[EM] IRV vs Plurality

[Abd ul-Rahman Lomax]

At 02:14 AM 1/13/2010, robert bristow-johnson wrote:

>>For instance in Aspen CO's most recent
>>IRV election, if 75 *fewer* voters had voted for one candidate that
>>candidate would have won.
>
>that's a pathology (and sounds like a worse one than Burlington VT in
>2009).  so let's get rid of it.  but *throwing* *away* information by
>reverting to the single-vote ballot from the ranked ballot doesn't
>make things better.  it makes it worse for the majority.

The alternative generally is not pure plurality, it's some kind of 
runoff voting. The candidate must gain a certain plurality or a 
majority in the primary, or a runoff is held. The method for 
determining who gets on the runoff ballot is not necessarily the 
best; IRV flops because it's possible that the majority preference 
(pairwise) is third or even lower in first rank, and is eliminated.

In the real examples I've seen, by the way, the majority preference 
was only a little behind the runner-up when eliminated. FairVote, in 
defending IRV against the center squeeze objection, makes it seem 
like the majority preference doesn't have "core support." Really, and 
usually, the "core support" is roughly equivalent to the runner-up. 
And "core support" and Later-No-Harm are a Bad Idea, for they reward 
partisan thinking, allowing compromise only after the first choice is 
dead, eliminated, impossible.

Now, suppose I have two choices: I can run my ballot as IRV, in which 
case my second choice doesn't count until my first choice is actually 
taken out back and shot, er, my first choice is eliminated, or I can 
run by ballot as Bucklin, in which case my first choice is the only 
one counted in the first round, but my lower ranked choices are 
counted in the round I've indicated. I could even leave an 
intermediate rank empty, so my second choice is only counted in the 
final round. Or second choices, plural. (Traditional Duluth Bucklin 
allowed multiple votes in the last rank but not in earlier ones; I'd 
allow them in all ranks, no reason not to, and it makes the method 
more flexible, my guess it would slightly improve result quality, but 
it would definitely decrease spoiled ballots.)

The voter wanting Later No Harm protection would indicate on the 
ballot that it's an IRV ballot. This means that lower-ranked votes on 
that ballot would never count against the election of a higher-ranked 
candidate. They would not be counted until the candidate were 
eliminated. So the ballot would be run as a Bucklin ballot, with only 
one vote on it, the first rank. Then, if no Bucklin majority appeared 
(which is a true majority), there would be an instant runoff. The 
votes from all ranks on Bucklin ballots would stand. They would not 
be eliminated. But the ballots from the IRV voters would be treated 
as IRV ballots, and the candidate in first preference on them, with 
the lowest vote count so far, would have those votes struck, in 
effect, and the highest ranked votes standing would be counted. Note 
that if all voters vote IRV ballots, the election becomes pure IRV 
and if all voters vote Bucklin ballots, the election becomes pure Bucklin.

In all cases, single winner election, the only vote on a ballot that 
actually counts is a vote for the winner. All other votes are moot, 
in the end, they could be struck and the result would not change.

If the Later No Harm argument were truly valid, surely most voters 
would want to vote that way. But I sure wouldn't, because preventing 
"harm" to my candidate from my second rank vote also prevents "help" 
for my candidate from other voters who would vote for my favorite in 
second rank.

Rather, to understand the situation, we should encourage people to 
look at repeated balloting as a very old and justly respected voting 
method. This is repeated balloting with no eliminations, it provides 
complete freedom to the voters, they can vote for their favorite 
without regard for what others think as long as they want. But if 
they keep doing that, and many keep doing that, the election won't 
resolve and they have to keep casting ballots, it's a form of 
selfishness that reaches a natural boundary. So what happens, in 
effect, is that voters lower their expectations and accept a result 
that they personally consider less than idea, but adequate and 
necessary in order to complete the election with a majority.

It's totally obvious that this method could be made a bit more 
efficient by allowing multiple approvals. Repeated approval elections 
with no eliminations is probably somewhat more efficient than 
repeated elections allowing only a vote for one candidate, as voters 
will begin, perhaps, with some level of minor compromise, and they 
can then lower the approval cutoff as necessary. Or if they have a 
clear favorite, that's how they start until they have more 
information about what is practical.

Bucklin simulates repeated balloting with a declining approval 
cutoff. An initial single preference vote is complemented by an 
additional approval with second rank, and another with third rank. 
(If a voter wants additional LNH protection, the voter can skip a 
rank, adding multiple approvals in last rank, say). A similar process 
can be done with Range ballots, and it's possible that it would 
encourage better provision of information by the voters.

>>2. they rank all candidates possible to rank on the ballot, say 3, but
>>there are more than 4 candidates in the contest with supporters.
>
>the number of ranking columns should equal the number of candidates
>that legally get on the ballot plus one, the write-in candidate.  (i
>guess, if some voter wants to write-in and rank two write-ins,
>they're sorta screwed, but i don't see that as a practical right to
>grant to the electorate; unlimited number of write-ins?)  it should
>never be the case, unless you're in Italy or somewhere with 29
>parties and candidates.  even with 29, they should be able to print
>30 columns of ovals to fill in.  the ballot would look messy, but
>there is no reason it can't be done.

There is plenty of reason that it's a waste of time. Most voters 
don't have sufficient information to intelligently rank 23 
candidates, as there would be in San Francisco. So you get 
truncation, and truncation leads to majority failure. That means that 
a majority of voters, effectively, voted against the candidate by 
voting for someone else and not the candidate.

I personally prefer to vote only for candidates that I know something 
about. Hence I really am voting against candidates when I vote for a 
different one, and in a ranked system, if I don't rank a candidate, 
that's a vote against the candidate. I do not want to see that 
candidate elected unless I get a chance to reconsider and vote for 
the candidate in a runoff. Period! (But I'll accept the candidate if 
a majority do voluntarily vote for the candidate.

Note that from my study of actual IRV ballots, it's fairly plain that 
many voters don't realize that by ranking a candidate, they are 
actually voting for that candidate. With IRV, the only way to not 
vote for a candidate is to not rank the candidate at all (unless you 
rank all candidates, and rank that one last, assuming that is 
possible, and if there are write-in candidates, you just, by doing 
this, voted for the candidate against even what turns out to be a 
popular write-in candidate -- you didn't realize this! -- and would 
you prefer an unknown write-in who is popular enough to almost win to 
the jerk you know is a jerk? I would. And that's why I would not vote 
for a candidate where I couldn't accept the result of my vote, where 
I might regret it.

I will not vote for the lesser of two evils because both are evil. 
But I will vote for a mediocre candidate or perhaps a "mixed" 
candidate, some good and some evil, over an evil one. If the system 
only presents me with two true evils, I might be better off not voting at all.

>   now if there were 1000 parties
>and candidates, that would be a pretty impossible ranked-order ballot
>that includes everyone, but i consider number 2. to be a very
>unlikely issue to worry about.  if the number of candidates greatly
>exceed the number of preferences that we can afford real-estate on
>the ballot, then it should be made clear the voter that everyone not
>ranked is tied for last place.  big deal.

It's actually a problem with three ranks. Dodgson realized that the 
majority of voters in his time, and it may still be true, only had 
enough information to vote for a favorite. Hence Plurality makes 
sense, to a degree. It levels the playing field, in a way. But there 
are obvious improvements possible. Dodgson's was Asset Voting, which 
is totally brilliant and it's a shame that it was overlooked for more 
than a century, in spite of being published. By the man who is now 
recognized as possibly the greatest expert on voting systems from the 
nineteenth century.

>>The only way IRV voters get to participate in the final counting round
>>is if they vote for one of the candidates that happens to survive to
>>the final round and since sometimes IRV eliminates the most popular
>>candidate like it did in Burlington, that can be difficult to
>>determine.
>
>what's difficult to determine?  who the most popular candidate was in
>2006 or 2009?  that's simple!  but before, with the old ballot, you
>did not collect as much information from the voters, and then you
>wouldn't even know it, if the election rejected the most popular
>candidate.  why does not collecting the information give us a better
>idea of what the electorate wants over collecting the information?

The point is that if you collect the information, and if it is useful 
information, you should use it! Single-winner IRV wastes a frightful 
percentage of the information collected. Consider this: a minor 
candidate might actually have broad support, but only as a second or 
third or lower choice. Still approved, in fact, it could be possible 
that a supermajority of voters would be quite happy with this candidate.

It is theoretically possible that if the true preferences of voters 
were collected, this "minor candidate" would win against every other 
candidate by a complete landslide. I.e., if there are ten candidates, 
say, and suppose that all candidates have their own factions where 
they are preferred, roughly equally. Under this condition, it's kind 
of a tossup who gets eliminated. But there is tendency for centrists, 
candidates with very broad appeal, to not be the absolute first 
choice, and center squeeze is encouraged, as the professor in a tape 
Clay posted to the Election Science Foundation list yesterday 
covered, by political considerations and where candidates place 
themselves. Center squeeze, with a political spectrum and more than 
two parties, is not rare, it's actually created by the system. So, 
good chance, this special compromise candidate, everyone's first or 
second choice, is eliminated with IRV, and then what happens? 
Toss-up, really, and it's possible that ninety percent of the 
electorate is really pissed. But if all the votes can be counted, and 
not just first preference votes, or first standing votes after 
eliminations, it becomes obvoius who should win. It's actually a 
landslide with Bucklin, demonstrating the true popularity of the 
candidate. As I described it, the election would be unanimous.

But FairVote would be complaining that Later No Harm was violated. 
Ninety percent of the voters, it is implied by the logic they use, 
will be pissed, since they can tell that obviously their second 
choice vote "harmed" their favorite. Really? Was that a harm? And is 
it harm to the candidates we should be considering, or harm to the 
society, to all of us or to, at least, the majority?

>it seems to me that what the anti-ranked ballot people want is the
>luxury of not having to be pinned down on Election Day as to whom
>they'd vote for as a fallback candidate.  is it that you'd rather
>have the ability to change your mind about your fallback candidate in
>the period between Election Day and Runoff Day?

That "luxury" is well known to be important. Absolutely. I want that 
right. I want to see how the candidates behave once they know that 
they might actually win. Comeback elections occur in one-third of 
nonpartisan runoffs. They do not occur with IRV. I know that FairVote 
activists argue that runoffs are Bad because voter turnout can be 
lower, but, in fact, when voters really care, they turn out. Low 
turnout in runoff elections would generally mean that voters are 
adequately satisfied with the result, or equally dissatisfied, 
roughly. They don't have the motivation to turn out. Perhaps they 
voted for a minor candidate in the primary and it's Tweedledum and 
Tweedledee for them as to the runoff. On the other hand, they get a 
chance to reconsider.

In some elections, I only know enough to vote for my favorite. If my 
favorite doesn't win, then I very much appreciate the opportunity to 
consider a reduced set of candidates, and if I then care, I'll turn 
out to vote in the runoff. But many voters, practically by 
definition, won't care, so they won't turn out, and that is perfectly 
okay. It raises the quality of results!

Only if some sector of the electorate is specially disadvantaged as 
to the difficulty of voting is low turnout a possible problem. Being 
poor, as one example, isn't much of a disadvantage, unless it's 
associated with ignorance and lack of general motivation, which 
doesn't mean that, for this particular voter, turning out to vote is 
actually going to do the voter some good! Rather, such voters may be 
more easily influenced by deceptive propaganda....

>i don't get it.  why is requiring voters to make up their mind by
>Election Day (even about second choices) such a violation of their
>rights as a voter?

Does Robert know that he's arguing against basic democratic process? 
People have known about IRV for a long time. Robert's Rules of Order 
even recommends Preferential Voting for situations "where it is 
considered impractical to hold the normal repeated balloting." But 
they still require a true majority, and they also note the problem 
that Robert seems so eager to discount: preferential voting, of the 
kind they describe, and they obviously are aware that there are other 
methods that might not do this, can pass over a compromise candidate 
that would be found in repeated balloting as well, and deprives the 
voters of the additional information that they might gain as a result 
of the first poll, this information helps them to focus and make 
better decisions the next time, to resolve the election with a 
satisfactory result. Supported by at least a majority of those voting 
in the final election.


>>IRV/STV is fundamentally unfair because a large group of persons whose
>>first choice loses, never has their 2nd choice counted,
>
>only if they don't mark their 2nd choice.

No. Damn it, I wish people would make more effort to understand the 
scope of the problem! Robert, I expect you to say Ooops!

I vote for A and for B as a second choice. In the first round of IRV 
counting, B is eliminated. My second choice is never counted, because 
my first choice, A, was still standing when B was eliminated.

That's the effect of the Later-No-Harm compliance of IRV, and a 
simple example of the destruction LNH wreaks. It doesn't matter if B 
is the Condorcet winner by a landslide, if B is the second choice of 
every voter, with vastly higher satisfaction overall if elected, that 
first preference vote, to the FairVote activists and anyone who 
drinks their Kool-Aid, is sacrosanct.

Even if a "first preference vote" is with very low or even absent 
preference strength. Try to equal rank in first preference, or any 
preference, and you know what the IRV counting rules will do, don't 
you. Tell me why the system throws that information out? It has an 
obvious meaning! It means that the voter is claiming to be equally 
satisfied by the election of either candidate!

IRV with equal ranking allowed would be a much better method; 
basically it would be Approval voting in first round, or later 
rounds. But Bucklin is more straightforward and far easier to 
understand. The elimination algorithm of IRV is just plain weird and 
chaotic, take a look at those Yee diagrams. It looks simple. It is 
far from simple.

>   otherwise, that's
>certainly not true.

Well, Robert has a golden opportunity now. He can recognize that he 
can be certain about a thing and be dead wrong. That is 
extraordinarily valuable! Take the opportunity, Robert, I promise you 
that the pain, if there is even any, will be transient, and, in the 
long run, you'll even be happy you made the mistake.

>   ("unfair" is subjective, but the latter half is
>untrue for voters who mark a 2nd choice with 3 credible candidates.
>or do you mean that if they vote for the 2 biggest losers, their 3rd
>or 4th choice that *does* count in the final round isn't their 2nd
>choice so it's unfair?)  IRV is screwed up, but you still have made
>no case that the weird pathologies of IRV are worse than the widely
>known pathologies of plurality in a multi-party or multi-candidate
>context.

The original comment was quite correct (but, yes, "unfair" is 
subjective unless qualified, but, indeed, it could be qualified with 
reasonable fairness criteria). Suppose there are three credible 
candidates, we'll call them A, B, and C. No, the example given isn't 
what's being talked about. The statement is literally true. Except by 
op-scan ballot images that show those second choice votes, they are 
literally never counted. With hand counting, they would never be 
counted. I think you'd better review how IRV works! When a candidate 
is eliminated, votes for that candidate are not counted from that 
point on, and if they haven't been counted before that, they are 
never counted. So with IRV, it is quite clearly true that it's 
possible that every single voter votes for a candidate, casts a vote 
for a candidate, and the candidate is not elected. You may think 
that's fair, but I don't. When that happens, not every vote was 
counted and my slogan is

Count All the Votes.

(And, of course, use them!)

To make this crystal clear, let's suppose that A, B, and C are 
relatively equal in popularity. And we can assume no complications 
with additional candidates, there are only three. So I vote A>B. But 
because B, it turns out, gets the lowest vote count, maybe it's, say, 
32%, 34%, 34%, or it's even 33%, 33%, 34% but B's is a vote less or 
loses the coin toss, B is eliminated. My second rank vote for B is 
never counted. And if the votes were this:

33% A>B
33% B
33% C>B

Consider what this means. It means that every single voter considered 
B acceptable, because every single voter voted for B. The B voters 
are in the middle, say, and really don't care which of A or C are 
elected, they are both equally unpalatable to them. They are 
centrists, and they recognize the harm to the country if an extremist 
is elected, which A and C are from their point of view.

But because of a small vote gap or a coin toss, B is eliminated, and 
only one-third of the actual votes for B are ever counted and 
considered. Now, tell me, is this fair. What do you think, really? 
Stand aside from everything you have thought or said before, does 
that sound fair?

Sure, this is an extreme example. But in real life, the margin of 
acceptability of the B candidate can be quite high. Montrose in 
Burlington wasn't just the Condorcet winner by some small margin, it 
was substantial. Montrose would have won under Bucklin, I have no 
doubt, Bucklin handled elections like this, historically, with ease. 
Montrose would likely have been the Range winner, and was almost 
certainly the social utility maximizer. (Which is slightly different 
from being the Range winner because voters will vote strategically to 
some degree or other).

>>unlike some other voters. It's a highly inequitable method.
>
>not as inequitable as plurality.   why is reducing the information
>obtained from the electorate more equitable?

It's equitable because it is equal. What's inequitable is collecting 
a lot of information from voters, then only using some of it in a way 
that effectively disenfranchises some.

And, please, remember this. The purpose of elections is to pick a 
winner. IRV in nonpartisan elections so frequently picks the 
candidate who is the first preference winner (by plurality plus, of 
course, when it happens by a majority, which is the norm with small 
candidate counts) that we might as well say "always." We can assume 
that voters, with IRV, under present conditions (not necessarily 
after they get wise to how the method works), are voting sincerely. 
So with Plurality, they will either vote sincerely and get the same 
result at much lower cost and certainly an easier voting decision, or 
they will modify their vote according to their understanding of 
what's practical.

Let me explain how I'd do this. With plurality, if I have an 
understanding of the situation and there are more than two viable 
candidates, as far as I know, I will probably vote for my favorite 
from among all those whom I consider might have a chance. With no 
such knowledge, I will probably just vote for my favorite without 
restriction. Why not?

People don't go through complicated game theory analysis to vote. In 
nonpartisan elections, they don't have the party endorsement to go 
by, so what do you think they do? They vote for their favorite!!!! 
And only if their favorite is truly minor, and they know that their 
favorite can't win, and we must assume that they will usually be 
right about that, will they vote strategically, for a favored 
candidate from among the realistic possibilities. The result, 
basically, Plurality will produce the same results. Period. With 
nonpartisan elections.

Now, when was the last time you heard the words "nonpartisan 
elections" from FairVote? Yet that is exactly where they have been 
selling IRV, the place where it is utterly and completely useless and 
quite possibly harmful. Burlington was partisan, so Burlington showed 
how, with partisan elections, the spoiler effect can be avoided, 
perhaps, but then IRV makes the wrong choice as to the winner.

And all those Wright>Montrose votes: they weren't counted for 
Montrose. They were moot, because Montrose was eliminated first, before Wright.