At 03:47 PM 5/15/2008, Juho wrote:
Here's one style of vulnerability of IRV that has not been discussed
that much.
In elections that have many candidates that represent small interest
groups, and when there are so many such candidates that practically
all voters have at least one such candidate that closely reflects
their views, then those candidates that appeal to many voters may not
be the first preferences of practically any voters.
Right. I think we need to remember how IRV was invented. First of
all, "IRV" is a neologism, a term coined in the 1990s by a friend of
Rob Richie and then promoted as the name for single-winner STV in the
U.S. For practical reasons, all the applications actually being
implemented are not full preference ballots. San Francisco, for
example, allows only three ranks, and with over *twenty* candidates
on the ballot for some city council seats, it's understandable that,
if people vote sincerely, a lot of people end up not influencing the
result. In fact, in 23 IRV elections in the U.S., held since 2004,
that went to runoff rounds (were not won by a majority in first
preference), only two (I don't think it was three) found a majority
of ballots. In San Francisco, by comparison, in one election there
was enough ballot data given to allow Bucklin analysis. Bucklin found
a majority in more elections. IRV loses votes, because they may be
buried. The very characteristics that FairVote touts as features are
the root cause of some of the problems.
Later-no-harm is incompatible with the discovery of a majority,
though I don't recall the details. And "core support," of course,
guarantees the weird scenario that Joho describes. IRV is sold as
allowing sincere voting of true first preference.
Okay, suppose voters take that literally, and write in the name of
the person they most trust for the office, sincerely, without
restrictions. Condorcet methods and Approval, etc., can handle this
without a problem. But IRV goes bonkers. If we have a three-candidate
election, with the three running neck-and-neck, but everyone prefers
one of the three as a second choice, it is quite obvious who the best
winner is. But with IRV, it's a tossup. One time out of three, the
Condorcet winner is disqualified in the first round, and the votes of
that winner are then distributed. What I've been seeing in real
elections is that these distributed votes are likely to be samples
from the same population. I.e., they will tend to be awarded to the
other candidates in the same proportion as with the population of all
those who voted for those other candidates. (This is a truly
remarkable finding, and I'd love it if someone would confirm it -- or
refute it. Data is available from San Francisco, Burlington, and
Cary, NC, I've looked at them. There may be some others.)
Anyway, we have a situation where IRV cannot predict who will win (it
depends on the second preference votes entirely, except that second
preference votes for the Condorcet winner are not seen or counted,
because that candidate was eliminated in the first round. So if this
is A, B, C, with the A:B pair, if counted with Bucklin or a Condorcet
method, would be a landslide for A, but then IRV may (one time out of
three) choose B or C. The difference is a few votes, radically
changing the result. This is why Yee diagrams for IRV are so chaotic
in this region. Bucklin shows the true situation: A is everyone's
second choice.
FairVote argues for the importance of "core support," a Criterion
that they made up because IRV supposedly satisfies it. (It doesn't,
actually.... but it would take very special circumstances for it to
fail.) The reality behind core support is that such support is
ordinarily necessary to (1) get on the ballot and (2) run a
successful campaign. If a candidate could actually win without
plurality core support, well, I'd be amazed. However, because of the
nature of IRV elections, "core support" does not mean that a
plurality (or even one voter) actually prefers the winner, first
preference, for the voters may not have voted their first preference,
since to do so is foolish, tosses away that vote. Systems which allow
full ranking, including a write-in, don't have this problem. Asset
Voting, in particular, would allow *totally* sincere voting, and it's
an extraordinarily simple method. Just count the votes..... Quota
required to win. Asset Voting was invented for multi-winner, but it
could easily be used for single-winner, and it totally avoids the
supposedly universal election paradoxes. By punting. Asset voting is
not a method that necessarily resolves in one ballot.
Anyway, yes, suppose that the voters are divided up into *many* small
factions. If people vote for their favorite, it can be quite quirky
and unpredictable what happens in the first round. So, easily, a
candidate who would be the second choice of nearly everyone could be
eliminated. To my mind, it is almost criminal to not consider all the
votes cast, which Condorcet methods, Approval, and Bucklin do.
(Bucklin is counted in rounds, and, if a jurisdiction didn't want to
spend the money, and a majority is found in the first round, later
votes might not be counted. But I'd say, if the voters put the effort
into voting, we should put the effort into counting. Whatever method is used.)
IRV is sold as replacing runoff elections, that was the very purpose
of the name invention. But IRV does not, apparently, behave at all
like top-two runoff. In one-third of the elections I examined, before
IRV, top-two runoff, in San Francisco and Cary, reversed the
first-round preference. The runner-up in the first round ended up
winning the runoff. But once IRV was implemented, there were no
reversals. Period. In every election I looked at, the leader in the
primary won the runoff, but, even more striking, *the runner-up in
the primary remained the runner-up in the runoff.*
We are starting to see why Plurality voting was used. It works as
well as IRV, almost always. Now, why did they have those runoffs in
the first place? Well, one time out of three, the election reversed.
That means one of two things: the former runner-up was really the
Condorcet winner, which plurality could not discover *because of lack
of sufficient "core support."* -- or, alternatively, differential
turnout explains the difference. Now that does not explain what
happened in Cary, NC, for the runoff was held with the general
election, the primary is the special election, i.e, not on the
general election ballot, held a month earlier. It might in San Francisco.
But if this reversal was a bad thing, it would have been much easier
and much cheaper to get rid of it entirely. Don't require a majority!
If we want to require a majority to support a candidate for the
candidate to win -- and that was the law in San Francisco -- then IRV
doesn't do it. It may be that *nothing* but a runoff can do it. Basic
democratic process doesn't use elimination in runoffs; it requires a
majority and if there is no majority, there may be voluntary
withdrawals or additional nominations, essentially a new election is
held. But as a compromise, whatever method is used for the primary,
if the results don't show a majority of support, then a runoff should
be held -- if one wants the winner to have such support.
In selling IRV, all this was, of course, not mentioned. It's not
clear that anyone even understood or expected it. The ballot
proposition in San Francisco that started up RCV there was
deceptively sold. Di they know it was deceptive? I think it is likely
that some knew. San Francisco had election code that required that
the winner receive a majority of ballots cast in the election, or
there would be a runoff. "Instant runoff voting," I'm sure, looked
very attractive. Have the runoff without holding a separate poll! The
voter information pamphlet explicitly said that the winner would
still be required to gain a majority of votes. But the actual
proposition struck the majority requirement from the code. If they
had implemented preferential voting without striking the requirement,
then the claim would have been true.
Why did they strike it? Someone had to understand that IRV, with
plurality winners, was violating the old provision. Otherwise they
would just have left it in place.
Now, some IRV supporters have seen this argument, and their response
has been, "But they wanted to get rid of the expensive runoffs."
Sure. Who wouldn't? But they *also* wanted majority winners, and the
"majority of last round votes" that IRV reports is *not* a majority
winner by any traditional definition. They had to get rid of that
legal provision, which meant that they were getting rid of the
majority requirement, not continuing to "require" it, as the voter
information pamplet claimed. If they wanted to avoid the expense,
they could have obtained the *same* election results as IRV, almost
certainly, by just dropping the requirement. Or, better, they could
have implemented Approval or, probably even better in terms of
allowing first preferences to be expressed, Bucklin. No big expense.
More likely to find a majority. And then, in the situations where a
majority is not found, *you hold a runoff*.
The characteristics of top-two runoff, or of other methods used with
majority-failure such as Bucklin or Range or Approval, have not been
much studied. (With Range, approval cutoff is needed; ballot
instructions could state that any vote of 50% or higher would be
considered approval for this purpose. Approval or Bucklin, of course,
explicitly approve.) What I've realized is that the runoff is a test
of preference strength. If the voters have a strong preference for
one of the choices over the other, they will preferentially turn out
to vote, compared to those with a weak preference. This would, then,
improve the average voter satisfaction with the result. It's a
Range-like result, and could explain the popularity of top-two
runoff. If this is true, then IRV is definitely a step backwards, and
when it is actually being used (i.e., when runoff rounds are
counted), it is producing worse results about a third of the time.
FairVote has always tried to frame the debate as IRV vs. Plurality,
but actual major victories have been in places where the heavy cost
of implementation could arguably be saved by avoiding runoffs. So the
real comparison, there, would be between IRV and top-two. IRV is
being used in some places, and is still being called IRV, with only a
two-rank ballot. It's presumed that voters know who the frontrunners
are, and so they will presumably vote for one of them in second rank.
I am finding such twisted logic in the pro-IRV propaganda, it amazes
me. The opponents of IRV in San Francisco were apparently completely
flummoxed. Nobody noticed the bait-and-switch. You want majority
winners? Great idea! But you don't like the cost? Hey, we can sell
you this nice little baby! A bit expensive, but you'll save every
election year, lots of money, so it will eventually pay for itself.
Find Majorities Without Runoff Elections!
The fine print: Majority not required. The "majority found" is found
by pretending that ballots not containing a vote for the two
frontrunners aren't important. One could get that result much more
easily. Just count a plurality poll with the IRV rules, in rounds.
Eliminate, in each around, candidates with the least number of votes.
Last round, two candidates, the one with a majority of remaining
ballots wins. Presto! A majority!
You should have seen the pro-IRV forces complaining about efforts to
clean up the IRV article on Wikipedia, so that it was less of a
propaganda piece. Edit wars over what might seem like trivial
differences. About the use of the word majority. The existing
language used majority over and over again to mean "majority after
exhausted ballots are set aside." I'd try to specific it, to make it
more accurate. "Pendantic! Confusing! Too much detail!"
And then, most recently, yesterday, actually, I notice that the
Robert's Rules' (RRONR) "recommendation" of "IRV" wasn't. Even less
than I'd been willing to acknowledge. Robert's Rules describes -- not
recommends, they clearly dislike it -- preferential voting. Not IRV.
But they do describe, *as an example of preferential voting*,
sequential elimination. Then RRONR also describe center squeeze as a
problem with sequential elimination. The IRV proponents want the
article to give the Robert's Rules "description" -- they abandoned
efforts to keep "recommendation," it was really unsupportable -- but
not, of course, the importance of center squeeze, or any example of
what Robert's Rules might mean by "preferential voting," in general,
for the text that may be interpreted as a recommendation in narrow
circumstances is actually of preferential ballot without specifying
how it is analyzed, so the obvious question for a reader would be,
what else is there? Try to mention Bucklin, why, that's "promoting"
Bucklin. Edit warring has begun over this, at least to some degree.
What did I find? Well, Robert's Rules, in describing sequential
elimination was not crystal clear what "finding a majority" meant.
One with a background in parliamentary procedure would, I think, get
it right. It means a majority of ballots cast, and there is,
elsewhere in Robert's Rules, some specification of this. But at the
very end of the section on preferential voting, there is a mention
that voters should be educated in the use of the method, something
like, "otherwise they may think that voting an additional preference
could hurt the chances of their favorite, and this could cause the
election to have to be repeated."
In other words, they are assuming that "majority" means true
majority. What they are describing is a method which requires a
majority or there is a runoff. Period. But then --search the internet
for Robert's Rules of Order and Instant Runoff Voting -- they are
using this "recommendation" to use IRV in a way that is quite
contrary from Robert's Rules. The Robert's Rules form of preferential
voting avoids runoffs *sometimes*. When there are many candidates, it
turns out, almost never. Unless you drop the majority requirement,
and we are circling back to the beginning. Bait and switch says it.
Majority-required voting systems perform much better than many
election methods people suspect. They are actually hybrid, closer to
deliberative process. There is no way to guarantee a majority unless
you constrain and coerce the voters, which they actually do in
Australia. (You *must* vote, it is against the law not to, and you
*must* rank all the candidates, or else your ballot is spoiled and
will not be counted.)
And if voting is as Robert's Rules actually suggests, there isn't
even a guarantee that an election will complete in two ballots, for
Robert's Rules, explicitly, dislikes any sort of involuntary
elimination. New election is what they say. Until the electorate gets
it together to vote a majority. When Robert's Rules says that
"preferential voting is fairer than election by a Plurality," that is
because they have described a method that still requires a majority.
It simply finds one more efficiently, perhaps. Sometimes a runoff
will be avoided. But with IRV, as distinct from other forms of
preferential voting, not very often at all.
And it is not clear how often we would see majority failure if the
method were, say, Bucklin. San Francisco had such high numbers of
candidates because they had top-two runoff, and voters could safely
vote for their favorite in the first round. (Strategy would say
otherwise, but most voters simply vote honestly when they think they
can). So a better election method, and top-two is better than simple
Plurality, is likely to encourage more candidates. These were
nonpartisan elections, so the party system was largely moot. And with
lots of candidates, no voting system, itself, can guarantee a
majority. But some can find it better than others. Counting all the
votes is, in my opinion, always a good idea!
----
Election-Methods mailing list - see http://electorama.com/em for list info