[EM] Why I think IRV isn't a serious alternative 2

[Abd ul-Rahman Lomax]

At 12:56 AM 12/21/2008, Kevin Venzke wrote:
>Hello,
>
>--- En date de : Ven 19.12.08, Abd ul-Rahman 
>Lomax <abd at lomaxdesign.com> a écrit :
>
> > With LNH, the "harm" is that the voter sees a
> > second preference candidate elected rather than the first
> > preference.
>
>Actually, the harm need not take that form. It could be that you add an
>additional preference and cause an even worse candidate to win instead of
>your favorite candidate.

That's not called LNH, I think. LNH: Adding an 
additional preference cannot cause a higher preference candidate to lose.

With Bucklin, the described behavior can't occur, 
if I'm correct. The additional vote maintains a 
vote for both candidates against all other 
candidates, and only abstains from that pair. No, 
*failing* to add an additional preference can cause the effect described.

Basically, the claim is that LNH is important 
because voter fear of LNH failure will supposedly 
cause voters to bullet vote, thus causing the third candidate to win.

But this is precisely a situation that the voters 
can handle. They can balance the benefit and 
risk; to do this zero knowledge is possible. I've 
stated that the voter should, in that case, 
assume that the voter's own preferences are an 
indication of the overall preference of society; 
making this fuzzy, understanding that it is 
fuzzy, can then indicate a reasonably voting 
strategy. I.e., the voter doesn't assume that the 
voter's own position is *absolutely* an accurate 
sample, just that it's a rough indication. The 
voter can also make different assumptions; if the 
voter knows that his or her own preferences are 
idiosyncratic, unusual, the voter will normally 
have *some idea* of how others generally feel. A 
voter who knows that his or her own position is 
far-left, as defined by society, will usually 
know this! So the voter can model, normally, how 
others are likely to vote. But it's better, of 
course, to put this information together with 
polls and other information, such as personal contact with other voters.

What's being said here is that voter knowledge of 
the *situation,* the *context* of an election, is 
an important element in voting and that it is 
*normal* and *functional* for voters to 
incorporate this judgement; that there exist some 
abstract utilities, totally independent of 
context, on which voter preferences are based, is 
actually quite a leap away from what we know, 
it's an unwarranted assumption. It seems that we 
are designed, more or less, to find VNM 
utilities, roughly, from the get-go, we do not 
sit down and consider all possibilities, our 
considerations are prefiltered following what we 
think possible in the first place.

Dhillon and Mertens actually address this: they 
start with a total candidate set, all possible 
candidates, but that set considers what is 
socially possible. It's prefiltered, in fact. 
"All possible" is the key word. It includes 
election rules and social norms. It's larger than 
the candidates on the ballot, but it is not unlimited.


>Yes, but the concern should not be that you personally will ruin the
>result, it's that you and voters of like mind and strategy will ruin the
>result.

There are two approaches: true utility for 
various vote patterns, which is the "last voter" 
utility, since if your vote doesn't affect the 
outcome, it has no utility (except personal 
satisfaction, which should, in fact, be in the 
models. There is a satisfaction in voting 
sincerely, all by itself, and this has been neglected in models.)

The other approach is the "what if many think 
like me?" approach. That's not been modeled, to 
my knowledge, but it's what I'm suggesting as an 
*element* in zero-knowledge strategy. It's 
particularly important with Approval! The 
"mediocre" results in some Approval examples 
proposed come from voters not trusting that their 
own opinions will find agreement from other 
voters, and if almost everyone votes that way, we 
get a mediocre result. This actually requires a 
preposterously ignorant electorate, using a bad strategy.

Optimal strategy shifts if a majority is 
required. Requiring a majority, in a first round, 
shifts optimal strategy toward the bullet vote, 
in Approval as well as in other methods. It 
shifts Range toward stronger expression of clear preference.

The electorate, then, has a solid basis for 
determining strategy in the next round! (This 
works in unlimited round systems, and classic 
Approval studies considered how the rounds would 
settle, but it also works with two rounds, just not as flexibly.)

 From the beginning, we should have questioned 
the tendency to believe that "strategic voting" 
was a Bad Thing. It's actually a necessary thing; 
VNM utilities are "strategic," and it's only the 
use of VNM utilities that can make a method 
Arrovian-compliant. At least that's what Dhillon 
and Mertens seem to claim to have proven, 
Relative Utilitarianism, they purport to show -- 
and it's been long enough, and enough work has 
built on their work, that it's unlikely the proof 
is false, it just hasn't been popularized -- is a 
unique solution to the Arrovian conditions they 
define, which aren't strange definitions at all, 
they are actually "weak." I.e., should be relatively easier to satisfy.

> > I'd have to look at it. How does MMPO work? I worry
> > about "nearly," [...]
>The "opposition" of candidate A to candidate B is the number of voters
>ranking A above B. (There are no pairwise contests as such, though the
>same data is collected as though there were.)
>
>Score each candidate as the greatest opposition they receive from another
>candidate.
>
>Elect the candidate with the lowest score.
>
>This satisfies LNHarm because by adding another preference, the only
>change you can make is that a worse candidate is defeated.

Okay, that's clear. Now, "nearly" a Condorcet 
method? But this is a peripheral issue for me. 
Reading about MMPO, my conclusion is that, absent 
far better explanations of the method and its 
implications than what I found looking, it's not 
possible as an alternative. Hence in my own 
mental VNM utilities for voting systems, it's got 
low preference strength, and I really don't know 
where to rank it.... It would be nice if we could 
get to the point that political practicality were 
more ... rational. But we aren't there.

I've just been doing some research on Bucklin and 
searching for preferential voting or the 
preferential system, the name by which it was 
often known in the U.S., I came across some IRV 
stuff, specifically the San Jose Measure M that 
passed in 1998. What is obvious is that the 
electorate there was deceived, with the 
opposition -- raising a legitimate concern -- 
being outmaneuvered by Steve Chessin. The 
"unbiased" description of Measure M, 
unfortunately, was dead wrong. And Chessin, I'd 
have to assume he knows the truth, lied in the 
ballot arguments. Directly and plainly. Unless, 
perhaps, the IRV that was being approved by 
voters there is different from every other IRV 
implementation in the U.S. That would be refreshing!

My point is that we need to focus better. San 
Jose is an active issue, there were efforts to 
get IRV in place this year, and those will 
continue. There is a real need for reliable 
information about IRV and other systems, and for 
that information to be made available to 
decision-makers and, for example, the county 
counsel who issued the "unbiased description."

"Preferential voting" has waited 10 years to be 
implemented there because of the serious counting 
difficulties. Yet, it appears, preferential 
voting, the "American" version, was used in San 
Francisco, I have not yet determined the details, 
but something like 1916 as an adoption date, and 
it lasted for some years. It was Bucklin. Which 
is far easier to count, and which is *more* 
likely to find a majority than IRV, and the only 
serious argument against it is LNH fear: will it 
cause voters to fail to specify additional 
preferences? But, in fact, if it's not considered 
"instant runoff," if it doesn't *replace* runoff 
elections, but merely makes them more uncommon, 
that isn't actually a problem. It appears that 
it's also, quite simply, not what happened with 
municipal elections using Bucklin. There were 
very healthy numbers of additional preferences expressed.

The LNH argument crops up in books about 
preferential voting, in particular one where the 
"American preferential vote" and the "English 
preferential vote" were being compared. (I.e., 
Bucklin and STV). Purely theoretical, no actual 
data of substance, no consideration of the meaning and context. Same as today.

There is no evidence, in fact, that voters will 
actually fail to add preferences with Bucklin 
more than they will with IRV. What's been missed 
is that bullet voting is a totally normal voting 
pattern, having little or nothing to do with LNH 
fears. Sure, *in theory*, LNH fears could 
suppress additional preferences .... but, then, 
strategic concerns could increase them. *Both* of 
these are forms of strategy. One proponent of 
Bucklin answered that voters, if they have some 
mild preference between two candidates, the 
favorite and one a bit less favored, would be 
more concerned about defeating a bad candidate 
than about their vote helping the less favored 
one beat the more favored one. And this is 
actually quite a sound argument, made in roughly 1920....

>DSC is harder to explain. Basically the method is trying to identify the
>largest "coalitions" of voters that prefer a given set of candidates to
>the others. The coalitions are ranked and evaluated in turn. By adding
>another preference, you can get lumped in with a coalition that you
>hadn't been. (Namely, the coalition that prefers all the candidates that
>you ranked, in some order, to all the others.) But this doesn't help
>the added candidate win if a different candidate supported by this
>coalition was already winning.

MMPO is easy to explain, but the *implications* 
aren't easy without quite a bit of study. DSC 
being harder to explain makes the implications 
even more obscure. Thanks for explaining, I 
appreciate the effort; but I'm prioritizing my 
time. I'll need to drop this particular 
discussion. If, however, a serious proposal is 
made for implementing one of these methods, I'll 
return. Right now, what I see is that Bucklin 
should get priority, as a form of Approval that 
*better* satisfies LNH concerns and the very 
legitimate desire to express a favorite, and as a 
very simple and easy to canvass and understand 
method. It will produce the same results as IRV 
in the favorite scenarios of FairVote: spoiler 
effect situations, but it continues to function 
-- and has been proved to function in history -- 
in situations where Center Squeeze would cause serious IRV failure.

And the fact that IRV is taking down the best 
voting system we have in current application in 
the U.S., based on a thoroughly misleading 
argument that IRV simulates Runoff Voting, and is 
supposedly cheaper, must be addressed. There is 
work to do. San Jose is about to, probably by 
next election, replace Runoff Voting with IRV, 
based on a ballot Measure passed ten years ago, 
that radically misinformed the electorate, in the 
*same way* as the San Francisco electorate was misinformed.

The local Libertarian Party chair caught it, but 
didn't explain it clearly enough. He was whacked 
down by other Libertarians who said he didn't 
have the authority to speak on behalf of the 
party, and Chessin made sure this was mentioned 
in the rebuttal. That rebuttal, however, 
explicitly affirmed that the winner would be 
required to obtain a vote from the "majority of 
ballots." He said this denying the concern raised 
by that Libertarian. Oops! That's a true 
majority, and cannot be found unless you require 
full ranking, which is almost certainly 
unconstitutional in the U.S. (Dove v. Ogleby, Oklahoma).

Chessin, did you realize you were deceiving the 
voters? That was a long time ago. The distinction 
has escaped a lot of people, even those who 
should know better, because they imagine full 
ranking, even though we *should* have known that 
full ranking simply doesn't happen unless you 
coerce the voters, as they tried in Oklahoma.

> > > In other words: I want to have a TTR election where
> > candidates risk being
> > > spoilers if they place worse than third.
> >
> > That would be a system where the candidate is risking
> > damage to the overall benefit of the election. Did you mean
> > to write it as you did? A spoiler typically will drop the
> > "spoiled" candidacy one rank, not two.
>
>That is what I meant to write, although I don't understand your second
>statement.

In real voting systems, candidates who aren't 
going to win *always* risk being spoilers in the 
sense stated. It's not avoidable, unless you 
coerce votes, which ain't gonna happen in the 
U.S., at least. However, the risk can be greatly reduced.

>As far as it being a "system where the candidate is risking damage to
>the overall benefit of the election": We already have this with FPP,
>with every candidate who places third or worse.

The problem is that there is disagreement as to 
the "overall benefit." Nader claimed that there 
was no difference between Gore and Bush. This is 
equivalent to a position that the benefit of a 
third party candidacy, expressing support for 
that party, is more important, of more benefit, 
than making a choice between Tweedledum and 
Tweedledee. I think the position stank, and still 
stinks -- Nader continues to defend it, in spite 
of the huge damage done, which damage would have 
been less likely than with Gore -- but I have to 
say that this was a decision, how to vote, made 
by each voter. Enough agreed with him, and it's 
revisionist for us to say that, "Well, they 
didn't mean it. They *really* preferred Gore." 
Why? They voted for a candidate but didn't 
believe what he was saying? No, they did *not* 
prefer Gore. Some of them thought that it was 
*better* for Bush to win, because then it would 
get really bad and then comes the revolution....

However, I'm pretty sure that if the method in 
Florida had been Approval or Bucklin or IRV, we'd 
have seen some additional Gore votes. But what's 
hard to figure out is what additional votes would 
have gone to Bush, there would have been those. 
My guess is that, since the election was very 
close with Plurality, it would have remained very 
close with any of the preferential voting methods 
or Approval, and possibly also Range. Now, 
imagine the flap over counting the vote if it had 
been IRV in Florida, and close! Bad enough with 
Plurality or one of the other sum-of-votes 
methods. Total nightmare with IRV....

>Basically I want a hybrid of FPP and TTR, that does better than either
>at providing an actual third choice that might be able to win. That
>everyone and their mother can be nominated fairly safely under TTR is
>nice and democratic, but I think it's a waste of potential.

Sure. Bucklin. It seems to have handled an 
election in Portland, OR, or was it Seattle, with 
92 candidates on the ballot. (It was 5-winner 
Bucklin, so voters had five votes in the first 
two ranks, and unlimited votes in the third 
rank). Next election, the excitement had worn off 
and there were fewer frivolous candidacies. And 
use Bucklin as a primary and runoff method, 
allowing write-ins in the runoff. Otherwise top 
two runoff. Want to make it more sophisticated? There are simple tweaks.

I'd like some credit for noticing that runoffs 
solve a lot of problems with voting systems, and 
for suggesting that using a better primary 
method, in particular, can greatly improve TTR. 
I've also pointed out, this seems to be original 
as well, that voter turnout in elections exerts a 
Range-like effect, by suppressing votes based on 
weak preferences. This has the strongest effect 
on a special election runoff; but it's also a way 
of validating preference strength in a runoff 
which is resolving a Condorcet conflict with a 
Range primary. If the Range preference strength 
estimates are accurate, the Range winner has a 
huge advantage! Whereas when I first suggested 
such a runoff, the response was immediate: the 
Condorcet winner will win, so why bother?

That's only in a fantasy world, where voters have 
fixed preferences, and the same voters will vote 
in the runoff who voted in the primary. Doesn't happen!

Again, I've been almost a voice crying in the 
wilderness, here, in pointing out the damage that 
is being done by the mindless campaign to replace 
Top Two Runoff with IRV, a campaign that only 
makes sense as a rather Machiavellian political 
strategy, thinking only of the eventual goal (PR) 
and the vulnerability of TTR (expense and inconvenience).

It's quite possible to argue with a straight face 
that IRV is better than Plurality in partisan 
elections. But the problem is that IRV isn't 
generally being implemented in that context, it 
has almost always been *nonpartisan* TTR elections, not Plurality.

And it's pretty certain that this is a step 
backwards. It took years of work to get as many 
TTR elections as we have, it was long considered 
a very important reform. It's the one that 
actually gives third parties the best shot at 
winning, certainly more than IRV. With a better 
primary method that avoids center squeeze, it 
becomes awfully close to ideal. So.... man the 
barricades!!! We have work to do.

It's been very enlightening reading the history 
of Bucklin in the U.S. It was all the rage for a 
few years, it was implemented, it appears in at 
least 55 municipalities, including some very big 
cities, such as, my latest finding, San 
Francisco. What happened? I don't know, it just 
seems to have disappeared. Politicians who lost 
elections fought back, that's clear, they blamed 
the method instead of their lack of support from 
voters. Preferential voting is easily seen as a 
threat by the two major parties; I just saw 
Fusion Voting shot down, two years ago I think it 
was, in Massachusetts. You can bet the Dems and 
Reps didn't support it! (And they said so!) 
Fusion Voting, of course, gives minor parties a 
toehold. Can't have that! Next thing, we might 
have a Socialist elected! Or one o them 
tree-hugging Greens! (Or one of them gun-toting 
Libertarians! Never mind that Libertarians supposedly renounce coercion!)

> > The *theory* of oscillation or endless regression based on
> > feedback between polls and voter decisions is just that, a
> > theory.
>
>What is the alternative? Do you think polls will settle on two
>frontrunners almost arbitrarily?

No, I think that the influence of polls on voters 
is damped. They simply won't respond as strongly 
as you think. And they don't depend exclusively 
on polls, they depend on a whole complex of 
communications, not only through media, but with 
neighbors, friends, co-workers. It's not the 
*polls* which settle on two frontrunners, but 
*society,* the polls merely reflect this, with 
more or less accuracy. Three-frontrunner 
situations are relatively rare, and are more 
likely to happen with minor nonpartisan elections.

In a strong two-party system, which we will have 
for the foreseeable future, it is a practical 
certainty that there will only be two, and the 
voters don't even have to think about it. Minor 
party supporters will know what they are facing. 
My recommendation to minor party supporters is to 
build party strength in two places: in minor 
elections, particularly nonpartisan ones, local 
offices, where party affiliation just isn't that 
important -- but the winner gains a kind of 
"bully pulpit" nevertheless -- and in coordinated 
political activity aside from elections. Work for 
Fusion Voting! It allows a party to build 
strength without spoiling elections, getting 
credit and ballot position for every vote it 
effectively awards to a major party. And it 
always has the option of running a third 
candidate, but it should really work, first, for 
voting system reform. Bucklin does it, even 
though Open Voting (Approval) is cheaper and 
simpler. Third parties will be better served by 
the explicit first preference vote; *that's* why 
they like IRV, they imagine that this will help 
them. It will, in some ways. But it makes it 
practically impossible to actually win, they have 
to go all the way from the bottom to the top *in 
one step.* TTR gives them an intermediate step 
that is more in reach: second place. They then 
have a real chance. No guarantee. Le Pen's 
supporters didn't have a chance, really, but they 
tried. They turned out, his runoff vote was a 
million votes higher than the primary vote. But 
the only reason he was in the runoff was center 
squeeze failure in the primary, Jospin was 
actually -- it's almost certain -- the Condorcet 
winner, possibly by almost as large a margin over 
Chirac as Chirac had over Le Pen (80%).

Better primary method! Bucklin will, I predict, 
avoid about half of the runoffs, even with a lot 
of candidates. It's possible that with a 
sophisticated enough method, even more runoffs 
could be avoided, but having the runoffs as a 
possibility does, in fact, guarantee, almost 
perfectly, majority support for a winner. (Want perfection? Asset Voting!!!!)

I see that Forest Simmons has proposed a Yes/No 
form of Approval with Delegated Voting. 
Essentially, as I recall, the voter designates on 
the ballot a proxy, to handle the "abstentions*. It's Asset.

Like all Asset methods, it sets up a deliberative 
process that isn't limited by number of ballots, 
possibly. (Simmon's method might require a single 
ballot, but that is no longer necessary once the 
effective electors have been drastically reduced 
in number, and if their votes are public.)

>The only alternative I can think of is that there would be no effective
>polls. And I suspect that would be just as bad as having polls that don't
>stabilize.

Suppression of information isn't a great idea for 
improving voter knowledge. Voter knowledge is 
*important,* and that includes context -- the 
position of the rest of the electorate -- as well 
as knowledge of the candidates themselves.

It doesn't matter if polls stabilize. The 
influence of polls on the voters is moderate. 
Some voters don't care at all about them, they 
will bullet vote no matter what you do, unless 
you coerce them. I find it fascinating that a mix 
of "strategic voters" -- read voting Approval 
style, and probably bullet voting -- and "sincere 
voters" -- voters who Range vote intermediate 
ratings -- seems to have lower Bayesian regret 
than either set alone. I really should nail that 
down, it's what I recall reading from an unpublished paper by Warren Smith.

See, all Range votes can -- and should -- be 
considered sincere. It's just that they are not 
full disclosure of preferences, the voter has 
chosen to express some and not others, to give 
more strength to some, as more important, than others.

Trying to force "sincerity" by forcing the 
expression of all preferences, or equating them 
all, is very, very misguided. Rather, the voter 
choices in terms of what preferences they express 
and what preferences they do not express is 
actually part of an efficient compromise process. 
I'm really not offended if a voter's choice to 
full rate A and B results in the election of B 
when the voter "really preferred A." The voter 
made the decision that the A>B preference wasn't 
important enough to express, compared to other 
considerations. The "other considerations" are 
the voter's view of the rest of the electorate, 
*and this is just what we do in real, direct 
negotiations.* It's part of the process whereby a 
good voting system simulates a deliberative 
process of finding an ideal compromise.

Want to judge real election success? Ask voters 
to ratify the election. If a majority don't 
ratify it, *it fails!* And then the measure of 
success is the ratification vote. The ideal is 
100%. It can be done with small groups, I've seen 
it and have described it. I've seen consensus 
organizations do a fairly good job. (But it's 
also well known that *demanding* consensus can be 
quite oppressive and results, sometimes, in a 
false appearance of consensus, or, alternatively, 
in a dominance of those benefited by the status 
quo, I've seen both of these as well.) It's 
probably impossible to get all the way there with 
large groups, but there is no particular level 
that is impossible. I'd bet that we could get, 
with good deliberative communications systems 
(FA/DP anyone?) to over 90%. *There is power in 
consensus, and people know that.*

(to be continued)