[EM] FairVote Propaganda: Bucklin Voting.

[Abd ul-Rahman Lomax]

This information is taken from the page http://www.fairvote.org/?page=2077

There is a shortage of information on Bucklin. FairVote, through this 
article, has been one of the main sources that others have used in 
describing it and its history. FairVote's analysis is hardly neutral, 
it's highly biased, drawing unsupported conclusions from the data; 
unfortunately, it's not easy to come by data, aside from this 
article. That's something that an enterprising student with access to 
old newspaper and other archives could do, report on the actual 
performance of Bucklin, and the actual arguments made with 
implementations and the replacement of Bucklin.

>An Early 20th Century American Use of an Alternative Voting Method

Bucklin was widely known simply as "preferential voting." (As in 
Duluth and the Minnesota Supreme Court decision in Brown v. 
Smallwood). The ballot was generally a three-rank ballot, like RCV. 
How the ballot was interpreted varied with the implementation.

>During the first decades of this century at least six states1used a 
>form of ranked voting in primaries known as Bucklin Voting (named 
>after its original promoter, James W. Bucklin). This system allowed 
>voters to give first and second choices2. In the ballot count, all 
>first choices are counted. If no candidate wins a majority, all 
>second choices are counted simultaneously and added to the totals of 
>the two candidates with the most first choices - meaning voters 
>could have two votes count if they ranked two people. The candidate 
>with the most votes wins. As states transitioned from nominating 
>candidates at conventions to nominating them with primary elections, 
>this systems was proposed as an alternative to accepting 
>unrepresentative plurality winners or the use of costly runoff 
>elections with reduced turnout.

This was inaccurate as to Duluth, anyway, and as to the Oklahoma 
method, as well; the ballot allowed three ranks, not two. My guess is 
that FairVote is accurate, here, as to the goals of Bucklin 
implementation; however, I'd note one caution: runoffs cost money, 
but reduced turnout in them is probably not a real problem, it only 
looks like one. Voters who don't care about the result, relatively 
speaking, don't turn out to vote in a runoff, reducing turnout. But 
increasing turnout, somehow, wouldn't necessarily change results for 
the better; rather, it would simply introduce noise. We can imagine 
that low turnout is due either to voters being disgusted with the 
limited choice, or satisfied with both candidates in a top-two 
runoff, they will support either one of them in the coming full election.

Bucklin was not only used for primaries. I know of a mayoral election 
that used Bucklin, and Smallwood was a judge elected by Bucklin.

http://query.nytimes.com/gst/abstract.html?res=9403E3DF173DE633A25756C0A9679D946296D6CF
http://query.nytimes.com/mem/archive-free/pdf?_r=1&res=9403E3DF173DE633A25756C0A9679D946296D6CF

>Bucklin voting, unlike instant runoff voting, or Condorcet voting, 
>falls in to the same class of voting methods as Approval voting, the 
>Borda count and Range voting.

Well, it's a hybrid method. It uses a ranked ballot, as it was 
implemented, and, in Duluth, only one vote was allowed in the first 
two ranks, with it being possible to approve multiple candidates in 
the third rank. The lower ranks are only counted if the previous 
ranks don't find a majority, so, unlike Approval and Range -- unless 
some device is added to them -- Bucklin satisfies the Majority 
Criterion under all interpretations. Bucklin is both a ranked method 
and an Approval method. It collapses into pure Approval if no 
majority is found in the first two rounds (Duluth). It collapses into 
a kind of Range in the Oklahoma implementation (where second and 
third rank votes were given fractional values.)

And here we get to the propaganda:

>  These voting methods suffer from an important flaw. Voters quickly 
> recognize that voting for an alternate choice may help defeat their 
> first-choice candidate.

Duh! However, this is probably moot for most voters, that's where the 
FairVote argument is highly deceptive. This is the argument that 
Later No Harm is important to voters. In pure Approval, it's 
definitely a major consideration, probably, but with Bucklin, first 
preference holds, unopposed by lower ranked votes *unless no majority 
is found.* And then, the lower votes don't actually "help defeat the 
first choice candidate," rather they, as they later state, represent 
a withdrawal from that particular pairwise election, but a new 
participation in, possibly, a series of other pairwise elections. The 
voter gains power.

Most voters, though, it seems, when full ranking is optional, don't 
add additional preferences. IRV conceals this, because the largest 
group of voters who don't add additional preferences, probably, are 
those who bullet vote for a frontrunner. Those who vote for the top 
two, their ballots are never uncovered, they are not exhausted, we 
only see exhausted ballots (in high percentages!) from those who vote 
for minor candidates. Generally, when one is voting for a minor 
candidate, there is *more* motivation to add additional preferences, 
so we can expect that there will be higher levels of bullet voting 
among those who vote for frontrunners.

Bucklin counts all these votes, so we can expect it to be more 
efficient than IRV in finding majorities, unless, indeed, the Later 
No Harm thinking causes more voters to not add additional preferences 
than with, say, Instant Runoff Voting. While I'd expect this effect 
to be relatively strong with Approval, the motivation for it is 
drastically reduced with Bucklin, and my expectation is that most 
voters would vote Bucklin exactly as they would vote IRV. But the 
Bucklin counting method is known to produce results that will, 
overall, more widely satisfy the voters.

>  For example, if both your first choice and second choice advanced 
> to the second round, your ballots would cancel each other out.

That's right. If no majority is found in the first round, the second 
round votes are added in. Suppose it's a three candidate election 
(most accurately, there are three frontrunners, running neck and 
neck, it seems -- that's rare, by the way). You vote for A in first 
rank. You have voted in the A/B and A/C pairwise elections, and have 
abstained from the B/C pairwise election. No majority for A, your 
vote for B is now added to the B total. So, yes, you have just 
effectively abstained from the A/B election, except that, of course, 
you did clearly indicate a preference for A, and it might be possible 
to use that later in a more sophisticated version of Bucklin. But you 
gained a new pairwise election: you have now voted in the B/C 
election. Is that important to you? If not -- don't add the vote for B!

But if you would rather see A or B elected, and not C, then you could 
add the vote. Which is more important to you? This is like Approval 
strategy, but improved, you did make an effort to elect your favorite 
A, and you have not withdrawn your vote in the A/C pair, it's still there.

In Oklahoma Bucklin, your second rank vote would have been added in 
as a half vote. So you'd still be maintaining an edge for A over B; 
but you haven't exerted a full vote in the B/C pair.

FairVote has not shown any evidence, ever, that Later No Harm 
considerations result in real and significant changes in how people 
will vote with a ranked ballot. Consider this: in most elections, 
there are two frontrunners, and most people know it. Under what 
conditions would LNH considerations -- as distinct from a pure 
tendency to bullet vote for a favorite, which affects any ranked 
method as well as Approval and Range -- affect the votes and the outcome?

The one who prefers a frontrunner is not going to add a second rank 
vote, except rarely, to the other frontrunner. That's not how people 
vote! They vote to make a realistic choice, most of the time, and 
they know that the real election is between (usually) two candidates, 
so they make sure to vote for one of them. In Plurality, that's 
simple! In methods with a more sophisticated ballot, though, they do 
have additional choices. If they prefer a minor candidate second, 
they may add that vote. Usually this is not going to affect the 
outcome, and they will normally not be concerned about "harming" the 
frontrunner they prefer. But it's really moot if they add a second 
preference vote or not, unless they add it for the other frontrunner.

Most people, by definition, prefer one of the frontrunners. If that 
weren't true, they wouldn't be the frontrunners! So most voters *may* 
truncate, add no additional vote, or they may add harmless ones, moot 
for the outcome, but possibly making some political statement. There 
is an example of that in the Cleveland election reported above: the 
leader in second and third preference votes was ... the Socialist 
Candidate! Were the supporters of the Democrat (or Republican? -- 
this was 1913!) afraid that their second rank vote would cause the 
Socialist to win? Or was it that they did not understand the system, 
they did not prefer the Democrat, but the Socialist, and voted 
insincerely, not realizing that there was no advantage in this. The 
article doesn't give overall vote totals, but there is enough 
information to infer that the lead of the Democrat and Republican 
over the Socialist was very large. This election, by the way, found a 
majority for the Democrat, in the second round. The third round 
wasn't necessary.

>  For this reason, in high stakes elections in which voters have 
> strong favorites, most voters opted to "bullet vote" and protect 
> the interests of their favorite choice be withholding any alternate 
> choices. In Alabama, for example, in the 16 primary election races 
> that used Bucklin Voting between 1916 and 1930, on average only 13% 
> of voters opted to indicate a second choice.

This is pure fantasy, imagining the motives of "most voters." Most 
voters bullet vote because that's a sensible vote if you are voting 
for a frontrunner ("most voters") and prefer that candidate to all 
the others! You only add an additional preference if you fear that 
your first rank vote won't be adequate to prevent the worst 
candidate(s) from winning. 13% of voters is probably quite good! 
Notice that in other elections where we have records, such as the 
Cleveland election cited above, and the Duluth election of Smallwood, 
there was much, much higher participation with additional choice votes.

>A flaw unique to Bucklin, with its limit of two choices was that if 
>a voter's second choice was not one of the top two initial count 
>candidates, their second choice vote was wasted.

First of all, all the Bucklin implementations I've looked at had 
three ranks. And the third rank allowed pure approval voting. So not 
voting for a frontrunner meant (1) you were ignorant, or (2) you 
didn't care, you thought the frontrunners were Tweedledum and 
Tweedledee. Since I haven't seen the two-rank version, I don't know 
if the second rank allowed more than one vote. If it did, the 
FairVote argument becomes even more spurious.

However, here, we see the result of a totally biased approach to 
describing voting systems other than IRV. In every system, including 
IRV and Bucklin, if you don't vote for a frontrunner, your *entire* 
vote is almost always "wasted." Except that people vote for reasons 
other than participating in the choice of a winner. Consider Nader 
voters in Florida in 2000. With Bucklin, your vote is always counted, 
apparently (that's a choice of election officials, perhaps; but the 
Cleveland election report in the New York Times showed totals for the 
second and third ranks even though only two rounds were counted to 
determine the winner.

There is nothing about Bucklin, in principle, that limits it to two 
ranks, as is implied here (A flaw unique to Bucklin....) No, that's a 
flaw of limiting the number of ranks, as with three-rank RCV. Same 
problem. In quite a few cases, with IRV, your second and third rank 
votes aren't counted at all (in fact, perhaps for most ballots). So 
to present this argument as an argument against Bucklin -- in a 
context where the comparison is clearly with IRV -- is little short 
of bizarre. FairVote calls whatever method uses something like single 
transferable vote "IRV," and they've included 2-rank versions. Almost 
all versions of IRV implemented in the U.S. have been three-rank, 
same as Bucklin in most implementations.

Sauce for the goose is sauce for the gander?

>  As a result, the winner with the plurality of combined first and 
> second choice votes could easily fall far short of a majority. 
> Using Alabama as an example again: With 16 primary elections 
> between 1915 and repeal in 1931, in no case did the addition of the 
> second choice votes give the winner a majority (the purported goal 
> of the system). And in only one case did the addition of the second 
> choice votes change the outcome from the original first choice 
> plurality candidate.

Fascinating! This is quite what I would expect. Why? Well, it appears 
that the same thing happens with IRV! The only difference is that IRV 
pretends to find a majority by *discarding and considering moot all 
ballots which don't contain a vote for the two frontrunners!* Primary 
elections are "nonpartisan elections," vote transfers don't happen as 
if the candidates were clones in some respect. (I.e., we may be able 
to predict that a Green supporter will second rank a Democrat, or the 
like. But not that a Joe supporter will second rank Mary.) 16 primary 
elections between 1915 and 1931. How many nonpartisan IRV elections 
have there been?

Before November, in the U.S., there were something like 32, I forget 
the exact number. In *every case* the winner of the election was the 
"first choice plurality candidate."

In a *few* cases, essentially close elections, vote transfers gave 
the winner a majority. And they certainly did that with Bucklin 
elections as well, just not the primary series cited. Single ballot 
methods, period, don't replace the need for runoff elections. IRV 
certainly doesn't.

The data given here by FairVote, selective and biased as it is, 
supports the possibility that Bucklin is better than IRV. In theory, 
Bucklin is better than IRV at finding majorities, since IRV conceals 
possible majorities underneath the first preference votes for the 
other frontrunner. (This may not be a common vote pattern, but it's 
more common in nonpartisan elections than in partisan ones, I'm 
sure.) If the concern with IRV were truly finding a majority, all the 
votes would be counted, including those concealed ones. But IRV is 
*replacing* majority requirements in the U.S., a major anti-democratic tragedy.

Real runoff elections produce a "comeback" election in about one 
election out of three where a runoff is needed. IRV is producing *no* 
comeback elections (in the nonpartisan context). Bucklin, I'd expect, 
would find *some* of these comeback winners, but not all. It does not 
replace runoffs, it merely avoids *some* of them, under *some* 
conditions. There was one comeback reported in 16 elections. That 
might be about right! If that ratio held, and IRV were equivalent to 
Bucklin in this respect, we'd have expected to see two comeback 
elections. The other one would be a result, perhaps, of voters 
getting a good look at the two candidates and deciding which one was 
better based on the new information they have, and, as well, being 
motivated to turn out and vote. Real runoff elections are far 
superior to "instant runoff," and pretending that single-winner 
single-transferable-vote is some simulation of real runoffs is highly 
deceptive.

>Some civic leaders suggested that a way to fix this problem would be 
>to require voters to express a second choice. This was rejected 
>partly out of respect for voters who voted on principle and 
>genuinely had no second choice in a particular field of candidates. 
>But even if adopted this still would be problematic as voters would 
>have an incentive to disingenuously pick a second choice they 
>believed had no chance to win, so that they would not accidentally 
>help defeat their first choice candidate.

This was done, actually, in Oklahoma. And it's done in Australia, 
usually held up as the shining example of "IRV" by FairVote. The 
Oklahoma requirement was found unconstitutional, for quite good 
reason, and it had nothing to do with the Bucklin method itself, it 
had to do with not counting bullet votes; with the concept of forcing 
voters to vote for someone they might detest. Apparently those who 
wanted election reform in Oklahoma were unable to pass it again.

(There was a dissent in the opinion that said that the full ranking 
requirement should be simply discarded, and that the rest of the law 
should stand. That this wasn't the decision is simply an indication 
to me of the usual: there are powerful forces against election reform.)

>There is a fundamental difference between this faulty "preference" 
>voting system and standard instant runoff voting in terms of wasted 
>votes and the encouragement of disingenuous strategic voting.

Perhaps. Most analysts think that "disingenuous strategic voting" 
isn't a common phenomenon. The Later No Harm criterion itself is 
controversial among voting systems theorists, some being "disgusted" 
by it, because it attacks the very foundation of democratic process, 
the finding of compromise. If my vote can't be interpreted to 
compromise until my candidate is effectively dead, then, sure, my 
second rank vote can't "harm" my candidate (usually the overall vote 
isn't a "harm," though that possibility exists with IRV, because IRV 
is a non-monotonic system where voting for a candidate can cause that 
candidate to lose), but neither can the second-rank votes from other 
voters "help" my candidate. It cuts both ways. IRV prevents voters 
from indicating a compromise *without* eliminating the favorite. 
Obviously, if you do that, the system may indeed find the compromise, 
but if the voter looks at the results, the voter would find that, if 
this wasn't the voter's favorite, the voter's vote, taken entirely 
out, wouldn't change the result. The bullet vote would simply have 
caused, perhaps, majority failure. Which is why runoff elections are 
the cat's meow.

Single-winner elections waste votes; if they are allowed to terminate 
with a plurality, they waste *most* votes. The bulk of 
lower-preference votes in IRV are *never* counted, at least not by 
the election officials. We don't know, unless we do direct ballot 
analysis ourselves, in the few places where, say, "ballot images" are 
available, how many voters even bothered to add in additional votes.

We know it with Bucklin, since, generally, all the votes are counted. 
(It's easy, just Count All the Votes, then add them up as needed. Any 
voting equipment can handle it, Bucklin is almost as easy to count 
and canvass as Plurality.)

>  There are several concepts that it is useful to keep in mind, when 
> evaluating voting systems, including: 1) Does the system discourage 
> manipulation through disingenuous strategic or bullet voting? 2) 
> Does the system minimize the number of "wasted votes?" 3) Does the 
> system promote majority winners?

Okay:

(1) Both Bucklin and IRV encourage sincere voting. Bullet Voting is a 
normal "strategy" followed, probably, by most voters, both with 
Bucklin and IRV. The cause isn't necessarily LNH, but rather that the 
voter doesn't have enough information to intelligently rank more than 
one or maybe two. It's an old problem with preferential voting 
systems. We don't have head-to-head comparisons, but the LNH 
violation by Bucklin is unusual and conditional. Most voters who 
would be motivated to add additional preferences, with either system, 
would have little or no fear of "harming" their favorite. The 
Socialist voters in Cleveland who, apparently, added second 
preference votes for the Democrat, in fairly large numbers, which 
took the Democrat over a majority, did they "fear" that this would 
cause the Democrat to win, instead of the Socialist? My expectation 
is that voting patterns under Bucklin and IRV would be almost 
identical. But with the same votes, Bucklin would choose better 
winners, because it is not afflicted with the most important problem 
of IRV: Center Squeeze.

(Center Squeeze is mentioned in Robert's Rules of Order, Newly 
Revised, 10th edition, as a problem with the method of preferential 
voting that they describe, which is, in round outline, IRV, but with 
a crucial difference: a true majority is required, not the false 
majority that IRV finds by rejecting ballots which don't contain a 
vote for a frontrunner. The problem is that a compromise winner, who 
might be preferred, even strongly preferred, by most voters (could be 
two-thirds of them!), can fail to get enough first preference votes, 
and so is eliminated. This is a known problem with real Top Two 
Runoff, and IRV only fixes it under some conditions, not in others. 
To avoid the problem with IRV, voters must vote insincerely, or risk 
the victory of a hated candidate. Is this "disingenuous strategic 
voting?" Certainly looks like it! With Bucklin, the voters can vote 
sincerely, they are simply more motivated, in a three-frontrunner 
situation, to add an additional ranked vote. Bucklin, because it 
doesn't eliminate candidates, then, can find the compromise winner.)

(2) This measure, "wasted votes" favors Bucklin, since all the votes 
are, at least, counted. FairVote is referring to votes that didn't 
rank a frontrunner, and IRV, in real implementations outside 
mandatory full ranking, wastes the same votes, plus most votes aren't 
actually counted, and most importantly, the lower rank votes for what 
would be a compromise winner, favored by most voters -- even by a 
large margin -- over the IRV winner, aren't counted. Talk about wasted votes!

Count All the Votes! That covers Approval, which is just Plurality 
with all the votes counted, including "overvotes." And it covers 
Range and Bucklin and *certainly not IRV*. (It also covers Condorcet 
methods, which count all the pairwise elections from a ranked ballot.)

3) Does the system promote majority winners? IRV does *not*, in most 
applications. In nonpartisan elections, it is a glorified and 
expensive form of Plurality, differing very little in result. Bucklin 
probably does a little better, finding a majority in situations where 
IRV won't report it, but indications are that *usually* IRV and 
Bucklin will find the same result. The exception might be with 
roughly one election out of 16, where Bucklin will find a compromise 
winner that IRV will miss.

However, the system that truly promotes majority winners is Top Two 
Runoff. In order to discredit and avoid this, IRV redefines majority 
in several ways. In order to imagine that IRV "promotes majority 
winners," they have to be referring to the "last round majority," 
which does great violence to the concept of a majority winner.

In Australia, where full ranking is required, the election procedures 
refer to an "absolute majority" as the target that a candidate must 
meet to win. However, when it becomes Optional Preferential Voting, 
the procedures change to read "majority of votes for continuing 
candidates." That is no longer a majority of votes, no longer 
represents a decision by a majority of voters to support the 
candidacy of the winner. And that is what we are looking for when we 
seek a majority. Under Robert's Rules of Order, that's what 
"majority" means, that a majority of those who cast a non-blank 
ballot decided to support the winner, regardless of voting system used.

There is no way to guarantee this beyond these:

(1) Require all voters to rank all candidates using a preferential 
system such as Bucklin or IRV.

(2) Use a multiple ballot system, where the vote may change.

The first option is probably unconstitutional in the U.S., it 
certainly was in Oklahoma, with a quite solid decision. The second is 
what has been implemented in the U.S. to further the goal, and, 
typically, it's been considered that the second election can 
terminate with a plurality.

In most places, it is pure Top Two Runoff, with vote-for-one in the 
primary, and, then, only the top two eligible in the runoff, which, 
then, *must* find a majority, since any ballot with any other vote is 
considered spoiled. While this is similar to IRV, in certain ways, 
there is an additional detail that makes it very different: in some 
places, such as in California by default, write-in votes are allowed 
in the runoff. And, in fact, in Long Beach, California, in a recent 
election for mayer, a write-in won the runoff. But only by a plurality.

Still, the only method which seeks a true majority is Top Two Runoff, 
and if we desire to continue this -- and FairVote is obviously 
arguing that majority results are valuable, and I'd agree -- then, 
for continued reform, we would need to use a better primary method 
and probably a better runoff method, than vote-for-one. (A 
vote-for-one primary, where a majority is not found, is a two-winner 
election, plurality at large, to determine who goes on a runoff 
ballot, and the same as the first round for IRV.)

To me, there is an obvious and simple candidate for reformed primary 
and runoff: Bucklin. The harm of bullet voting there would simply be 
an increased need for a runoff. The runoff, if two-rank Bucklin, 
would allow write-ins without creating spoilers, while, unlike IRV, 
not creating a Center Squeeze effect in the runoff. Most voters will 
not use the additional rank, unless there is some serious defect in 
the primary making a write-in candidacy important. Bucklin, I would 
guess, based on estimates from looking at the RCV ballots in San 
Francisco, would avoid as many as half of the runoffs, and it doesn't 
incur the major counting costs that IRV does. So a Bucklin-runoff 
system, my estimate, would be cheaper than an IRV system without 
runoffs, and *far* more democratic -- in perhaps 5% of elections or 
so. Otherwise the same.

It is also possible to set the majority requirement in a more 
sophisticated way. Where there is adequate margin between the 
frontrunner and the runner-up, it would take a huge -- and very, very 
unusual -- shift in the voting patterns to see a comeback election. 
So, if saving money is considered very important, some  of the 
runoffs might be avoided with a specification of some margin in votes 
rather than insisting on the purely democratic majority. We'd know 
better with more experience with runoff/Bucklin or even runoff/IRV.

The important thing is to seek majorities. IRV -- "instant runoff 
voting" -- was misnamed, as a brilliant piece of political promotion, 
implying that it does what it does not do. Real runoff voting has 
long been considered an important reform, and it's the only voting 
system known, in actual practice, to encourage strong multiparty 
systems. IRV, quite simply, having practically the same effect as 
Plurality (but occasionally switching the result from one major party 
to the other, practically *never* to a minor party), doesn't do that. 
The proof is in the pudding: IRV isn't used by any true multiparty democracy.

>To accurately aggregate individual opinions into a community 
>decision a voting system should encourage citizens to honestly vote 
>according to their consciences.

Sure. Bucklin does that, actually, the LNH fear isn't particularly 
rational with Bucklin. The bullet voters in the Alabama primaries? 
They were sincerely voting their consciences, they simply didn't care 
enough about the other candidates to rank them. That's going to 
continue no matter what voting system is used; if we saw 13% 
additional candidate ranking with 2-rank Bucklin, what I'd want to 
know -- and they don't tell us -- is what the vote totals were, and 
the lead of the frontrunner in the final vote, not just whether or 
not a majority was found. Sometimes it's pretty obvious that a real 
runoff won't find anything different, the lead is too great. 
Primaries may be full of relatively unknown candidates, with voters 
not having as good an idea of who might be leading. The same voters 
would probably bullet vote with IRV. And, apparently, they did, when 
IRV was used for political primaries. Or do I have that confused with 
Bucklin -- also called "preferential voting."

In any case, we know that even with partisan elections, in Australian 
IRV, when full ranking is optional (They call it Optional 
Preferential Voting), bullet voting is quite common, and the results 
don't show the full extent of it, since raw ballot data is generally 
not available and as long as the voter ranks one of the frontrunners, 
the ballot is never considered exhausted and we never know if it 
contained additional votes. It is probably *very* common.)

In San Francisco, in one IRV election, the winner got less than 40% 
of the vote. San Francisco *wanted* to keep the principle of majority 
vote, but what they *actually* did was strike the majority 
requirement from the election code; the committee that put together 
the voter information pamphlet for the Proposition that implemented 
"Ranked Choice Voting" there was highly misleading on this point, 
saying that the "winner would still be required to gain a majority."

(There is no such "requirement" on the winner, the kind of "majority" 
that IRV finds is mathematically guaranteed, since any ballot which 
doesn't contain a vote for the top two, but only for eliminated 
candiates, is excluded from consideration in the basis for this "last 
round majority." If this were adequate as "majority," then it would 
be simple to get the same with Plurality. Just exclude all the 
ballots which don't contain a vote for the top two. The remaining 
ballots show a majority winner, by definition. In fact, with IRV, 
just carry the elimination one more step and Presto! Unanimity for 
the winner. Now, wouldn't that be nice?)

>  Approval, Borda, Range, Bucklin, and at-large multi-seat voting, 
> for example, fail this test.. Many "smart" voters with a strong 
> first preference will "bullet vote," refraining from expressing any 
> alternate choices. This is not a difficulty with systems such as 
> instant runoff voting (IRV), however, since voters who select 
> second or subsequent choices cannot thereby help defeat their first choice.

With any system, most voters with a "strong first preference" will 
bullet vote, period. If the first preference is a frontrunner -- by 
definition, most voters -- this is actually quite an optimal 
"strategy," and it is a sincere vote. Let's forget about at-large 
multiseat voting, where methods designed for single winner don't 
apply, unless additional procedures are implemented (there are Range 
and Approval multiwinner methods).

FairVote repeats this interpretation of Later No Harm over and over. 
"Cannot thereby help defeat." *If* we think of the voter as being two 
voters, we could say that the second voter can "help defeat" the 
first voter's preference. But the voter is not two voters, and only 
casts one ballot, and this ballot *cannot*, with Approval or Bucklin, 
"help defeat" the favorite. Rather, at most, it pulls back and 
abstains from the single pairwise election between the Favorite and 
the Lower Preference.

The *voter* did not "help to defeat the first choice." Rather, take 
the voter's ballot away, the result doesn't change, so we know that 
there was no "help." What happened is that there wasn't a majority in 
the first round, where the voter expressed a totally sincere 
preference. (In fact, if we want to increase sincerity in Bucklin, we 
*allow* multiple first preference votes, but that's another story; 
all the Bucklin implementations did not allow that.)

Does the voter prefer a frontrunner? Two-candidate frontrunner case: 
Voting for only one of them is fully strategic *and* sincere. 
Additional votes are *moot* for determining the winner, except if the 
voter is radically mistaken about the frontrunners: so they are 
generally harmless. If one would be distressed by ranking a minor 
candidate, and the candidate won, *don't* add the vote! Don't vote 
for anyone you wouldn't be *relatively* pleased by seeing win!

Three-candidate frontrunner case (rare). This is clearly a more 
difficult situation, and it's really up to the voters to decide 
what's more important: making sure that the favored frontrunner wins, 
or making sure that the worst frontrunner loses. In most situations, 
actually, this isn't a terribly difficult decision for most voters. 
Bucklin is nice, compared to pure Approval, because it does allow 
that totally sincere first preference vote, or even a totally sincere 
second rank vote, in the Duluth implementation. The third rank is 
probably enough to cover most voters, and because it was an approval 
round, it can be reserved for an AntiPlurality vote: vote for all 
possible frontrunners against the worst frontrunner.

(So someone seriously worried about C winning but really preferring A 
to B with reasonable strength could, with good rules, vote A first 
round, leave second rank blank (good rules), and then vote for B in 
the *third* round.)

Most voters will simply vote a ranked Approval vote. First rank, 
Favorite. Any others "approved?" -- i.e., the voter would be relieved 
or even pleased to learn that this was the outcome -- then add second 
rank, and maybe third, but most voters won't need that.

The actual votes depend on context. The magic amalgamation of 
"sincere preferences" into some ideal social order is impossible, 
that's what Arrow showed. There is a system known, however, to 
satisfy a reasonable interpretation of Arrow's conditions, and, to 
make a long story short, the systems that reasonably satisfy it are 
Range and Approval. Part of the system, if a voter cares to maximize 
impact on the result, involves the voters estimating election 
probabilities, which is what they are entirely accustomed to doing, 
Plurality requires this of us. (We don't waste our vote, most of us, 
writing in the name of the number one best possible person for the 
office being elected. Neither will we waste our voting power in *any 
method* doing this, but methods like Bucklin allow us to, indeed, to 
write in an absolute favorite who has no chance of winning and 
*still* exercise, with practically no loss, our full effective vote.)

IRV in theory does this as well; however, with RCV, there are only 
three ranks, and overvoting isn't allowed in any of them. So if you 
do it with RCV, you only have two ranks left. Reserve one for a 
frontrunner, that's it. If it's a three-frontrunner election, and you 
prefer another candidate on the ballot, you don't have enough ranks, 
your choices have become quite limited. With a three-rank Bucklin 
ballot, you could write in the absolute first preference, then your 
ballot preference, then add two approval-style votes in third rank, 
so as to help defeat the worst.

>The Bucklin system failed to achieve its intended goal of preventing 
>split votes resulting in minority selected nominees. It was in fact 
>no better than the plurality rules common today.

He's describing IRV, in fact, not Bucklin. Bucklin *did* find 
majorities in some of the elections, he has just looked at the 16 
Alabama primaries. Look at the election in Brown v. Smallwood. Look 
at the Cleveland mayoral election in 1913, referenced above. I'm sure 
there are others.

It ought to be common sense: there isn't any way to "prevent split 
votes" with a single-ballot method except to exclude the supporters 
of minor candidates who don't care to vote for a frontrunner. IRV 
can't magically create a "majority," no method can, without coercing 
votes. Top two runoff, as an overall system, will almost always find 
an ultimate majority, and *usually* it will be a legitimate result. 
Sometimes it's a result forced by Center Squeeze, the automatic IRV 
result, but, at least, with a 2-candidate-limited runoff, the 
electorate explicitly makes its choice. Allow write-ins and that can 
be fixed. Use a better primary and runoff method, the result will be 
almost perfect.

IRV, in nonpartisan elections, the little secret that FairVote 
doesn't want us to know, almost always -- no exceptions so far in the 
U.S. with plenty of examples, now -- elects the Plurality winner. 
Something like two-thirds of the time, there is a majority in the 
first round. This means that, if the votes were sincere, and we have 
little reason to think they were not, this same candidate would have 
been elected with Plurality. (FairVote made this argument above re 
Bucklin. Defending IRV, I've seen them change their tune a bit: but 
what if the supporters of that candidate didn't realize that they 
were in the majority? Well, if the supporters of a frontrunner don't 
realize that the candidate is a *frontrunner*, they are seriously out 
of touch; in Plurality, they will do what we naturally do: vote for 
our Favorite. It actually works, Plurality is not a *terrible* 
method, it's much better than pure voting theory sometimes shows, 
because people vote it intelligently, which means they make the 
necessary compromises in deciding how to vote.)

However, Top-Two-Runoff, in roughly one election out of three (29% of 
Texas primary elections, in a study by FairVote), shows a "comeback 
election," where the electorate explicitly, by majority vote, rejects 
what would almost certainly have been the IRV winner, in favor of the 
runner-up in the primary. Bucklin would find some of these, quite 
likely, making a better result without needing a runoff. The other 
situations would cause majority failure, and the only way to fix that 
is with a real runoff.

>The debates over repeal of Bucklin voting systems often offered 
>second round runoff elections as a majority rule alternative to the 
>minority rule Bucklin system.

And now, what we see, as a result of the diligent efforts of 
FairVote, is a reversal of this, where second round runoff elections 
are being replaced by the minority rule IRV system, a system clearly 
worse, in terms of result, than both Bucklin and second round runoff.

Second round runoff probably produces results that are a little 
better than Bucklin. That second round is a powerful tool! But there 
was a smarter choice: Use Bucklin first round, and still require a 
majority. Now, if we looked at that Alabama series, we see one 
election out of 16 where the result shifted. That's not bad! It's 
probably worth having the extra ranks on the ballot for that. It's 
possible to have Range/Bucklin/runoff, where much better data about 
candidate preferences is collected.

Ironically, the Oklahoma Bucklin was really a kind of Bucklin/Range 
or Bucklin/Borda method, with weights assigned to ranks. In other 
words, we've had advance voting methods in the U.S. before. We really 
should study what happened, and not with some polemic intent, but 
seeking understanding.

The lower weight assigned to lower preferences would tend to 
encourage more sincere voting, since the potential "Later No Harm" 
impact would be lessened even more. By focusing on IRV, FairVote 
missed the opportunity to find better voting systems; by targeting 
Runoff Voting, FairVote ignored real democratic practicality and 
genuine majority rule, in favor of what, in practice, boils down to 
cost. IRV is supposedly cheaper than real runoff elections.

That, in fact, is probably false. Bucklin would be a lot cheaper than 
real runoff elections, but if we want a majority, the crucial reform 
to protect is runoff voting. FairVote's ultimate goal is really 
proportional representation. It's a noble goal, but, by setting 
themselves a narrow strategic path, they are actually damaging 
democracy in the name of ultimately making it better. That's not 
necessary. Why did they do this? The method used for one of the best 
forms, and certainly the best widely-used form, of proportional 
representation is Single Transferable Vote. But it's complicated and 
expensive to canvass. FairVote knew that this would be an obstacle.

But, aha! They could present the same counting method, used 
single-winner (which makes the worst features of STV into prominent 
defects), and sell it as "instant runoff voting," a totally new name 
for a very old system -- one long considered defective by voting 
systems experts, I've seen the papers back into the nineteenth 
century. This was pure political strategy, and is an example of 
what's wrong with our systems, where political activists who get a 
little funding can promote an idea far beyond its appropriate level.

However, we in the U.S., have a home-grown system, Bucklin Voting. It 
works. It's theoretically better than IRV. It's cheaper to use. 
Unlike the even cheaper Approval Voting (Just Count All the Votes!), 
it preserves and satisfies our desire to vote exclusively and 
sincerely for our favorite, and only shifts from this when a majority 
isn't found (and usually a majority will be found, that's how real 
elections work, unless you get *many* candidates); it doesn't shift 
far. Bucklin was very popular, where used in municipal elections, 
which is where almost all the IRV implementations are going.


>Sources:
>
>Albert P. Brewer, First- and Second-Choice Votes in Alabama, The 
>Alabama Review, A Quarterly Review of Alabama History, April 1993.
>
>Maggie Burgin, The Direct Primary System in Alabama, Masters Thesis, 
>University of Alabama, 1931, page 49.
>
>Charles E. Merriam & Louise Overacker, Primary Elections, 1928, pages 52, 83.
>
>J. Bradley King, Indiana Secretary of State's Office.
>
>Footnotes
>    * The states with years of adoption and repeal (if known) are: 
> Alabama 1915-1931, Idaho 1909-1919, Louisiana 1916-1922, North 
> Dakota 1911-1913, Oklahoma 1925, Washington 1907-1917.
>    * Oklahoma, uniquely, required voters to make three choices. If 
> there was no majority first-choice winner, the second choices on 
> all ballots were added to the first-choice totals at a value of one 
> half vote. If still no majority was achieved, all third-choices 
> were added at one third vote value. If there was still no majority, 
> the plurality candidate was declared nominated. This variation on 
> Bucklin was declared unconstitutional by the Oklahoma supreme court 
> (Dove v. Ogleby) because it required the voter to make second and 
> third choices.
FairVote is basically correct, but it would be more accurate to say 
that Dove v. Ogleby rejected Bucklin because the rules required full 
ranking, or the vote would not be counted. Note that most of the 
theory behind IRV assumes full ranking by voters (hence the claim 
that IRV "finds majorities"), and that the most significant 
application of IRV in the world, Australian elections for the lower 
legislative house, usually requires full ranking. What's remarkable 
is that Oklahoma did *not* find Bucklin, as a method, to be 
unconstitutional, only that additional rule. There was a dissent 
filed that concurred with the mandatory ranking conclusion, but which 
opined that the court should have simply voided that part of the 
legislation; that this was not done would indicate to me that there 
were other forces at work.

In Minnesota, Duluth Bucklin (I call it "instant runoff Approval") 
was found unconstitutional, in Brown v. Smallwood. FairVote has 
attempted to represent this as being due to Later No Harm violation, 
but that is based on one sentence taken out of context; the decision 
is quite clear overall that the reason for rejection was the very 
idea of a voter voting for one candidate being faced with first one 
opponent, then another, and so on; the Court was clearly rejecting 
any kind of alternative vote, approving only of, single-winner, 
voting for one. The Minnesota decision has a brilliant dissent filed, 
noting the utter bankruptcy of the majority arguments, which flew in 
the face of precedent in other states and the common legal opinion of 
the time. But, once again, there were clearly other forces at work 
than pure logic and reason. The Brown v. Smallwood court actually 
presented a cogent argument that Bucklin did not violate one-person, 
one-vote, quoting another court with approval, that what mattered 
wasn't the number of marks on the ballots, but the number of *voters* 
supporting each candidate, but then proceeded to totally ignore the 
argument just made, counting the number of marks on the ballots in 
the Smallwood election, as if this proved something negative (more 
marks than voters!!!). That, to me, when I see this kind of logic in 
a court decision, indicates that the court has an agenda and it is 
seeking rationalizations for it.