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

[Abd ul-Rahman Lomax]

At 06:58 PM 5/10/2008, Kathy Dopp wrote:
>Friends,
>
>I need help to rebut these points about IRV voting that are being
>spread via a US League of Women Voters' email list  that is tracked by
>persons in the US office of the LWV (and apparently the LWV of Mass.
>has already officially taken a position in favor of IRV voting).

Steve Chessin is, of course, a strong FairVote activist, and he is 
presenting the familiar package of arguments.

>I added a few comments in []s.  Please help me because I do not have
>information on this topic at my fingertips and do not have time
>currently to research a reply.  Thank you.
>
>
>---------- Forwarded message ----------
>From:  <LWVTopics at yahoogroups.com>
>Date: Sat, May 10, 2008 at 7:22 AM
>
>
>There are 3 messages in this issue.
>
>Topics in this digest:
>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.  The mathematical proof of this is known as Arrow's Impossibility
>Theorem, after Stanford economist Kenneth Arrow, who received the Nobel
>Prize in 1972 for proving his theorem in 1951.  Pathological scenarios
>can be constructed for all voting systems, including whatever ones Kathy
>may prefer to IRV.

This is a common theme in FairVote propaganda. They know that IRV is 
seriously defective. Consider what Robert's Rules of Order Newly 
Revised says about it. After noting that, for mail elections, 
"preferential voting" is better than electing by a plurality, it 
gives an example of preferential voting that is the sequential 
elimination method of IRV. However, in discussion, it's made clear 
that the default for this still retains a majority requirement, 
unlike every implementation of IRV so far in the U.S. (IRV is touted 
as "finding majority winners," but, in fact, in almost every IRV 
election that didn't find a majority in the first round -- i.e, first 
preference majority -- the winner ended up getting less than a 
majority of votes cast. The so-called "majority" that IRV finds is 
discovered by throwing out the votes that were not for one of the top 
two, and we could get that result with plurality. Just throw out the 
minor candidate votes, and, presto! -- a majority.)

And then Robert's Rules goes on to criticize the method it has just 
explained (not necessarily "preferential voting" in general). It 
notes that the method can "fail to find a compromise winner." [I'm 
writing this from memory, I could get exact quotes but don't have 
time at the moment]. This is a reference to "Center Squeeze," where a 
candidate who might be preferred by a two-thirds majority to the IRV 
winner nevertheless loses because that candidate doesn't get enough 
first-preference votes and so is eliminated before the 
second-preference votes are discovered. For this reason, some think 
that IRV violates one-person, one-vote, because not all votes are 
treated equally.

Now, about Arrow's theorem. FairVote likes to quote this because they 
can then say, well, IRV has flaws, but no method is perfect. Thus 
glossing over the fact that nearly all election method theorists 
prefer other methods than IRV. For example, the vast majority would 
recommend, as a first, very simple reform with no cost -- and most 
voters can continue to vote exactly as before -- Approval Voting, or 
what I call Count All the Votes. It's that simple, just stop 
discarding overvotes. In fact, if you do this with IRV, it becomes a 
better method!

Is it true that Arrow's theorem proves what they claim? No. It's 
false, and if Mr. Chessin doesn't know it, it's not because he's not 
been informed. Arrow's theorem is a narrow proof that applies to 
extracting a complete "social order" from individual rankings. The 
rankings are exclusive, and Arrow's theorem thus applies only to 
ranked methods; Approval Voting and Range Voting are not "voting 
methods" as defined by Arrow's theorem. This is an attempt to call 
down prestigious authority to resolve a question that isn't answered 
by that authority. Range Voting (and Approval is the simplest form of 
Range Voting) "violates" Arrow's Theorem, in the sense that it 
satisfies the election criteria that Arrow considered, but it also 
"violates" the initial definitions, so it's more accurate that 
Arrow's theorem does not apply. Yet it's a "voting system," so this 
makes mincemeat of Chessin's claim.

Now, Chessin will surely propose a "pathological scenaro" for Range 
Voting, or the more immediate and practical Approval. Let's do 
Approval: he would say, for example, that it is possible that a 
majority prefer a candidate, but somebody else wins. That's true. If 
a majority prefers a candidate, but enough of that majority also 
decides to vote for another as well, the majority preference can fail 
to win. In this case, the majority has effectively *consented* to this!

And then, consider how likely this is, under anything like present 
conditions. In most elections, there are two frontrunners, with other 
candidates quite unlikely to win. In order for a majority preference 
not to be elected, there must be more than one candidate with a 
majority. Which requires that the supporters of one major party 
candidate also approve the other major party candidate. Not very 
likely. Gore voters also voting for Bush. Instead, what happens with 
Approval is that Nader voters, some of them, vote also for Gore. 
Approval generally can be expected to fix the spoiler effect, without 
any fuss or complex ballot or nasty surprises.

The nasty surprise with IRV is also relatively rare. It happens if a 
third party rises in significance to the point where it could win. If 
there are three parties at rough parity, then we can see Center 
Squeeze, as mentioned in Robert's Rules. Center Squeeze is a truly 
bad result, whereas in the "problem" described with Approval, a 
candidate wins who was approved by more voters than any other, and 
that was by a majority. Further, in the very rare situation that 
there is such a case, i.e, multiple majorities, it's possible to fix 
it in the same way as top-two runoff. If there is a rule that an 
election must be won by a majority of the votes cast, IRV cannot 
reliably accomplish this, apparently. Approval, and an older ranked 
form of Approval, Bucklin Voting, can, more reliably (because it 
counts all the votes), and, if no candidate gains a majority, a 
runoff can be held, where gaining a majority of votes is a practical 
certainty. The rare situation where there is a multiple majority 
could be considered one where the majority choice is not clear, and a 
runoff held.

But, by far, the simplest and easiest reform is to stop discarding 
single-winner ballots with more than one vote on them. It's not a 
violation of one-person, one-vote because, in the end, only one vote 
counts, one cast for a winner, and all the other votes can be seen to 
be ultimately moot alternative votes. Never more than one counts for 
the winner.

>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.

Lie. What's been done is to redefine majority. IRV is declaring 
winners, currently, based on less than a majority of ballots cast. 
It's even worse a situation because what is generally being 
implemented is not fully-ranked IRV (as described in Robert's Rules) 
but "RCV", Ranked Choice Voting, with only three ranks allowed. In a 
few elections, that would be enough, but in San Francisco, for City 
Counselor, they may have twenty candidates on the ballot. San 
Francisco was running a lot of runoff elections because of those 
races. IRV didn't fix the majority problem. Bucklin Voting, which 
used, in Duluth, Minnesota, the same three-rank ballot, but with 
slightly different rules, would more often find a majority, because 
it counts all the votes. But no method can guarantee a majority in a 
single round without candidate elimination. From all the RCV 
elections in the U.S. so far, in this recent history since San 
Francisco started up in 2004, the exact same result would have 
occurred with Plurality as with IRV, as long as people would have 
voted the same first preference as they did under IRV. Where there 
was no winner by a majority in first preference, there was no 
majority found through the vote transfers, with, I think, one 
exception, maybe two. Out of about 23.

Now, what is truly remarkable is that IRV is not producing the same 
outcomes as top-two runoff. But the top two in first preference is 
remaining the top two after vote transfers. IRV is, instead, always 
awarding the victory to the candidate with the most votes in the 
first round, for, generally, when votes of eliminated candidates are 
transferred, it is as if the population of those voters, preferring 
an eliminated candidate, were a sample of the same population that 
voted for the candidates that remained. So the relative positions of 
the candidates remains the same. I didn't expect to find this! 
Historically, IRV has been known to reverse the first round, 
apparently, I think Ann Arbor may have shown an example of this. But 
it is unusual, apparently.

All that expense to avoid runoff elections -- and the conversion 
expense is quite significant. IRV has been "passed" in many 
communities, but actually implemented in many fewer, for lack of 
funding. Approval -- no cost. Just strike a couple of lines in the 
election code. It's *simpler.* Just Count All the Votes!

Before IRV, in San Francisco, about one election out of three that 
went to a runoff reversed the first round plurality victory. It's not 
happening with IRV. So IRV does not merely eliminated extra election 
expense, it also eliminates the right of the majority to make a 
choice. Instead, it's creating plurality winners, winners who did not 
gain a majority of votes from those voting in that election. It would 
have been even cheaper -- much cheaper -- to simply eliminate the 
runoffs. And then, if one wants to regain majorities, use Approval 
Voting or Bucklin Voting.

Bucklin was used in the U.S. for some years early in the last 
century. It was, in Duluth, quite popular, as is apparent from the 
proceedings in Brown v. Smallwood, where Bucklin was ruled 
unconstitutional in Minnesota, following arguments that were not 
accepted anywhere else. Contrary to FairVote propaganda, that 
decision would have outlawed IRV as well, for it was against the 
concept of any alternative vote at all, not merely the kind of vote 
involved in Bucklin. Because the majority opinion (which was fairly 
confused from my perspective) mentioned that a vote for a second rank 
candidate in Bucklin could harm the election prospect of one's 
favorite -- which is true -- FairVote has claimed that the decision 
was *only* about Bucklin, but this neglects most of the language of 
the decision. This quality of IRV, called Later-No-Harm by election 
methods experts, is just about the one reasonably desirable quality 
of IRV, but it conflicts with finding compromise winners who will be 
more broadly satisfactory.

I should mention how Bucklin works, I'll give the Duluth 
implementation, there were others. Voters were allowed to vote only 
for one candidate in first rank, one in second rank, and as many as 
they liked in third rank. If a majority was found in the first round, 
that was the winner. If not, the second round votes were added in -- 
no eliminations, so the total now can exceed the number of voters -- 
and if a majority (of the number of valid ballots) is found, we're 
done. And if not, then the third round. For an example of a Bucklin 
election, see Brown v. Smallwood, the results of the election 
involved are given. IRV probably would have resulted in the same 
winner, with a lot more counting complexity and expense.

Bucklin simply involves counting all the votes and adding them up. 
Not so with IRV. IRV is not precinct summable, you can't just take 
the totals from precincts and add them together, for the result from 
a precinct (after the first round), depends on the results from all 
the other precincts in the previous round. IRV is rare among election 
methods for failing the "summability criterion." (Condorcet methods 
can be precinct summed, what is summed is a matrix of votes, because 
a Condorcet method can be considered to be a series of simultaneous 
pairwise elections, for all the pairs involved.)

>   It
>eliminates the spoiler effect,

Yes. It eliminates the first-order spoiler effect, not center 
squeeze, which it creates through candidate elimination. Candidate 
elimination is one way of saying "vote elimination." Approval 
similarly eliminates the spoiler affect. If the third party 
approaches parity, some voters have more difficult decisions to make, 
but no big surprise like center squeeze.

>  and it encourages positive
>issue-oriented campaigns instead of mudslinging opponent-attacking
>campaigns.

Well, I laughed at this one. It's not true, according to reports in 
the San Francisco media. It's a *theory*, wishful thinking, that came 
to be promoted as a fact, as part of the campaign to get RCV in San 
Francisco, but apparently there is no evidence that there actually 
was this effect. I'd really wonder if Chessin has *any* evidence at 
all that this happens. By now there have been a few elections, after all!

>  It is also easy to understand and easy to administer.

!!! IRV is, in fact, quite difficult to understand. It *seems* easy. 
How long, Mr. Chessin, did San Franscisco take to release their 
election results last November? A single tabulation error in one 
round can affect the sequence of eliminations, and, essentially, it 
all has to be done over again. It is not just a matter of adding up 
votes. It's extraordinarily complex.

Just thinking about IRV, most people won't think about Center 
Squeeze, for example. They won't even imagine that such a thing could 
happen, that the IRV winner could be opposed by two out of three 
voters, in favor of another. Would this be some rare thing?

Yes. It's not common. It takes three candidates running about even in 
first preference, most commonly. That's rare. How often does it 
happen in real IRV elections? Well, this is an interesting question, 
don't you think. *We don't know, because the necessary data isn't 
generally reported. When a candidate has been eliminated, they stop 
reporting the votes for that candidate, so we don't know the second 
preference votes for a candidate after they are eliminated. We only 
know the second preference votes of the eliminated candidate's 
supporter's ballots, those are the ones that show up in the totals. 
IRV does not count all the votes, far from it. It doesn't even count 
most of the votes, period. For example, when a candidate wins in the 
first round, which is normal, except in those zillion-candidate San 
Francisco elections -- which happens for only a few council seats -- 
all the lower ranked votes aren't counted. That could be done with 
Bucklin, too, to save money. But nowadays, with automated equipment, 
why not just Count All the Votes.

yes, I like to repeat that. I think it's a great slogan, don't you?

As I mentioned, even though IRV is horribly complex, it gets better 
if voters can vote for more than one in a rank. Bucklin considered 
ballots with votes for more than one in first and second rank to be 
spoiled, I see no good reason for that. In general, method 
performance improves if all the votes are counted. IRV, for example, 
if overvoting is allowed, becomes more like Approval, which performs 
better. But if one is going to use a preferential ballot, Bucklin 
makes more sense and is cheaper to count and does find more 
majorities, if that matters. (IRV is not finding a majority because 
once it has come down to two candidates, the eliminations stop. It is 
possible that some of the supporters of the runner-up also voted for 
the IRV winner, so *maybe* there was actually a majority. We don't 
know. In some cases, though, this was very unlikely. Bucklin does 
uncover all those votes.

>   Exit
>polls of voters show that they vastly prefer IRV to two-round runoff
>elections, and this is across all ethnic groups.

Well, I've looked at several studies and I find them much less 
glowing than that, and there is a study from England that found 
problems with handicapped voters and IRV.

Are those voters aware that IRV is *not* finding majority winners? 
They've basically been lied to. When the proposition establishing RCV 
was presented in 2002, the voter pamphlet, explaining the method, 
said that "the winner will still be required to win a majority of the 
votes." For some reason, all the opponents presenting pamphlet 
arguments didn't pick up on this, detailed knowledge of election 
methods is not widespread. The RCV initiative actually removed the 
majority requirement from the election code. If they had left it in, 
and if they counted all the votes, i.e., checked to see if there were 
any votes to be transferred from the runner-up (which I'd consider 
fairly as being "for the candidate" in some way), they would have had 
a couple of runoffs, not nearly as many as before. But they could 
have done this much more cheaply with Bucklin or just by tossing the 
no-overvote rule.

>I'm not sure how Kathy defines "support", but it must be different from
>how I would define it.  IRV uses the same basic logic as a traditional
>runoff election to find a majority winner.

No. If it did, then the results would be the same. They aren't. One 
out of three runoff elections in San Francisco resulted in reversal 
of the primary plurality result. None of the IRV elections did. A 
major cause is probably the large numbers of exhausted ballots, 
produced because either voters did not rank enough additional 
candidates or because the ballot only allowed them to rank three. In 
a real runoff, voters are faced with a specific choice, and there is 
more opportunity to examine those candidates closely and see how they compare.

Chessin is following a strategy for promoting IRV that was developed 
in the 1990s. Before that, nobody compared IRV with top-two runoff, 
as far as I know. IRV is the single-winner form of STV, which is a 
pretty good method for doing proportional representation, albeit 
probably more complicated than necessary. This method of sequential 
elimination is more widely used for PR, and PR was the original goal 
of what became FairVote. Most election theorists consider PR a 
reasonable and proper goal; the problem is that when the method is 
used for single-winner, defects that aren't so important when you are 
using it to elect many winners, instead loom large. Apparently, 
FairVote decided that going directly for PR here was too difficult, 
and one obstacle was the complex method used for voting. But if they 
could sell the method as replacing top-two runoff, the cost savings 
might pay for the conversion. And they got stuck on this strategy, 
and no matter how much election method theorists pointed out to them 
that they were promoting a pig, they simply continued to find ways to 
present it, to put makeup on the pig.

I used to think that IRV was, at least, better than plurality. I've 
come to be not quite so sure. Top-two, in any case, is much better 
than commonly thought. What sometimes is considered a defect, the 
allegedly lower turnout with runoffs, may actually improve the 
results by effectively introducing a kind of rough Range Voting 
effect: an inconvenient voting round tends to measure preference 
strength. I have actually proposed that runoffs be used with Approval 
and with Range whenever the results don't clearly indicate a majority 
approval of the result. And simulations of Range Voting, which is the 
best single-ballot method known (it actually uses *as the method* the 
only objective measure of election performance), show a small 
improvement in result quality if there is a top-two runoff.

(How could that be, if it's ideal? Well, there is a technical 
problem, which is that the ideal method minimizes overall "Bayesian 
regret," which is a measure of how far the result falls short of 
optimal, and that is generally based on simulated internal utilities 
for the election of each candidate, and the range of these utilities 
for each voter varies. But voters don't vote absolute utilities, they 
vote votes, and normally they will vote a full vote, which in Range 
means that for at least one candidate they vote the maximum rating 
and for one they vote the minimum. This means that different 
preference strengths end up being equalized. Normally, to some 
extent, this averages out, but top-two tends to fix it, because the 
votes get renormalized. And in real runoffs, there is an additional 
effect, which to my knowledge hasn't ever been simulated, that the 
people who vote in a runoff are generally more highly motivated, 
which means that they *actually* have stronger preferences.)

In San Francisoco, instead of RCV, it was argued, they could have set 
the primary election *early*, with the runoff at the general 
election. That's what was done in Cary, NC. Those runoffs, held with 
the general Novemeber election, *also* reversed the first round 
winner one time out of three. IRV is destroying majority choice. And 
I think those polls results will start changing if people realize 
that. There are much better election reforms that are cheaper and 
don't do that.

Top-two runoff has its problems. Really some of the same problems as 
IRV. The best winner can easily lose in the first round and not make 
it to the runoff. It is quite easy to fix this! Just start counting 
all the votes, as the simplest solution. Use something like Bucklin 
if you want more voting flexibility. Use Range if you want more 
accuracy. Range can be low-resolution (very simple ballot, one form 
that I've seen is, for each candidate, you can vote -1, 0, or +1. 
This is a Range 2 ballot, one step up from Approval) or 
high-resolution (anything from 0-9 or 0-10 to 0-100).

But by far the simplest reform is Approval. No cost. Same ballot as 
plurality. Finds more majorities. Note that FairVote will claim that 
people will bullet vote. Yes, they will. Most people. But not the 
supporters of third parties, generally, that don't have a hope of 
winning. That is how it fixes the spoiler effect, which usually 
involves only a few percent of voters, rarely more than 10%.

>   Any argument along the
>lines Kathy presents is equally true of IRV AND traditional runoff
>elections.

Well, not knowing what Kathy wrote, I can't comment in detail. But I 
just mentioned that there are similarities between top-two runoff and 
IRV; but those similarities end at the first round. The second round 
in top-two is a new election, with the candidates being now very 
visible and very clearly being compared, which often is obscured in 
the first round. This is quite likely the main reason for the 
reversals we saw with top-two that we don't see with IRV>

>   Imagine a traditional runoff election (or IRV election)
>with three candidates, A, B, and C.  If A gets 48%, B gets 46% and C
>gets 6%, the runoff would be between A or B.  Now let's say all the C
>voters prefer B over A, so B ends up with 52% to A's 48%.

Preposterous assumption. Where is he going with this? Here is what 
really happens. The C voters tend to vote about the same ratio for A 
and B, with some of them not adding any additional preference. This 
election is close. So it's quite hard to predict.

(The result I've seen has been born out in every IRV election that 
went to elimination rounds. It is not expected, it is 
counter-intuitive, perhaps, though, when you think about it, it does 
start to make sense. Voters, quite simply, aren't lined up as neatly 
as we think. Partisans of the top two, we don't now much about, 
because we aren't seeing their second preference votes. But voters 
for the other candidates seem to be, pretty much, coming from the 
same population, and their votes get distributed without changing the 
candidate rankings. A reversal like Chessin suggests simply isn't 
happening in these recent elections, it may happen in some special cases.


>I think Kathy's statement is based on the following scenario, in which
>someone might claim that C (the one who got 6%) is actually supported
>by the MOST voters.  Here is how that reasoning would work:  It COULD
>be that all of the supporters of B would prefer C over A and all the
>supporters of A would prefer C over B, thus (according to this view) C
>is in some sense "supported" by 94% of the voters (all of A and B's
>supporters).  But in this scenario C is still only the first choice of
>6% of the voters, and 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.

Okay, I see what Chessin is getting at. He's proposed a very weak 
example. This is center squeeze, made to look really as bad as he 
could. We should expose, right now, that he is arguing against 
Robert's Rules of Order and what is well known.

FairVote essentially invented a new election criterion about first 
preference votes, they call it the Core Support Criterion, one 
recognized by nobody else because it, quite simply, is not a 
desirable thing in itself, though it is associated with something 
that might, under some circumstances, be desirable, and you'd better 
believe that these propagandists want you to make that connection. 
Center squeeze with the above numbers is extraordinarily unlikely, 
because someone who is truly the second choice of nearly everyone is 
probably going to have more than 6% of the vote.

Suppose instead that we have these preferences:

34-: A>B>C
33+: C>B>A
33:  B>C>A

The voters are more or less equally divided between the three 
candidates, in first preference. B, however, loses the first round by 
a small margin. Since B is eliminated, C wins. Apparently by 66:34, a 
landslide. Actually, though, if there were to be a face-up between B 
and C, B wins by 66:34.

IRV discards, does not count, the B votes concealed underneath the A 
and C votes. Any Condorcet method would detect this and declare B the winner.

Approval might or might not, it depends on preference strengths, it's 
hard to predict that from a ranked ballot description. Probably, 
though, some of the non-B voters would vote multiply, and there are 
twice as many of them as there are of B voters -- who might also vote 
multiply -- so I can say that B woudd *probably* win.

Bucklin would have no problem with this election.

                A   B  C
First round:  34  33 33 no majority
Second round:  0  67 33 second rank votes
total:        34 100 66 B wins.

Note that as B was the first or second choice of all the voters, B 
actually enjoys a kind of unanimous vote.

Now, can Chessin look at this election with a straight face and claim 
that, because B was in third place in the first round, she should not 
win the election? It's preposterous.

> >I belong to an email list which discusses various alternatives for
> >voting schemes, and although I don't pay close attention to this list,
> >it is generally agreed on the list that IRV does not work and that
> >other approaches might work better.

Yes. She does.


>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.  But all
>of them agree that our current election method, plurality elections
>(complete with the spoiler problem), is the worst method.

Not all. Many of us used to think that. Now, the list she is talking 
about is, of course, the Election Methods mailing list. It's an open 
list, it is not controlled by some faction, but, yes, long ago, IRV 
supporters disappeared, for the most part, from this list. It's 
because of the *arguments*, stupid!!!

Essentially, it's impossible to maintain the arguments for IRV in the 
presence of people who understand them. Many, many of these people 
started out as supporters of IRV. But then they learned about 
election methods. IRV is very rarely used for direct, single-winner 
elections. It is *not* better than top-two runoff. Not if you think 
election quality is important. But it shares with top-two certain problems.

Yes, the EM list has proponents of many different methods. And nearly 
all of them agree that Approval would be an excellent first step. As 
to Range, the general agreement is that it's an excellent method if 
voters would only vote "honestly," and that's a very complex 
question, it turns out. But there is some relatively objective 
evidence that Range is indeed the best method, even with "strategic 
voters." If everyone votes strategically, with good knowledge, the 
election reduces to Approval. Which is not a bad outcome. (But when 
you don't have good knowledge, the safest vote is actually the 
so-called "sincere" Range vote.)

I've arranged the simple methods in a simple hierarchy:

Plurality: Vote for one, candidate with the most votes wins.
Approval: Vote for as many as you choose, candidate with the most 
votes wins. Same ballot as Plurality.
Range: Vote for as many as you choose, fractional votes allowed. 
Slightly more complicated ballot.

Then there are more complex methods, such as Bucklin, with its 
"instant runoff approval voting."

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

This is pure deception. Let's take a look at the evidence he 
presents. It happens that I've had occasion to look into this. I've 
been quite surprised how easily this falls apart.

>   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.

That really sounds significant, doesn't it. After all, *Political 
Scientists*. They must really think that IRV is hot stuff. Well, does 
it say IRV? No. Actually, that provision was put in the APSA 
constitution a long time ago. As far as I've been able to tell, it 
has never been used.

Notice how this was phrased. It is always phrased that way. Because 
it's true if stated that way. But what if I wrote that "APSA elects 
its national president using IRV." That's what they want you to 
think, after all! It would be a lie. That the claim is always stated 
in exactly the right way to make it true is a pretty strong clue that 
Chessin and the others know that they are being deceptive. Or they 
are simply good at memorizing the party line. I do have inside 
information that they tell people exactly how to promote IRV, use 
this language, not that language, etc.

Now, let's assume that the political scientist really know what they 
are doing. That's what's being implied, right? Perhaps we should 
elect our President this way, right?

So what do they do?

Well, there is a Nominating Committee. The nominating committee 
nominates the President-elect. So can the members, through some 
process. If there is no nomination other than the one from the 
nominating committee, then the President-elect is ratified at the 
Annual Meeting or whatever they call it.

Yes, the bylaws provide that if there is another nomination, there is 
a mail ballot. And if there are more than three nominees, then a 
preferential ballot is used, and the method is described, and it's 
the same as IRV.

To my knowledge, it has *never* been used. It may have been there 
since the founding of the organization, preferential voting was big 
in the U.S. then.

This is really typical FairVote propaganda. Spin.

Look, anyone who doubts this and who is interested in Election 
Methods, join the EM list like Kathy did. Ask questions. You will get 
answers. Some participants are blunt, but most are polite. IRV is not 
popular on that list because it's a lousy method, it's pretty simple. 
The people on the list, collectively, have no axe to grind, though 
some have become a bit disgusted by the consistent refusal of what 
used to be called the Center for Voting and Democracy to actually use 
any sort of democratic process to make decisions. And by the repeated 
and totally spurious arguments that get repeated over and over again, 
they pop up like a whack-a-mole game, completely undeterred and 
unaffected by facts and patient explanation.

>   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.

Uh, there are people on the list who come from Australia and who 
think STV is pretty good. It's true that at present the use of 
methods other than Plurality, Top-two, and IRV for single-winner are 
rare, but Borda Count is used (Kiribati and Nauru). And Borda Count 
is actually Range with a defect that is easily fixed. Just allow 
voters to place more than one candidate in a single rank and allow 
ranks to be vacant and, presto! Range Voting. But there are a couple 
of people who still think Borda is Best.

In the past, though, Bucklin was used in the U.S. Approval is used by 
deliberative bodies, sometimes, the United Nations General Secretary, 
it could be said, is elected by Approval. Approval Voting was also 
used for the election of popes and for the election of Doges in 
Venice, for 500 years, it was *very* stable. (The method was complex, 
involving stages and the drawing of lots, but the votes that were 
part of it were Approval votes.)

Used to be we thought of the U.S. as leading the world in democracy, 
but now, apparently, FairVote considers that we should follow. If a 
couple of places around the world, all with strong two-party systems 
(or sometimes two-coalition systems), use IRV, then, why, we should 
take it up! Right?

I would definitely not support IRV, ever, and the reason is that it's 
an expensive change. There is a very simple change that *also* 
addresses the spoiler effect, and that can gradually be modified to 
be even better, either by adding ranks (Bucklin) or by allowing 
fractional votes (Range), or, if it is so decided, by adding ranks 
counted as such rather than phased in as approvals as with Bucklin. 
There are lots of possibilities, but the first one is so simple and 
clear that we will wonder, years from now, why we waited so long. 
Just Count All the Votes!

>   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.

Because the votes aren't being counted that would tell us. From the 
similarities of top-two and IRV, we can tell that the problems are 
real, and they do show up in the simulations. The source code for the 
simulations is available, and anyone can critique it. But there 
aren't any actual election method theorists, knowledgeable enough to 
do that work, who support IRV, not that I've ever seen.


> >Also some other reasons not to promote IRV (besides the fact that it
> >often ends up putting candidates into office that are not supported by
> >- or even are opposed by - a majority of voters) include:
>
>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.

The problem is common, but not rampant. Chessin should know that. His 
comment about no candidate having majority support is, of course 
true, but he's confined that to Condorcet cycles, when, in fact, when 
you have enough candidates, it's common to have no candidate with 
majority support even when there is a majority winner. I've had 
occasion to follow debates with Chessin before, and he's not 
particularly knowledgeable.

Again, that's the comparison he wants you to make: compare IRV with 
plurality. Or with top-two. Usually the latter. With top-two, the 
argument is money. Save money on elections. But when you look at the 
results, saving the money by replacing simpler methods with IRV 
involves a high investment that *might* pay off after some years. But 
we can fix that spoiler effect by just counting all the votes, and 
where we need to spend money is in making sure that we do that 
accurately and honestly. Not by making it harder to count votes and 
audit the counts.

What Chessin has totally avoided and attempted to divert attention 
from is the possibility of an IRV election that *massively* fails to 
elect a *landslide* majority winner that would be obvious from the 
IRV ballots if they are all counted. Sure it's rare.

Why is it rare? It's rare because it is rare that there are more than 
two viable candidates. When there are only two candidates, Plurality 
actually does a fairly good job. Usually. Its the exceptions that 
worry us interested in election reform. IRV came out of a complex 
method, STV, designed for a complex task: proportional representation 
where the results are only decided by the voters, not by parties. 
Take that method and reduce it to one result, presto! Instant Runoff 
Voting. Now, the credit for inventing IRV is generally give to Ware, 
in the latter part of the 19th century. I found reference from that 
time to his work, and it was not well received by people very 
familiar with STV. The defects were immediately recognized. So why 
was IRV implemented, say, in Australia? Well, they were already using 
STV. So it was convenient.... Actually, I've not checked that, I 
should.....  But definitely, STV was invented first.

And a vastly better method for proportional representation and which 
could also be used for single winner, that *might* be even better 
than Range, was invented and published in 1886 by Lewis Carroll 
(Charles Dodgson). It was rediscovered by Warren Smith in, I think, 
2000; he called it Asset Voting. In Dodgson's version, vote for one. 
Candidates who receive votes may reassign them. They essentially 
become electors, public voters. To be elected requires a quota of 
votes; Dodgson used the Droop quota, used by most PR systems. No 
votes are wasted, except those given to a candidate who refuses to 
negotiate and compromise. Very, very simple method, that essentially 
chooses an assembly deliberatively. And I've written a fair amount 
about it.... Because no votes are wasted, you can vote for anyone you 
choose. There aren't losers, in that everyone's vote goes to a 
winner. That's *representation*. Somehow, we got stuck "electing" 
representatives. It was a bad idea. We should elect officers and, in 
fact, the best place to elect them is through a representative 
assembly, where very simple election methods, because ballots can be 
repeated, become very sophisticated and powerful.

> >1. It can be very difficult and time-consuming to count manually, and
> >so makes auditing elections very difficult.
>
>[Notice his response to my comment begins by employing a
>misinterpretation of my comment.]
>
>IRV has been counted easily and quickly in recent U.S. elections in
>places like Burlington, Vermont, and Cary, North Carolina.

Cary had lots of problems, so I don't know where he's getting his 
information. Burlington was a fairly small election, as I recall. San 
Francisco has had huge problems, delaying the full reporting of the 
vote by *months*.

>   Auditing
>can be more complicated, simply because there is more information from
>each voter to double check, but can also be more thorough.  San
>Francisco and Burlington, for example, posted a record of every
>ballot's set of rankings on the Internet, allowing anyone to tally
>those elections themselves.

It's a huge task. By the way, I tried to decode that San Francisco 
data and I wasn't able to figure it out. I'd love to see a reference 
to it that explains it. I don't think that full reporting has 
generally been done; the more recent San Francisco figures I've seen 
were not as fully reported as they were at first.

> >3. IRV makes it virtually impossible to conduct post-election exit
> >poll discrepancy and vote count pattern analysis to detect suspicious
> >patterns that are consistent with vote miscount and voter
> >disenfranchisement. I.e. IRV makes any problems with vote count
> >integrity harder to detect.
>
>This is false.  Exit polls can and have been conducted for IRV
>elections.  The fact that voters are giving more information about
>their preferences does not make detecting misconduct any harder.

I'd defer to an expert on this, which Kathy is. Exit polls are 
problematic enough with single votes. A lot of voters won't remember 
what the sequence was, but I'm not sure how big a problem this will 
be. With a lot of this stuff, we need experimental data because, I've 
already discovered, the real world results can be quite different 
than expected. I did not expect to find such poor results for IRV. I 
thought it would find a majority winner more often. And I expected to 
see the reversal of the first round plurality result, in some 
elections. Not once, in many elections, so far. (most of them have 
been in San Francisco, there has been, really, only a handful of 
elections elsewhere.)


>[He is incorrect in making the above statement, but I do have the
>expertise to rebut this one incorrect statement.]
>
>As explained previously, there are no perfect voting methods, although
>I believe that IRV is the best method.  Many of the advocates of other
>reforms seek to denigrate IRV in hopes of advancing their favorite
>reform method.  (I note that none of them have been successful in
>getting any jurisdiction to adopt their methods.)

We weren't funded back in the 1990s. And there is a common political 
phenomenon. Real political reform can be quite difficult. FairVote 
sells IRV to third parties based on giving them hope, but, in fact, 
IRV pretty much guarantees that third parties won't win elections, if 
we can expect what has happened elsewhere. In fact, though, IRV helps 
two-party systems function without disruption from third parties 
through the spoiler effect. So it may, indeed, be easier to implement 
the complex, expensive reform of IRV, and more difficult to implement 
the simple, let's count all the votes reform of Approval.

>As to IRV not achieving its "claims":  There are two claims often made
>for IRV that may be somewhat over-stated.  Although political
>scientists and courts have recognized IRV as a majority voting method,
>it is possible that some voters will not rank either of the two
>finalists in the runoff count.

Possible? Its a certainty, for many reasons. I'm not aware of any 
court decisions that have "recognized IRV as a majority method," but 
IRV does satisfy the Majority Criterion, which does not mean that it 
elects a candidate preferred by a majority over all other candidates. 
And that's with full ranking.


>   In this case, it is possible that the
>runoff winner will have less than 50% of the votes cast in the original
>election.

Notice: "possible." He isn't kidding! In nearly every RCV election in 
the U.S., in San Francisco, in Burlington, in Cary, NC, and in Takoma 
Park, Maryland (where IRV was truly a fish bicycle, but, hey, Rob 
Richie lives there), when it went to runoff -- i.e., a majority 
wasn't found in the first round, a majority wasn't found later, 
either. Sometimes the winner got as little as about 40% of the vote.

RCV in San Franciso is limited to three ranks, which with twenty 
candidates or more, makes it pretty likely one would see exhausted 
ballots. In an IRV election, many voters won't know enough candidates 
to rank even three. In a real runoff, they'd have had an opportunity 
to study the specific candidates that made it to the runoff.

>   This is the same dynamic as in a traditional runoff
>election, in which some voters don't show up at the second election,
>and the runoff winner may end up with fewer votes than 50% of those
>voting in the first round.  This "non-majority" winner is FAR less of a
>problem with IRV than it is with our current plurality elections, or
>with separate runoff elections (where turnout generally drops
>sharply).  Thus, while IRV will elect a majority winner among those
>voters expressing a preference between the final winner and the other
>candidates, one could claim this does not assure a majority winner
>among all the voters.

Turnout is lower in runoff elections held inconveniently. That is not 
necessarily a bad thing. In Cary, NC, where the primary was held in 
October and the runoff, if needed with the general election in 
November, turnout was the same for primary and runoff, where I 
looked. It will depend on how much voters care. If voters are 
indifferent to the top two (either considering them both equally 
good, equally mediocre, or equally bad), they are much less likely to 
turn out. But basic democratic principle, majority rule: a majority 
of voters voting. Not of voters not voting.

>The second claim of IRV that may be over-stated is that it eliminates
>the "spoiler problem."  While compared to our current voting method,
>IRV indeed solves the spoiler problem when independents or third
>parties run against two major parties.  However, concerns about
>"spoilers" could possibly resurface in some scenarios when there are
>three or more major parties.  But none of those scenarios have appeared
>in practice in those countries that use IRV and have more than two
>major parties.

None of them have more than two major parties. It appears that way in 
some places where there are really two  *coalitions.* I'll 
congratulate Chessin, though, for at least mentioning the problem.

> >I took a spreadsheet once and it took me just a few minutes very easy
> >to create examples when IRV voting doesn't work as anyone would want
> >to. I  recommend taking a few minutes to try that yourself before you
> >recommend IRV again.
>
>As Arrow's Impossibility Theorem shows, it is impossible to invent a
>perfect voting method, as reasonable and desirable features that one
>would want in a voting method are mutually exclusive.

Nope. Arrow's Impossibility Theorem does no such thing. Because 
people, at the time, were only considering ranked methods, where 
preference strength isn't collected, a lot of people drew that 
conclusion from Arrow's work, but that is not what Arrow claimed and 
it isn't what Arrow proved.

>   IRV does better
>than most in real world experience, and virtually all elections methods
>experts agree it is superior to plurality elections we use currently.

In the abstract, without consideration of cost, security, and some 
areas that quite simply have not been explored much. I've never seen 
expert analysis of the effect of preference strength on runoff 
elections, for example. The fact is, though, the vast majority of 
election methods experts, those who study election methods -- as 
distinct from political scientists, who are often quite naive about 
election methods -- would agree that Approval, simply counting all 
the votes, is a better reform than IRV. And it costs nothing.

>While it is easy to construct pathological scenarios for any voting
>system, none of the pathological scenarios for IRV have ever been
>observed in practice.

Because the data is hidden. Wonder why it's hidden? We now have 
*several* elections with visible data. The problem scenarios are rare 
in strong two-party systems. So... we can estimate how often the 
problems arise with simulations, and, ideally, we could check the 
results of the simulations against real elections, but this work 
takes time. It is ongoing. And believe, me, the people working with 
election methods -- as distinct from political activists like Chessin 
-- are really interested in the truth, they'd rather eat their hat 
than distort research.

> >Fair Vote is a misnomered organization.
>
>It saddens me that Kathy thinks this.  I would suggest people visit
>www.fairvote.org and come to their own conclusions.

Would you like a guided tour? FairVote has very carefully constructed 
arguments that can be quite difficult to disentangle. Some of their 
stuff fooled me for years. It's like being led into a salted mine. 
Everything is nicely laid out for the reader to conclude what they 
want you to conclude. It took me at least a year to figure out that 
Robert's Rules did not actually recommend Instant Runoff Voting. If 
one isn't really familiar with what is involved, and after reading 
the FairVote introduction, why, what's that Lomax saying, he's crazy, 
*obviously* it recommends IRV. What's he talking about? Well, all I 
can say is that FairVote did a great job. I should write up the 
Robert's Rules story. It's in Talk archives on Wikipedia for the IRV 
article. They describe IRV not because they recommend it, they 
actually criticize it, and what they recommend -- in circumstances 
where they clearly think it a last resort -- is "preferential 
voting." Like Bucklin, for example, or any Condorcet method. But they 
don't describe those, because they are not in wide use, and Robert's 
Rules describes current practice. Actually, with all deliberative 
bodies, Robert's Rules prefers repeated balloting without 
eliminations and, naturally, requires a true majority vote. It 
requires special bylaws to allow anything else, including IRV without 
a true majority requirement. When I started pointing this out on 
Wikipedia, it was like I was saying that two plus two doesn't equal four.

When I pointed out that the IRV version given in the Rules required a 
true majority vote, they called me silly and obstinate. And then, 
just the other day, in fact, I noticed that a few pages later the 
Rules note that "sometimes voters don't add additional preferences, 
mistakenly thinking that this will help elect their favorite, but 
this can cause the election to fail." [please remember, I'm quoting 
from memory, this may not be exact.] It can only cause the election 
to fail if "majority" means "majority of all the valid votes cast." I 
had been previously relying on opinions from parliamentarians on the 
web, elsewhere....

Now, it is often said that adding a later preference, with IRV, can't 
hurt your first preference. That's true, with plurality-election IRV. 
If a majority is required, it's not true. With the IRV legislation 
Terrill Bouricius proposed when he was in the Vermont legislature, 
that claim about later-no-harm was given in the specified ballot 
instructions. But the Vermont Constitution requires a majority, and 
they knew full well that the IRV "majority" wasn't a true 
constitutional majority. And your second rank vote might cause the 
election to complete, so no election in the Legislature. (If no 
majority, the Legislature chooses the Governor by secret ballot from 
the top three in the regular election). If there were an election in 
the legislature, and your favorite was one of the top three, then 
your favorite could win. Adding a second preference *could* harm your favorite.

But that would not stop me from adding a preference. I happen to 
think later-no-harm is actually undesirable as an election quality. 
If I agree with my neighbor about a compromise, I'm "hurting" the 
"chances" of my first choice. And that is exactly what I should do, 
if the compromise is a reasonable one. Elections are not a zero-sum game.

> >"Sophistry" is a statement or claim that at first glance sounds good,
> >but upon closer examination is found to be incorrect or inaccurate.
>
>And not applicable, I hope, to anything I've said about IRV.

Unfortunately, it does.

I know that Chessin has seen most of this before, I've followed the 
correspondence....