[Election-Methods] Fwd: [LWVTopics] IRV Voting

[Abd ul-Rahman Lomax]

I commented, and posted it directly to Ms. Dopp 
and to the list. The list has a 40KB limit, 
beyond which messages require moderation.... I 
presume it will show up, otherwise I'll chop it in two.

At 08:01 AM 5/11/2008, Juho wrote:
>On May 11, 2008, at 1:58 , Kathy Dopp wrote:
>
>>From:  <LWVTopics at yahoogroups.com>
>>Date: Sat, May 10, 2008 at 7:22 AM
>>1a. Re: IRV Voting is a really Bad Idea
>>    Posted by: "Steve Chessin" steve.chessin at alum.mit.edu steve_chessin
>>    Date: Fri May 9, 2008 10:43 pm ((PDT))
>>
>>Kathy doesn't say what voting method she prefers, but it's well-known
>>among electoral experts that there is no such thing as a perfect
>>voting
>>system.

Juho didn't pick up on this. It's a very common 
Fairvote argument, often repeated with about the 
same exact phrasing. Essentially, if someone 
points out how IRV has some lousy characteristic, 
this is trotted out. IRV is actually 
spectacularly bad, but anything they can say to 
deflect the criticism, they will say. And it's, 
simply, and plainly, not true. Arrow's theorem 
does not apply to all voting methods, only to 
strictly ranked methods. Arrow's Theorem points 
out that a certain narrow set of characteristics, 
which seem, at first sight, to be reasonable, are 
mutually contradictory. There is another theorem, 
Gibbard-Satterthwaite if I remember the name 
correctly, which goes a little further, but neither covers Range Voting.

Ranked methods in general make an assumption that 
is blatantly false: that all preferences have the 
same strength. Just to give an example of how 
this distorts thinking, it's very common to 
assume that the Condorcet Criterion is 
practically golden for election methods (even 
though IRV fails it). If there is a candidate 
preferred by a majority over all other 
candidates, that candidate should win. But we 
routinely avoid taking that choice in small 
groups and organizations, when the preference of 
the majority is small and the preference of the 
minority is great (and there is not a large gulf 
in numbers between the majority and minority). 
Thus the ideal winner of an election, in 
situations where all voters will, after the poll, 
agree that this was the best result, is not 
uncommonly other than the Condorcet winner. But 
ranked methods can't collect that information, 
the ballot form does not allow it. One ranked 
method comes close: Borda count, and it does it 
by making a rough approximation: that the 
preference strength between two candidates is 
proportional to the number of candidates ranked 
between those two by the voter. Borda is often 
considered a bad method because this makes it 
vulnerable to strategic nomination, Borda fails 
independence from clones, which is one of the 
criteria considered by Arrow. But Range doesn't, 
because no such assumption is made by Range. If a 
Borda election is slightly modified, so that, 
say, there are as many ranks as candidates (true 
for Borda), but the voters can equal-rank and, 
the corollary, leave ranks blank, Borda becomes 
Range. But the ranking is no longer strict, and 
Arrow's theorem ceases to apply.

>>Yes.  I prefer IRV to all other systems for electing a single
>>winner to
>>an executive office because it elects a majority winner in a single
>>election without the need for a separate expensive runoff election.
>
>Yes.

Again, what I've seen with many students of 
election methods is that they start out with 
this. But the incorporated assumption just stated 
is false. IRV doesn't find a majority winner in a 
single election, and not just rarely. Frequently. 
When an IRV election, in the real world in the 
U.S., where ranks are normally limited to, say, 
two or three, if a majority doesn't appear in the 
first round, it doesn't appear in the later 
rounds. Almost always. So Juho, you just agreed to something false.

That's not an expected result, I was quite 
surprised to find it once I started actually 
looking at the elections. There is a bit of a 
word trick here. By redefining "majority" the 
proponents of IRV have managed to make it *seem* 
that a majority is found, and they report 
results, typically, showing a majority. But it's 
not a majority of votes cast, it's a majority of 
votes left standing after ballots with 
inconvenient votes on them ("exhausted ballots") 
are discarded and not included. You could get 
this result with Plurality. Just use the same 
elimination procedure. (It's *really* easy to do 
so, since *nobody* in Plurality has cast a second 
preference vote!). But we wouldn't consider that 
the Plurality winner was a "majority" winner just 
because that winner had a majority of the votes cast for the top-two.

This is an old debate trick. Instead of stating 
an argument directly, it is stated indirectly, as 
an incorporated assumption, as part of an 
argument about something else. Most people will 
be distracted by the something else, and the 
incorporated assumption just slides in 
practically unnoticed. Does IRV "find a majority 
winner?" What does that mean? It has a meaning 
with all other methods that is *different* from 
what the IRV people make it mean when considering 
IRV. Top-two runoff, the common method that is 
being replaced by IRV, actually does (almost) 
guarantee a winner, by having two elections when 
needed. The winner gains a majority in the second 
election, or not. In theory, in some 
jurisdictions, the electorate, if it does not 
want to accept the top-two, could vote for a 
write-in candidate, so if the preference strength 
is great enough, top-two is Condorcet compliant. 
As a practical reality, we have a radically 
unorganized electorate, so it's quite unlikely to happen.

But IRV can declare a "majority winner" -- and 
even a large majority of votes cast -- when, in 
fact, a *large* majority of voters have voted 
against the IRV winner, as shown on the ballots, 
because IRV hasn't counted all the votes, it has 
eliminated some candidates before finding the 
votes for them over the IRV winner. This is 
important enough, and clearly common enough in 
the relevant situations, that Robert's Rules 
mentions it as a flaw of the system. But FairVote 
is only going to quote what seems approving from 
Robert's Rules, and, I've seen it, struggled 
mightily against including the *rest* of the 
information from Robert's Rules relevant to IRV 
in the article on "Instant-runoff voting" on Wikipedia.

>>   It
>>eliminates the spoiler effect
>
>Yes it does when there are two major candidates and few minor
>candidates. The IRV related problems mainly emerge when there are
>more than two candidates with serious chances of being elected.

Yes. FairVote will allege that the problems of 
IRV are rare and that nobody is pointing to 
actual elections displaying them, but (1) IRV is 
only being used, in the places where it has been 
long used, where there is a strong two-party or 
two-coalition system, as with Australia. IRV 
protects the major parties from the spoiler 
effect, when the minor parties are small, so it 
actually favors them. (STV, with many-member 
districts, is much kinder to minor parties.) and 
(2) Very few elections have been reported in 
sufficient detail to allow a showing of the 
problem. Remember, IRV doesn't even count the 
votes that would be involved. The data is on the 
ballots, but access to that data is difficult or impossible.

However, most of the elections using RCV 
(three-rank IRV) in the U.S. are nonpartisan. 
Still, the extra rounds are turning out to be 
moot. They aren't changing the results over 
first-round plurality. If we were comparing IRV 
with simple plurality, that would be one thing, 
but normally IRV -- in most of the elections -- 
has replaced top-two runoff, that's pretty much 
the only pre-existing condition that can justify 
the huge expense. (Many of the jurisdictions that 
FairVote touts as having "approved" IRV actually 
have not, if approving it includes finding the funding to implement it.)

Top-two runoff seems to reverse the primary 
result about one-third of the time. IRV has yet 
to reverse the first-round result even once, in 
about 23 elections that went to multiple rounds. 
That alone was surprising. But which really 
should have those interested in the performance 
of election methods paying close attention is 
that, in none of these elections did the *runner-up* shift.

Now, there are a couple of possible explanations 
for this. Perhaps the first round votes are more 
sincere with IRV than with top-two runoff, though 
that seems unlikely to be causing such a massive 
effect. It seems to me that top-two sufficiently 
allows most voters to vote sincerely in the first 
round, under most circumstances, such that most 
first round votes would be sincerely preferred, 
even though the voter can vote for only one.

However, suppose it is true that IRV is making 
the right choice in the first round, it's helping 
voters to be sincere. If so, Bucklin would do the 
same job at far lower expense. Bucklin, in the 
Duluth form, had three ranks. The first two ranks 
were like an IRV ballot, and the only difference 
with the third was that the voter could vote for 
as many candidates as desired in third rank. The 
third rank can be used, for example, as an 
"anybody but Joe" rank, by voting for all 
candidates but one. (Which does not harm one's 
favorite, because that vote is moot in every 
other pairwise election except the one with Joe.)

Historically, in some Bucklin elections -- but 
not the ones I have examined in detail -- 
additional rank votes were only used by perhaps 
ten percent of voters. But this simply means that 
voters did vote sincerely in the first rank, 
without constraint. Bucklin has, then the 
desirable characteristics alleged for IRV, 
without the problems, and especially without the 
expense. Approval is even simpler than Bucklin, 
and is essentially free, it only takes a very 
minor change in counting procedure: just count 
all the votes. Stop discarding overvotes. (Which 
with automated equipment, is *more* work. So 
Approval might actually be cheaper than Plurality.)

>>, and it encourages positive
>>issue-oriented campaigns instead of mudslinging opponent-attacking
>>campaigns.
>
>Probably better than plurality here but not as good argument against
>other methods.

Again, Juho swallows the argument because it 
sounds reasonable. It hasn't done it. Show some 
evidence to the contrary, if you want to claim this benefit.

>>   It is also easy to understand and easy to administer.
>
>Basic understanding is easy enough. Inability to count the local
>results locally maybe makes "administering" not as easy as in some
>other methods.

Juho got that one. However, IRV is more 
complicated than meets the eye. Sometimes voting 
sincerely in IRV is a very bad idea. Especially 
when there are many candidates and only three 
ranks. Depends. IRV fails the Favorite Betrayal 
Criterion, which means that sometimes it pays to 
downrank one's favorite in order to improve the result for the voter.

>>[...] I don't think most people would support a system
>>that would declare the winner of a 3-way race the candidate who was
>>the
>>first choice of only 6% of the voters.
>
>This reaction is possible in the US with respect to the presidential
>elections where there is a strong tradition that the winner always
>comes from one of the two major parties. In some other elections (and
>maybe in US presidential elections too) people might also be happy to
>get a candidate that is liked by all instead of getting one of the
>extremists that is not liked by half of the voters.

What if we actually ask the people? Of course, 
we'd have to educate them first. The implications 
aren't necessarily obvious from the get-go. In 
fact, when leading IRV advocates have debated 
directly with critics, in privately-circulating 
mails (seen by small groups of people), they have 
often claimed that "Fine, you've got your 
theoretically superior method, but people aren't 
going to accept it, they aren't sufficiently 
educated, and you can't accomplish that. You 
should join us because we have the political momentum."

Essentially, by coming up with this "IRV is like 
your familiar top-two runoff, but you don't have 
to have that inconvenient and expensive runoff 
election," argument -- which, historically, was 
not at all an argument for IRV (which wasn't 
known by that name, it was called "preference 
voting" or "alternative vote"). Because it's a 
quite spurious argument, IRV differs greatly from 
top-two runoff because with the latter, the 
voters can actually consider the top two afresh, 
whereas often when there are many candidates, 
voters aren't sufficiently familiar with enough 
of them to intelligently rank more than one or 
two. IRV essentially prevents these voters from 
taking part in the decision, on the grounds that 
by not adding the extra ranks, they are 
"indifferent," and therefore it is as if they did not vote in the runoff.

A real top-two runoff only requires that the 
voter research or know about two candidates, not 
many. What does the voter do, presented with 23 
candidates in a San Francisco single-winner RCV 
election? What they used to do was vote for their 
favorite, plain and simple, in the first round. 
And many still do that, you can tell by the large 
numbers of exhausted ballots. Do they not add 
ranked votes because they are indifferent? No, 
because they are ignorant, in a very 
understandable way. They are actually moderately 
well-informed citizens if they are familiar with 
and support one or two of the candidates! To be 
able to rank three starts to become unusually 
well-informed, and, then, even if they do rank 
three, if it's the wrong three, i.e., they 
sincerely voted for their preferred three, their 
vote may easily be disregarded entirely and much of it not even counted.

Sure, voters can easily think IRV is an 
improvement. But they haven't examined what's 
under the hood, and it may take a few years for 
those who know to be able to explain it to them. 
Don't bank on IRV continuing to be used in places 
that have adopted it. Historically, it was used 
in the U.S. in a few different places, such as 
Ann Arbor, Michigan. Once. Revoked by referendum 
immediately. The method has been around for, 
what, 150 years? It was used here. And then not. 
So if a jurisdiction spends a lot of money to 
implement IRV, on the theory that use for a few 
years may pay back the increased expense, they 
just might be setting up yet another expensive 
election boondoggle, more money for the companies 
that sell election equipment, and more trouble auditing the results.


>>Well, given that that list (if it's the one I think it is) is made up
>>primarily of advocates of other systems (Condorcet, Borda, Approval,
>>Range, to name a few), I am not surprised that they agree that "IRV
>>does not work" and that their own pet approach works better.
>
>I think the EM list does not systematically promote some certain
>methods nor attack some others. Individual members do often have
>strong and stable opinions but there is no unified front. There are
>not that many active IRV promoters but I'm sure they are just as
>welcome as promoters of some other methods.

They've been here in the past. Quite simply, they 
can't take the heat. The bankruptcy of the 
arguments like those Chessin is making -- the 
same arguments have been made for years -- 
becomes way too obvious. What we've seen is that 
FairVote activists will pop up in forums around 
the internet, and mailing lists, and make these 
arguments. When it happens that election 
theorists show up and refute the arguments, they 
just stop arguing and move elsewhere and repeat 
the arguments. They know that most people don't 
continue reading past a certain point. They are 
quite willing to repeat defective arguments if 
they know that they sound good at first blush, 
and the major accomplishment of FairVote has been 
to come up with these arguments.

Frankly, though, this style is a major part of 
what is wrong with politics as usual. It comes to 
be about winning instead of about discovering the 
truth and making sound collective decisions. Rob 
Richie, the Executive Director of FairVote, has 
generally treated election methods experts with 
contempt, as "ivory-tower theorists," and implies 
that they are just as biased as him (just as 
Chessin above impugned the Election Methods list, 
which is open and which has no institutional 
bias). In the view of those at the core of 
FairVote, everyone is out for themselves, and 
will essentially deceive to get what they want. 
It's a corrupt point of view, self-reinforcing. 
Most people are *not* like that. Most people 
understand that when collective decisions are 
carefully considered and are broadly accepted, it 
benefits everyone, even if one does not always "win."

>>   But all
>>of them agree that our current election method, plurality elections
>>(complete with the spoiler problem), is the worst method.
>
>Many dislike plurality but I wouldn't say that all consider it to be
>the worst.

Essentially Chessin just makes it up. We've asked 
them who these experts are that think what they 
claim (they claim substantial support for IRV 
from experts), and the answers have been few and 
far between. And sometimes quite deceptive. Such 
as Tideman, who has also deeply criticized IRV 
and has gone so far as to call it 
"insupportable." But because he made a few 
positive statements about it, you can guess what 
they quote. (And Tideman also made some serious 
errors in certain of his analyses, but that's another story.)

>>The vast majority of informed opinion (not just that of the
>>IRV-dislikers) is that IRV is the best option for single-seat
>>elections.

I'm just going to say that he's lying. And the 
APSA claim I went into in detail. From what I've 
been able to find, APSA has never used IRV, they 
use something else: a nominating committee (by 
majority vote, but deliberatively, which happens 
to be Condorcet-compliant) nominates a 
President-election, who is then formally elected 
at the annual meeting. I found no evidence that 
the nomination was ever opposed. So it would be, 
at best, unusual that there was any contending 
candidate. And for IRV to be used there would 
have to be two contending candidates, i.e., three 
total, or more. It wouldn't be surprising to find 
that this has never occurred in the almost 
century that APSA has been around, and it would 
also not be surprising to find that the provision 
for it is that old. Preference voting was all the 
rage when APSA was founded, and they simply chose 
the sequential elimination form; in those days 
political scientists didn't have much knowledge 
of the various alternatives, and that remains 
true for most. It's not taught, particularly, and 
the field has, in recent decades, rapidly 
expanded as economists and mathematicians became 
interested (and some of the  interest in voting 
methods always came from mathematicians and the 
like, such as Dodgson.) Range Voting falls out of 
economics theory, game theory, i.e., advanced 
decision-making theory. Even though it's quite 
easy to understand, actually. (When the votes 
reflect real-world, verifiable quantities, such 
as known economic benefits of various 
alternatives to the voters, it's obvious that 
Range Voting is optimal. It simply finds the 
highest sum of benefits, the "greatest total 
good." And then this good is broadly distributed 
because there is not just one election, and the 
"majority" and "minority" are not some fixed group of people.

Political scientists? Few have shown much 
interest in election methods, which I've found 
surprising. The expansion in knowledge of how 
election methods perform, and about the 
theoretical alternatives, is coming from 
mathematics, economics, and from amateurs who follow an interest in the field.

So the *real* method used by APSA? Deliberative 
process. If the nominating committee is broadly 
representative (I think it is), what we *really* 
have is a kind of parliamentary election, except 
for a fixed term.) It's a deliberative method, 
and it's likely to choose the best candidate from 
the outset, without any mail ballot at all. Do 
you know that we could easily do the like with 
public elections? It's what Dodgson (Lewis 
Carroll., yes, of *that* Alice) proposed in 1886. 
Very simple. Very powerful. Very accurate and 
full proportional representation, and the 
"electoral college" created could then be used to 
pick a single winner deliberatively.

>>   For example, the organization of political science
>>professors, the American Political Science Association, has
>>incorporated IRV into their constitution to elect their own national
>>president.  While the members of Kathy's list [Note: I never said that
>>this was "my" list] have their pet methods,
>>none of those methods are actually used for governmental elections
>>anywhere in the world.  In contrast, IRV is used in the United States,
>>and has been used for generations in places like Australia and
>>Ireland.  Also, none of the pathological scenarios the IRV-dislikers
>>like to put forth have ever surfaced as a problem.

See, there is a list of these bullet points. He 
just crammed a whole series of false and/or 
misleading statements into one short paragraph. 
Someone who responds to him will normally, if 
being civil, respond with far more words, but 
Chessin et al know that the lies and deceptions, 
many of them, will stick. After all, people will 
think, he wouldn't dare lie, when his claims could be checked.

But: APSA, very, very misleading. No actual elections.

(Why didn't he say, "APSA uses IRV." Well, he 
knows they don't use it! He would not have said 
it the way he said it when he could have said it 
much simpler if he didn't know. This propaganda was designed by a spin-master.)

"Kathy's list." Not the most notable mailing list 
in the world for discussion of election methods. 
Open. Associated with a wiki that IRV people are fully welcome to edit.

None of these methods are actually used.... not 
true. Borda is used. Other methods *have* been 
used, some of them for a very long time.

"Places like Australia and Ireland," is a device 
for making two sound like more than two. Also, 
what are really different methods are lumped 
together as "IRV" by FairVote activists. For 
example, they will tell you that the Mayor of 
London is elected by "IRV," but it is what is 
known as the Sri Lankan Contingent Vote. Batch 
elimination, two rounds only. Closer to actual 
top-two runoff, which is more accurately simulates.

"None of the scenarios ... have ever surfaced." 
Again, notice the language, very carefully 
crafted. Why not "haven't happened"? Well, 
because he knows that we don't know that, for 
reasons I've given above. However, if we look at 
top-two runoff, there have been quite a few 
elections showing center squeeze, and IRV, in the 
first round, behaves similarly (and especially 
Sri Lankan Contingent Vote does, precisely the 
same). In France, for example, and in many other 
places. The candidate who was almost certainly 
the most popular, by comparison with the winner, 
was passed over because of being third in first 
preference votes. So, again, the way he makes the 
statement, it is sort-of true, but it is also clearly deceptive.

And I call being deliberately deception by making 
true statements "lying." Lying is about intent to 
deceive, and the best I can come up with in 
thinking about Chessin, Richie, and the like, is 
that they display reckless disregard for the 
truth. Once can easily pass off a few statements 
like what Chessin has made here as being simple 
disagreements or errors in expression. But when 
it's consistent over the years, as it has been, I 
have to personally come to a different 
conclusion. These are skilled political 
activists, they really don't care about the 
truth, they care about winning. It's my hope that 
their day is past, and I think I know how to ensure that.

It's not about which voting method is best. 
Voting methods aren't really the problem. The 
real problem is a heuristic one: how do large 
numbers of people make collective decisions as to 
how to communicate, coordinate, and cooperate? 
Voting methods are practically a detail, comparatively.

>I think two-round runoff is more commonly used than IRV.

Right. Top-two runoff is *very* widely used. And 
I claim, in fact, that top-two runoff is better 
than IRV, and it's actually easy to prove from 
the actual elections, reasonably well. Is it more 
expensive? Probably. But that's easily fixed with 
a *real* instant runoff system that counts all 
the votes, and that would make runoff elections 
rare. In our present political system, it is 
crucial that we find the best winners in certain 
elections, and if it takes a runoff and extra 
inconvenience and expense, it is well worth it. 
We are spending trillions of dollars in Iraq, so 
what would I now think about the expense of, say, a runoff in 2000?

>>It is theoretically possible to have an election in which NO candidate
>>has majority support, like a rock, paper, scissors endless loop,
>>but IRV
>>is FAR more likely to elect a true majority choice than our current
>>plurality election method with its rampant spoiler problem.
>
>IRV would help in this one problem. So would some other methods.

Again, Juho isn't being sufficiently strong, 
probably because he's unaware of the actual 
performance of IRV> It is not finding true 
majority choices. Notice the missing object of comparison.

IRV is *probably* far more likely to elect a 
majority choice than simple plurality, when there 
is a spoiler, which is relatively rare (but still 
significant, to be sure). But IRV can fail 
miserably, as noted, if any of those third 
parties actually are so uppity as to become 
realistic possibilities for winning. Warren Smith 
has called IRV "political suicide" for third 
parties, which have sometimes naively supported 
it. That's beginning to change, as, for example, 
some Libertarian leaders have come out, now, 
against IRV and for Approval and Range Voting. 
I'm pretty sure they'd like Bucklin, as well.

>IRV elects the majority winner when comparing the
>_two_last_remaining_candidates_.

Right. The FairVote activists claim that this 
simulates top-two runoff, except that with real 
top-two runoff there is a real election, with a 
real campaign, and thus an opportunity for voters 
to get to know the top two much better. This is 
the most obvious reason for reversal of the first 
round in one third of top-two runoffs, which 
isn't even coming close to happening with IRV. 
(If it were coming close to happening, we'd 
expect to see the runner up changing, as well. 
The runner-up isn't changing. The leader and the 
runner up is the same before and after all the complicated vote eliminations.

>  It may fail e.g. to elect the
>Condorcet winner, a candidate that wins all others by majority in
>pairwise comparisons. Instead of the 6% example discussed above one
>could also look at the following example where electing the centrist
>C (with 30% first place support and 100% first or second place
>support) instead of A and B makes more sense.
>
>35% A>C>B
>15% C>A>B
>15% C>B>A
>35% B>C>A

I gave a slightly simpler example. Same idea. 
Chessin wants people to believe that, okay, maybe 
there is this odd situation, but really all it 
takes is an election with three frontrunners, 
three in range of winning. First preference, 
above, they are roughly tied. That's the 
dangerous situation, where the elimination 
becomes rather arbitrary. There are other forms 
of preferential voting which consider all the 
votes, these are much better. Simplest is probably Bucklin.

>With IRV it may also happen that there are e.g. two right wing
>candidates in the race and the method eliminates first the one that
>would have also considerable left wing support, and this would leave
>the right wing at the last rounds with a candidate that left wing
>(and centrists) definitely do not want to support. Left wing could
>then win (to the disappointment of the right wing), or the less
>popular right wing candidate would win (to the disappointment of the
>left wing).

Complicated explanation. But, yes. It's called 
Center Squeeze. And it is what Robert's Rules 
talks about as a defect. FairVote propaganda 
about Robert's Rules is all over the internet, 
and I've seen it quoted on governmental web 
sites. They have very carefully designed this 
stuff to influence the masses and to influence 
the influential. And quite deceptively. That's 
what I object to, even more than the method they 
are promoting. The original goal of the Center 
for Proportional Representation was good, and 
supported by most election theorists. But the 
movement was quickly taken over by a core of 
activists, at least one of them making it his 
profession, and they were not about to allow 
democratic process to confuse their strategies. 
CPR became the Center for Voting and Democracy 
and then FairVote. They don't actually believe in 
democracy, they believe in winning.


>One reason why Condorcet promoters are less active is that there are
>many different Condorcet methods and the proponents of those methods
>each promote their own variant of the method. This means that they
>will also fight against each others and do not promote the Condorcet
>methods in general.

What I've seen, though, in every forum where the 
question has been asked, the experts have agreed 
that Approval makes very good sense as a simple 
reform; quite a few have, of late, as well, come 
to understand Range Voting and to recognize it as 
theoretically ideal. There are still widespread 
concerns about alleged tactical voting, but this 
is a holdover, in my considered opinion, from the 
thinking underlying ranked voting theory, which 
was defective. Approval Voting is a bit misnamed. 
It's Voting. And just count all the votes. Votes 
are not sentiments, and by treating them as such, 
and by assuming that there is some absolute 
trait, and absolute relationship between a voter 
and a candidate called "Approval," it becomes 
possible to treat a vote where the voter votes 
for his or her favorite alone as "dishonest." 
Same arguments arise with Range Voting, it is 
simply easier to understand with Approval. In 
fact, Approval and Range *never* reward reversing 
preference, which is not true for any ranked 
method of which I'm aware. But by creating a new 
kind of "dishonesty" -- voting equal preference 
when one has an actual preference (in some 
considerations by critics) or voting preference 
when one allegedly "approves" of both -- critics 
worry about the "vulnerability" of Range Voting 
to strategic voting. But the methods they would 
propose, instead, *force* strategic voting, under some conditions.

Reminds me of episiotomies. I was a midwife, for 
a time, and we, working at home, of course, never 
did episiotomies (making an incision in the 
perineum to open up the birth canal). We did take 
measures to encourage stretching. Occasionally 
there was some tearing. So what do the doctors 
do? They routinely do the incision. Since it 
might tear, cut it. Wait a minute! How often does 
it tear? (Usually not, or not much.) If it tears, 
is there a problem with healing? (No, the rough 
tear can be sutured together and heals better 
than the clean cut). Nice article on it on 
Wikipedia. Apparently, since my time in the 
field, the medical profession began to realize 
that they were doing more harm than good, and episiotomies are becoming rare.

In any case there are potential forms of Approval 
or Range Voting that are Condorcet compliant. (It 
always involves a runoff triggered by the 
presence of a Condorcet winner, which can 
generally be detected with Range ballots, and 
it's easy to make Approval ballots usable for 
that (Range ballots with at least as many ratings 
as candidates, preferably more, can be used for 
this with complete accuracy, under some 
definitions of the Condorcet Criterion.)

>  Some of the methods are also clearly more complex
>than IRV, but some are also very simple and easily understandable.
>The vulnerability of different Condorcet methods to strategic voting
>is slightly different, which often makes the discussion very strategy
>oriented, and makes Condorcet methods look like being plagued by
>numerous strategy problems (more than they actually are).

The problem with the whole discussion of 
strategic voting is that it's often backwards. 
Voters use strategic voting to improve the 
outcome. Some Condorcet methods make strategic 
voting difficult. But that's making it difficult 
for voters to improve the outcome!

What is really lost is that ranked methods don't 
collect the information necessary to determine 
the best winner, because the assumptions about 
pairwise victories are defective. Preference 
strength information is crucial to make a 
judgement of "best winner," and the information 
is simply missing from a pure ranked ballot. Now, 
when Condorcet methods allow equal ranking, they 
get better, and if the ballot has a fixed number 
of ranks and voters can place the candidates 
without restriction in each rank, and can leave 
ranks blank (thus indicating a stronger 
preference strength between two candidates than 
if there is no blank rank intervening), the 
information is there. And that is Borda, modified to make it into Range Voting.

If you then do pairwise analysis and discover 
that there is a candidate who beats the Range 
winner, you hold a runoff, you have made Range 
Condorcet compliant by allowing the contingency 
of a runoff. From simulations, it should be rare. 
Range *usually* chooses the Condorcet winner.

Why would you hold an actual runoff? After all, 
many election theorists would quickly opine, the 
voters have already voted for the Condorcet 
winner over the Range winner. Yes, they have. But 
they haven't made that explicit choice, and, if 
they have voted sincerely, the Range winner is, 
in fact, the best winner. So we test it with a 
runoff. Is the Range winner really the best? 
Whenever there is a candidate who beats the Range 
winner, that victory is, by definition, weak, 
compared to the preference strength involved with 
the Range votes that made someone other than the 
Condorcet winner be the Range Winner.

Weak preference discourages turning out for a 
runoff election. Further, some people who prefer 
A over B, weakly, will change their preference if 
they realize that overall satisfaction will be 
higher if B is chosen. People do this routinely 
with ordinary choices, they give up their first 
preference to make a bigger difference with 
others, and the fact that most people do this 
most of the time makes society run much more 
smoothly and much more for common benefit, which 
helps everyone. What goes around comes around.

So I expect that if the votes in the first round 
were sincere, contrary to the expectations of 
some election theorists (including some Range 
supporters, by the way, who don't like the runoff 
because they fear it will elect the Condorcet 
winner), the Range winner is most likely to prevail in the runoff.

But there is another possibility. The Range Votes 
have typically been distorted by two factors: 
first, normalization, where voters consider the 
set of realistic winners and make their 
preference range for those candidates conform to 
the span of one full vote. So the *actual* 
underlying preference strength expressed by a 
particular vote is not proportional true absolute 
expected satisfaction, comparable among the 
voters. It's just a kind of approximation. Range 
Voting does not produce perfect results, it can 
fail to find, in simulations, always, the perfect 
winner, i.e., the winner who truly maximizes the 
assumed absolute satisfaction level. (This kind 
of work can only be done in simulation, because 
the simulator *starts* with assume absolute 
satisfaction, which is then converted into 
relative satisfaction when the votes are 
translated to the ballot. It doesn't matter what 
absolute satisfaction means, specifically; the 
trick is getting a realistic distribution of 
such; and there is a lot of work remaining to be 
done with testing the predictions of the 
simulators with actual voting behavior. But it is 
the best work we have, so far, toward objectively 
analyzing election performance. And there are 
real situations where absolute utilities are 
known. It's just that we have no way, at least 
not yet, of directly measuring this with 
abstractions like satisfaction with an election result.)

Then, as well, voters take into consideration 
election probabilities. They don't want to waste 
a portion of their vote, assigning preference 
strength to a pairwise election that is not a 
real contest. So, again, the Range Votes deviate 
from actual preference strengths. Necessarily so.

But a runoff tests real, absolute preference, 
roughly. This is why, I expect, top-two runoff 
does much better than one might think. And it can 
be greatly improved by better methods of choosing 
the top two, if it is to be cut down to two. 
(Remember, basic democratic election process 
doesn't use eliminations at all, it uses repeated 
balloting, and never elects without a majority. 
That's standard Robert's Rules.) So, I've 
suggested, the top two should be the Range winner 
and the Condorcet winner, if they differ. Thus, 
with that simple change, Range becomes Condorcet 
compliant, satisfying the single most significant 
objection made by theorists against it, that it 
can elect a candidate who would be beaten by 
another. Yes it can, though not in such an 
offensive way as IRV can. But we can fix it.

(Besides, it's rare, and we might find from 
experience that Range results are never reversed, 
so we might, in the end, decide to get rid of the 
runoffs. What ever happened to the idea of 
*trying* something? At least, folks, we should 
start using Range ballots to collect the 
information. Even if we still awarded the victory 
to the "plurality" winner. (Though if it is a 
Range ballot we should, at least, allow multiple votes, i.e., Approval.)

>It would be interesting to know if the IRV promoters find also
>Condorcet to be a good method and worth promoting.

No. They don't. They criticize Condorcet. From 
http://www.fairvote.org/?page=1920
(where FairVote examines, with their usual 
deceptive arguments, other election methods.)

>Bottom-line: Condorcet-type voting violates the 
>principle of requiring a minimum level of core 
>support by permitting a candidate to win who 
>would not win a single vote in a plurality election.

In other words, the Core Support Criterion -- 
which FairVote basically made up, the criterion 
has no support elsewhere, and it really doesn't 
make sense -- has become a basic principle. 
That's turning a sow's ear into a silk purse, 
alright! Here is their justification:


>Problem 1: With these rules, a candidate can win 
>without being a single voter’s first choice. By 
>putting such heavy emphasis on breadth of 
>support, Condorcet-type systems, like approval 
>voting, encourage candidates to be seen as the 
>“least offensive” candidate rather than leaders 
>who take strong positions that might alienate some voters.

Apparently, even the candidate's mother doesn't 
vote for her? I guess that *would* be bad! Or 
maybe she died, or couldn't make it to the polls. 
The candidate herself doesn't vote for herself. 
Definitely milquetoast. Look, if such a person 
can actually win, it would be an amazing thing, 
it's so extraordinarily unlikely. Notice the 
talking ot of both sides of the mouth. An alleged 
feature of IRV is reduction of negative 
campaigning, as candidates will allegedly seek to 
gain additional ranked votes from supporters of other candidates.

Sure. But that doesn't apply to frontrunners, who 
stand, generally, *not* to gain such votes. How 
many Gore voters were going to vote for Bush? And 
if it's a good thing with IRV, why is it a bad 
thing with a Condorcet method? There is no 
evidence at all, from real elections, of the 
effect that Richie claims on the FairVote site. 
The truth is that single-winner leads to 
polarization, where candidates seek to become the 
favorite by being noticed for some striking 
position. Or by knocking down their perceived 
opponent. If we take proportional representation 
to an extreme, this kind of motivation entirely 
disappears. In fact, elections, really, could 
cease to become contests, but merely be a process 
by which the people choose their representatives, 
with no losers, with compromise being necessary 
only in certain respects due to the necessity to 
limit access to the floor of an assembly, or else 
it will become impossible to transact business.

Besides, which would you have as President, Mr. 
Milquetoast or Adolf Hitler? The strong leader? 
Yes, there are people who think like that! I'm 
surprised, though, to see the thinking from a 
group that used to call itself the Center for 
Voting and Democracy. "Strong leader" is actually 
one of the greatest hazards democracies face, for 
strong leaders can sometimes win elections based 
on their "core support" being the largest group. 
And thus the largest private army, whether that 
army fights literally with crowds and guns, or 
less disruptively with money and corrupt influence.

>Problem 2: Condorcet comparisons can yield a 
>situation where, in an election among Candidates 
>A, B and C, Candidate A is preferred to B, B 
>preferred to C, and C preferred to A. In this 
>situation, there is no winner, and a “fallback” 
>method must break the cycle. When this fallback 
>is needed, sincere voters can be punished. 
>Finally, Condorcet-type rules are difficult to 
>count by hand in big elections. Hand-counting is 
>important if problems emerge with voting machines or software.

Unmentioned in this, of course, the severe 
difficulties that can be encountered in close 
elections with IRV and hand-counting. It's done, 
to be sure. But it can be quite a mess. In any 
case, hand counting would be used to enter totals 
into a spreadsheet containing the Condorcet 
matrix. It's not difficult, and it does not 
require "voting machines or software." The 
spreadsheet data is then transmitted.

IRV requires either central counting of many, 
many ballots -- which is pretty difficult -- or 
local counting in rounds. That's the way it's 
done, I understand, in Australia. Each precinct 
counts its local first preference totals. It 
transmits those totals to a central office. The 
central office collects all those totals and adds 
them up. It then sends back the loser, the 
candidate to be eliminated, and these votes are 
distributed to the remaining candidates, and the 
totals sent in again. And the process is repeated 
until the central office declares a winner. In 
San Francisco, there were something like twenty 
rounds involved in one election. IRV provides 
many, many opportunities for ties, and tie 
resolution methods are problematic, but the real 
problem is that when an election is very close, a 
single vote error is much more likely to affect 
the result. Whenever hand counts are done, in my 
experience, vote totals from one count don't 
match the next, there are almost always errors, 
even with only a few hundred votes (surprisingly 
many, there are better ways of counting that 
obviously aren't being used; the counting method 
used here minimizes handling of the votes and 
maximizes the ability of observers to supervise 
each action, which can make the whole thing 
totally tedioius. I've argued that counters 
shouldn't *touch* the actual ballots, that they 
should be counting *copies* (physical or on 
computers) of the ballots. (Which is what 
actually happens with voting machines, by the 
way). And then counting could take place in the 
most efficient way, and counting could be done by 
multiple independent teams, so it's redundant 
from the get-go, and errors would practically disappear.)

The point: an error in a precinct could change 
the winner of a round and thus all subsequent 
transmissions. If such an error is found, all the 
later rounds of counting have to be repeated, 
because they now involve different candidates. 
And thus one does not accumulate a set of 
validated, identical totals, so the error rate 
does not fall with repeated counting, to that 
degree. It is no wonder that San Francisco wasn't 
able to release accurate results for *months* after November, 2007.

FairVote has made all these arguments in public 
places, so they know the defects in the 
arguments, or at least the counter-arguments. But 
FairVote isn't about examining the issues 
neutrally. It is an advocacy organization that 
has it's mind made up, and it is in the 
convincing others stage, and has been for years. 
They have, as long as I've known about them, been 
quite uninterested in cooperating with supporters 
of other methods. They've been approached, I've 
been cc'd on some of the emails, and they responded with contempt and insult.

>  Maybe their
>preference of IRV is partially based on the fact that it seems to be
>achievable right now. If the basic need is to change plurality to
>something that eliminates the current (small party) spoiler problem I
>think both IRV and Condorcet will do the job.

And so would simply counting all the votes. 
Condorcet does count all the votes. But 
apparently, the simple count-all-the-votes and 
add them up is actually a better performing 
method than Condorcet. What I'd really like to 
see is the use of Range ballots, with full ballot 
data being released (copies of individual 
ballots   -- did you know that, for example, in 
Florida, ballots are public record and anyone 
willing to pay the cost of having a guard stand 
there can photograph them? What if multiple 
independent election observers had digital 
cameras and could photograph the ballots 
immediately when the boxes are opened? *It would 
become impossible to manipulate the election 
through control of the counting process*! No 
reliance on voting machines, reliance only on 
images of the primary voting records. And Range 
ballots can be analyzed many different ways, 
including Condorcet, sequential elimination, Approval, etc.

But for a first step, folks, let's realize it: 
the obvious first step is to count all the votes. 
Approval Voting. Should have been done a long 
time ago; certain knee-jerk arguments against it 
(what about "one person, one vote"?) turn out to 
be based on an easy and false impression. (IRV 
allows multiple votes, with only one being 
effective at a time, which is why IRV supporters 
consider it one-person, one-vote; but with 
Approval, more than one vote can be active during 
the process, but, in the end, only one vote 
actually counts and all the others from a voter's 
ballot could be eliminated without any effect on 
the result. So the voter has only cast one 
*effective* vote, which is the simplest 
explanation I know of why Approval still 
satisfies one person, one vote. Basically, the 
extra votes are still alternative votes, never 
can both end up being used. It's just that they 
are equally alternative, not sequenced in priority.)

Robert's Rules instructs the clerk to disregard 
ballots with more than one vote. Why? Because the 
clerk cannot then discern the true intention of 
the voter. But this isn't a reason to disallow 
it, this is an instruction given with an 
assumption that it is disallowed, and therefore 
must be a mistake. Ballots generally don't say, 
"Vote for your favorite," they say, "Vote for 
one." If we assume that voters are voting for 
their favorite (an assumption that actually holds 
true in most society elections that hold repeated 
balloting, which is what Robert's Rules is 
talking about), then, yes, the clerk cannot 
discern who is the favorite. But once we have 
elimination and single-ballot elections come hell 
or high water, "vote for your favorite" has 
already gone out the window. And thus alternative 
votes make sense, and the easiest way to allow 
them is to stop disregarding them. Just count 
them. A voter can decide, for example to vote 
*against* a candidate, by voting for everyone 
else preferred to that candidate. That can be 
quite sincere, but, of course, it abstains from 
the rest of the election. It makes sense if the voter really feels that way.

But, most obviously, supporters of minor 
candidates in public elections will often choose 
to also vote for a frontrunner, thus, quite 
likely, ameliorating the spoiler effect. And 
ranked Approval (Bucklin) should completely solve 
it, for there is little reason for a third party 
supporter, now able to express exclusive 
preference his favorite, not to add a lower 
preference vote for a realistic winner.

Yes, Bucklin does not satisfy later-no-harm. Your 
second preference vote can hurt your first 
preference. But if your first preference and 
second preference are frontrunners, you aren't 
going to add that second preference. Unless there 
are *three* frontrunners where the whole thing becomes more commplicated.

Bucklin, though, could not elect the nightmare 
scenario we'd described with IRV, where an 
extremist candidate wins even though two-thirds 
of the voters opposed him in favor of another 
candidate, because of low "core support" -- first 
preference support -- for the compromise winner. 
That's because Bucklin counts all the votes. Same 
as Condorcet, just in a different way. These 
methods *usually* elect the Condorcet winner when 
one exists, and usually a member of the cycle 
when a unique one does not exist, and when they 
don't -- particularly with Approval or Range -- 
it can be argued that the Condorcet winner was 
not the optimal winner, the one most beneficial for society overall.

Oh. I forgot. According to these champions of 
Voting and Democracy, the best winner is the 
"strong leader." The one not afraid to offend, 
perhaps to tell the enemies of society, those 
disruptive and unpatriotic denizens of the other 
side, how terrible they are, and to take strong 
action against these enemies of all that is good 
and true about our great nation.

>All methods do have strategy problems and careful analysis is needed
>(especially since strategies are different in different methods).

Actually, the only 'strategy problems' faced by 
Range methods (including Approval) are with 
strategy as it was redefined to create an 
impression of vulnerability, once the proponents 
of ranked methods, and especially of instant 
runoff voting, realized how bad it looked that 
Approval -- the original target of this -- was strategy free.

Approval was *designed* to be strategy-free. So 
the spinmasters redefined strategy, considering 
and labelling a vote, for example, exclusively 
for the voter's favorite as "dishonest." An 
"honest vote," they claim, would have approved 
some other candidate. On what basis?

We don't need no stinking basis, we assume it. We 
assume that the voter approves of two. But then 
selfishly votes for his favorite.

When one looks closely at what this means, it 
vanishes like mist in the morning. "approve" is a 
relative term, it is not a quality of the 
candidate in the voter's eyes unless the context 
is known, what the realistic choices are. The 
economists know that, to make intelligent, 
optimized choices, one needs to know, before 
investing in a particular option, what the 
probability is of success. Approval allows one to 
express a vote only between two sets of 
candidates, so if a candidate pair is moot, one 
does not want to waste the vote expressing that 
pair unless it can be done without cost. One 
must, with Approval, vote by setting the approval 
cutoff between the frontrunners, if there are 
only two, and somewhere in the middle of the set 
of frontrunners, in the rare instance that there 
are more than two. Only if one doesn't know the 
frontrunners -- a rare voter -- would one vote 
according to some vague notion of "approval."

Really. The whole question of the vulnerability 
of Range Voting to strategy is quite problematic, 
based on fuzzy definitions created specifically 
to try to make Range appear vulnerable.

And then a *different* definition of sincere vote 
is used to determine if Approval passes the 
Majority Criterion. And that's another story by 
itself. Just remember this: a "strategic vote" in 
Range simply means a vote where the voter does 
not reverse preference, but decides how strong a 
vote to express in various candidate pairs, most 
intelligently. And, in fact, I strongly advise 
all voters to do this in the most effective way. 
We *want* the voters to tell us what matters to 
them, and how strongly it matters, and there is 
no such thing as exaggerating for personal gain, 
they are already expressing, we hope, what 
maximizes their personal gain, and if everyone 
does that intelligently, we get a very good result, definitely not a bad one.

What complicates it is that if, somehow, everyone 
could vote accurate, absolute utilities, we would 
then see that some voters could improve their 
personal outcome, at general overall expense (but 
slight expense), by exaggerating from the 
absolute utilities. But that's not a realistic 
scenario, there is no way to even know what the 
absolute utilities are. As a voter, I only know 
my personal utilities, not absolute utilities. 
Absolute utilities for analysis of elections are 
a theoretical construct that is assumed and then 
used to generate realistic internal utilities and 
from them, realistic voting patterns according to 
various strategies employed in various methods. 
That all this is true is then used to make it 
seem that voters can exaggerate for personal 
gain. Not in a real system. Only under two 
conditions: people know absolute utilities and 
are voting them -- something that is practically 
impossible -- and very good knowledge of how 
other voters will vote. To the extent that other 
voter's votes are unclear, truly exaggerated 
voting can be a serious mistake. They system might give you what you asked for.

Bullet vote for your favorite, A, when you prefer 
A>B>>C, you might find that the system, if the 
true choice turns out to be between B and C, 
takes you at your vote that you don't care.

On the other hand, if the true race is between A 
and B, it makes sense to vote full preference for 
A. That's not dishonest, it is simply making a 
choice between the realistic options. This is 
what they call "strategic voting." It's a voting 
strategy, yes. A simple one, easy to understand, 
and it is how most people will vote. So what's the problem?


>  IRV
>would have done fine e.g. in the recent US presidential elections
>with two major candidates and some small "spoilers". When the
>political environment changes and there will be numerous viable
>candidates, then Condorcet maybe performs better. If IRV will face
>these problems (and people will recognize them as problems instead of
>just as "some interesting randomness in the competition") then
>further transition to Condorcet is an option.
>
>In USA it would make most sense to join forces and promote transition
>from plurality to any of the better methods (e.g. Condorcet, IRV,
>maybe also some others). (That is, if large part of the US citizens
>feel that plurality and the pure two-party system idea that it builds
>on should be changed to something else.)
>
>Juho
>
>
>>----
>>Election-Methods mailing list - see http://electorama.com/em for
>>list info
>
>
>
>___________________________________________________________ 
>Try the all-new Yahoo! Mail. "The New Version is 
>radically easier to use" ­ The Wall Street 
>Journal http://uk.docs.yahoo.com/nowyoucan.html
>
>
>
>----
>Election-Methods mailing list - see http://electorama.com/em for list info