means *no compromise unless you eliminate my
candidate totally!* That kind of position will readily be seen as
fanatic, intransigent, and selfish, in normal negotiation situations.
LNH in a system *enforces* this, requiring all voters to be just this
intransigent.
It is no wonder that a referee, reviewing Woodall's original paper
describing and naming Later No Harm, called it "disgusting." (This is
reported by Woodall in the paper.) So this is not just my view, James.
When preferential voting systems are first introduced to them, it
is a common reaction for them to
say "I'll vote only for my first preference because any later
preference would count against my most preferred candidate". It is
That's right. That's how they are used to voting. And, in fact, it
seems, this is how most vote when presented with a preferential
ballot, even if it *is* Later No Harm compliant.
Now, the real question is not whether or not Bucklin satisfies Later
No Harm, but whether or not it *sufficiently* satisfies it to mollify
the voters, who will bullet vote in fairly large numbers no matter
what system you feed them. Carroll, over 120 years ago, understood
why; he knew that with STV, most voters would still bullet vote,
because that's what they are clear about: they know who their
favorite is. They don't necessarily know how to rank the other candidates.
Look, when I see a nonpartisan local election for some minor office,
I'm lucky if I can even identify a single candidate! This isn't just
really ignorant voters, it's *ordinary* voters. Given that Bucklin
only looks at additional votes if there is majority failure, that the
vote doesn't really "harm" the favorite unless the additional vote is
considered in isolation from the first, I really doubt that real
voters would vote differently, in significant numbers, with a
three-rank Bucklin ballot vs a three-rank RCV (IRV) ballot.
You have to look at the forces which cause voters to bullet vote, and
LNH violation in a method would not be the most prominent reason, and
avoiding later harm, strategically, matters in only isolated and
relatively rare situations.
"Ordinary electors?" Consider this: ordinary electors, by the
definition of "frontrunner" prefer frontrunners. Usually there are
two; in a two-party system, almost always, in partisan elections,
there are two. These voters aren't going to add a lower preference
vote for the other major candidate unless you force them. Why should
they? And if they add a lower preference vote for a minor candidate,
it would mean that they support this candidate enough to make that
statement -- and to risk the possibility that, by gum, a plurality
thought the same way! Most people will bullet vote, if they prefer a
frontrunner, unless, again, you force them to do otherwise, but with
IRV, we don't even know, since those ballots are almost never exhausted.
So where does it make a difference? Well, if the voter favors a minor
candidate! *And* has a strong preference for another candidate, a
frontrunner, over the worst frontrunner.
Is this voter afraid to add a second rank vote? I don't think so.
These are the voters who will use additional ranks. And LNH is
meaningless to them because they know that their favorite is not
going to win. Their second rank vote is not going to "help defeat"
their favorite. (Even if that were what would happen with a one-vote
victory by the lower-preference candidate over the favorite; in fact,
take the voter's vote out, the result doesn't change. The voter did
*not* help defeat the favorite, it's merely that by voting for the
second preference candidate, the voter consented to that election in
the event that *other* voters preferred that candidate. The voter
will only do this if the voter either has low preference in the first
place, or fears the election of a third candidate.
LNH doesn't consider the context of failure. Nor, apparently, did you
or the electors you talked to. In reality, IRV suffers from, quite
possibly, the same level of truncation as Bucklin did. With Bucklin,
we can tell, because, generally, all the ranks were counted and
reported (in one case, election of the Cleveland mayor in 1913, even
when not needed). With IRV, we cannot tell how much truncation there
is because *not all the votes are counted.* In fact, it's possible
that *most* votes aren't counted.
By how you present the facts to a relatively ignorant voter (and most
people are ignorant about the implications of various voting
systems), you can elicit quite varied responses. I don't wonder at
your report, James, I think it probably is accurate. But that doesn't
affect what I said. How many people think of voting systems as ways
of negotiating a social compromise? How many are familiar with
Arrow's approach and Theorem? (The task: to construct a social
preference order from a collection of individual preference orders.)
only when it is explained to them that under the counting rules that
will actually be used, a second or later preference can never
harm their first preference, that they begin to see the merit in
marking all the preferences they really have. So Later-No-Harm
does seem to be important to ordinary electors, at least here in the UK.
Sure. No doubt about it. But garbage in, garbage out. First of all,
if a system requires a majority, which is known to be the only way to
ensure democratic results and truly foster healthy multiparty
systems, LNH failure is intrinsic, though it depends on the details
of the runoff. In a fully democratic system, there are no actual
eliminations, simply a reconsideration by the electorate, with minor
candidates, with no hope of winning but having made their point as to
their support, dropping out, or other compromises being made, and
there is no limitation on the number of ballots.
So, consider the first ballot in such a fully democratic process. A
voter is deciding whether or not to add a second preference. What's
the optimal strategy?
It's quite obvious: Unless the preference strength is relatively low,
you bullet vote. Suppose the method is IRV. If you add a second
preference vote, this could cause the lower-preference candidate to
gain a majority and win. Otherwise, there is a runoff and your
candidate, not being "eliminated," could proceed, with new
opportunities, to win, a very real possibility if the candidate is in
third place only by a small margin. (Consider Jospin in France, 2004,
tiny margin behind Le Pen, whom he would have defeated by a landslide
in a runoff, and he probably almost as strongly have defeated Chirac.)
(For IRV and LNH, I have the image of, they take them out back and
shoot them, that guarantees that your vote can't harm them since the
*method* has taken care of it for you. By preventing you from helping
other candidates, until your candidate is disposed of, other voters
are quite the same prevented from helping your favorite. Hence LNH
compliance is intimately connected with Center Squeeze.)
There are two very different situations in which to consider Abd's
assertion that purpose of public elections should be to simulate a
process of compromise.
When possible, people vote directly in meetings; standard
deliberative process avoids multiple-choice questions, but when they
are used, as is often the case with elections, requiring a majority
is the norm. We hold public elections because it's considered
impractical to directly elect through deliberative process and
unlimited ballots, even though we know quite well that this is the
best method from a democratic perspective. (It's possible to argue
that Range can do better, but I consider that purely theoretical, in
practice I would just as strongly assert that a Range result should
be approved of by a majority, relatively easy to do: an explicit
approval cutoff. And then, of course, you could get majority failure
and the need for a runoff.)
Taking the general first, where an assembly of some kind is being
elected (e.g. city council, state legislature, House of
Representative, Federal Senate), the fundamental requirement in a
representative democracy is for such an assembly to be
representative of all significant viewpoints among those who vote
(as expressed by their votes for the candidates who offer
themselves for election). So the purpose of such an election should
be to reflect that diversity. It should not be the purpose of
the election to manufacture some consensus in the determination of
the candidates who are to be elected.
I'll agree absolutely. Single-winner elections, actually, should
*not* be about *general* compromise, but the lesser kind of
compromise which is necessary for a smaller group to agree on a
single representative. There actually is a near-perfect method for
doing this: Asset Voting. Beyond that, STV does a quite good job. And
*in this context*, Later-No-Harm makes much more sense. But Asset
Voting was a tweak on STV, allowing voters to bullet vote, as Dodgson
knew many would, and still find representation. Necessarily, it means
that candidates aren't again, totally eliminated, they simply don't
get immediate votes; rather, the method looks for quotas and when a
quota is found with standing votes (not "eliminated"), the seat is
awarded. But, then, because there are exhausted ballots, those who
were in first preference position on those ballots may later exercise
them, to create additional seats. I don't know if Dodgson fully
realized the implications. With a system like this, I'd expect bullet
voting to become so common that the STV ballot would be dropped. If
your goal is to create representation, why not create it *absolutely*
and through a deliberative process where you are unconditionally
represented. Vote for your absolute favorite, with practically no
restriction. (There could be *many* candidates.)
The Dodgson system, later called Asset Voting by Warren Smith (and
implemented in a Range-like environment with the constraint that your
votes added up to one full vote -- only a mathematician!), used with
an STV ballot and canvassing method, is not LNH compatible, since
your second preference vote could deprive your first preference of
your vote for later exercise. Would any voter care? Sure. They would
bullet vote! Unless they didn't care, in which case they might add a
second preference, though it is really beyond me why anyone would do
so rationally. I trust Smith enough to prefer him for the office,
where he will effectively vote for me, and the rest of the quota, on
issues of importance, but I don't trust him to make a good decision
on whom to elect as that representative? This is a strange kind of
trust? It's more like cynicism, in fact.
Reflecting the diversity of voters' views is, of course, impossible
when a single winner is required in a single-office election
(e.g. city mayor, state governor). In this situation there MAY be a
case for suggesting that one of the purposes of the public
election should be to simulate compromise.
Of course. My entire discussion was about single-winner elections.
There is some kind of compromise necessary, it's highly unlikely that
everyone will get their first preference!
(In fact, with good deliberative process, people, as they learn the
preferences of others, shift their first preference to consider the
value of overall agreement. We could call this compromise, but it's
more than that. It is seeking a more sophisticated kind of
preference, one which recognizes the power of consensus, that it's
far easier to implement decisions when everyone is on board,
everything gets better for everyone, just about. But it is *really
difficult,* we think, to do this on a very large scale. And certainly
it's difficult trying to do it with simple direct discussion, except
under narrow circumstances.)
However, even then, most of our voters would expect the winner to
be the candidate who
has a majority of the first preferences even if some other candidate
had greater overall "compromise" support, i.e. they would
expect LNH to apply and operate.
Sure. This is because the concept of preference strength has been so
obscured for centuries, and even rejected out-of-hand by people like
Arrow. Yet it is obviously used and applied in ordinary
decision-making, including collective decision-making. Take these
same people and present them with what I've called the "pizza
election," and they will quite easily agree that the first preference
of the majority isn't the highest standard. It's an *approximation*
that *often* works, but not always.
However, James, you applied the majority criterion here. That's not
really fair. With the LNH violating methods like Bucklin, the second
rank vote, in the situation described, wouldn't produce a violating
result. It's only with majority failure that the possibility of LNH
violation comes up, and it's been misnamed; if it's presented
accurately to these electors, they might not be so quick to think it necessary.
Here is what I've seen with real people: in deliberative process, it
was clear that a strong majority favored one option. However, the
organization was one which valued unity, and certainly it was not
going to make a decision without hearing from everyone who wanted to
speak. And that's what happened (and deliberative process normally
requires a two-thirds majority to close discussion, so this is normal
that there would be full discussion of alternatives before voting on
a conclusion.) There were something on the order of a half-dozen
alternatives. So it was decided to do a poll. For each alternative,
voters were to indicate what they would be willing to "accept." This
was *not* going to decide the result. It was just a poll.
Previously, I'd heard a member of this group say that the status quo,
the majority favorite, wasn't going to change except "over my dead
body." That's how strong feeling was. And there were others, a
relatively small number, who felt that the status quo was utterly unacceptable.
It turned out that the majority favorite was acceptable to maybe 70%
of those polled. There was another option that was acceptable to all
but one. (Something like 60:1). Now, for the first time, a motion was
presented. To adopt the alternative. It was seconded and the question
was called, without objection. What was the vote for adoption?
James, it was 100%. Apparently the single holdout changed her mind.
Majority vote would have come up, quite apparently, with a defective
decision, if it were done prematurely, without adequate discussion
and information about the general opinion of the electorate. Was this
a "mediocre result." I don't think so. Years of experience have shown
the new option, the one adopted, to be far more effective for the
purposes of the group than the prior status quo, which was steadily
discouraging small numbers of members, accumulating to many over the
years. Yet it had been used for about fifty years, and there were
reasons why some were quite attached to it.
Now, the ultimate decision was still made by majority vote. But after
a process that included an Approval poll. A Range poll would have
been even more informative, but the Approval method was sufficient,
as it would often be. Approval is a lot simpler to canvass, just
count all the votes, it can be done by show of hands, as it was in this case.
When there is no majority winner they may well be prepared to take
a compromising view, but there
are some very real difficulties in putting that into effect for
public elections.
Bucklin did it. It worked. There are some reasons for believing that
Bucklin had roughly the same level of bullet voting as IRV
(single-winner) does, when ranking is optional.
Most voters, quite simply, coming out of a Plurality environment,
where LNH simply isn't on their radar, are going to vote with
reasonable sincerity. If they have a strong preference for their
favorite, they will bullet vote. If that preference is weak, or if
they fear that their favorite doesn't have a chance, they will add
additional preferences. Every preference they express is sincere. The
averaging causes this process to show preference strength estimates.
The voters who think that their favorite doesn't have a chance are
quite accustomed to making compromises. They won't give a fig about
LNH. They'll be happy to be able to express their first preference,
without harm to their voting power.
Show these voters a Center Squeeze result from IRV, and they will be
horrified. Point out to them that if their preference for the
compromise candidate, over the remaining candidate, the worst of
three from their point of view, who goes on to win the IRV election,
which preference they took the trouble to express on the ballot, and
in which they agree with two-thirds of the voters, can't be
considered until it's too late, and won't be counted at all, they
will start to have less respect for the Later No Harm criterion.
Later No Harm has been elevated by FairVote to practically the most
important election criterion. The far more significant Condorcet
Criterion is totally neglected by them. This is raw appeal to
ignorance, LNH *sounds good*, but actually stinks. *Other things
being equal,* sure, LNH could be a good thing. But they are *not*
equal and, I hope I've shown, LNH conflicts directly with highly
desirable characteristics of voting systems.
I suggested, years ago, a system called A+P/W, or Approval Plus,
Pairwise. There is another name for it, but I don't recall. It was
Approval Voting, with a Preference indicator (the "Plus"). If the
Plus indicator was used, the vote would remain an exclusive vote for
the preferred candidate in the pairwise elections involving other
approved candidates. It's a two-expressed rank system, with equal
ranking allowed (I would have allowed multiple use of Plus, just
because I don't like discarding ballots. Count All the Votes.) ....
This would be LNH compliant. But I think Bucklin, simpler to canvass,
is quite adequate and sufficiently protects voters from a reasonable
fear of "harming their favorite."
Approval is maximally vulnerable to this LH fear, in a
three-frontrunner environment. It's simply moot with two
frontrunners, which is most elections, by far. Range ameliorates it
(you can rate intermediate ratings and, sure, that can reduce the
power of your vote for A over B, but it will increase the power of
your vote for B over C. It's a tradeoff.) Bucklin is even more
protective. IRV totally protects, but at great cost. All the methods
get better if real runoffs are used when a majority hasn't been
found. (IRV should be used as a two-winner method, perhaps, if it's a
primary stage. In Vermont, gubernatorial election, plurality, no
majority? Top Three go to the legislature for secret ballot choice.
Plurality, I think. Yes, sometimes the third place candidate won.
Politics, certainly FairVote claimed that. But there are other
possibilities. That could have been the compromise candidate,
possibly eliminated with both IRV and Top Two Runoff.
----
Election-Methods mailing list - see http://electorama.com/em for list info