At 07:55 PM 1/7/2010, Kathy Dopp wrote:
I've answered that question on this list before and Abd ul also answered it.
There are *many* good alternative voting methods that do solve the
spoiler problem, are monotonic, and elect majority winners and are
precinct summable. I don't know of any alternative voting methods as
bad as IRV/STV (although there must be one somewhere), so I would
probably support almost any alternative method that lacks the
multitude of flaws that IRV/STV have. Abd ul has convinced me that
regular top-two runoffs are good too.
Top two runoff, of course, shares the center-squeeze problem that IRV
suffers from, but, interestingly enough, that problem may not be as
serious if write-in votes are allowed in the runoff, as they are by
default in California and perhaps some other jurisdictions.
Not widely known is the fact that the spoiler effect is connected
with partisan elections. IRV in nonpartisan elections seems to
reproduce Plurality closely. With top-two runoff in nonpartisan
elections, the runner up in the primary wins roughly one-third of the
time, per a study by FairVote. However, with IRV, these "comeback
elections" hardly ever occur.
IRV does fix what I call the "first-order spoiler effect," where a
minor party candidate draws away votes from a major party candidate,
causing the less popular of the major party candidates to win.
However, as a recent Burlington election shows, IRV can award victory
to a candidate who would, by the votes expressed on the ballots, lose
in a direct contest to an eliminated candidate, because of the
peculiar significance that IRV gives to the first choice. It is
entirely possible that without the promise of IRV as a fair system,
the same configuration of candidates would not exist, and the more
popular candidate would have won.
There is another system that uses the same ranked choice ballot as
IRV, but that is probably much better at handling the center squeeze
situation, and that was at one time widely used in the U.S. (Far more
widely than the recent IRV fad.) That's Bucklin voting. It could be
called "Instant Runoff Approval." It's much easier to canvass than
IRV, the totals for each rank are simply collected from each
precinct, and the handling, if there is no majority in first
preference, is simple addition; the difference from Approval is that
the approvals are ranked, so additional approvals are only considered
if nobody gets a majority in first preferences.
Why was Bucklin rejected? Partly, it may have been for similar
reasons to the prior and present rejections of IRV. IRV has been sold
on a false promise: to find majorities without runoff elections.
Bucklin was sold in the same way, and it fails to find majorities
reliably for the same reason as IRV fails: people don't rank enough
candidates. This has been a known problem with STV for more than a
century, and whenever a candidate doesn't get a majority in first
preferences, it is *normal* for IRV to never find a majority even
after vote transfers, the IRV "majority" is a faux majority, a new
invention, a "last round majority," based on an entirely new concept
of a majority that isn't the traditional one: a majority of ballots
case. Bucklin is based on "majority of ballots cast," as are standard
repeated ballot systems. All ballots are considered.
Bucklin, because of the lack of "candidate eliminations," which
really means ballot eliminations in actual practice, is more
efficient at finding majorities, however, because it will find votes
concealed under votes for a leading candidate. We tend to think of
partisan elections, for some reason, where a voter for one of the top
two candidates would one rarely also approve one of the other top two.
But in nonpartisan elections, which are the vast majority of recent
IRV applications, a supporter of one of the leading candidates might
well express support for another leading candidate. Not highly
partisan supporters, but general voters. IRV conceals these votes,
Bucklin finds and counts them.
The error with prior implementations was in the false promise: when
it was realized that Bucklin wasn't actually finding majorities in
some of the places where it was used, because of enough voters doing
the traditional vote-for-one thing, Bucklin was dumped entirely. It,
as also happened with IRV in some prior situations, it was replaced
with top-two runoff, which usually finds a majority. Instead, Bucklin
should have been used as a method of avoiding unnecessary runoffs.
I like to think of this as the voter's strategy. With Bucklin, I will
unconditionally vote in first preference for my favorite. There is
never a reason not to. (With IRV, there can be circumstances where
voting for your favorite will turn out to be foolish, it can cause a
much worse outcome, it can even cause your favorite to lose. That's
what non-monotonicity means.)
Then, as to adding other, lower-ranked approvals, the question I'd
ask myself: Would I prefer this candidate to the delay and expense of
a real runoff election? If so, my preference is not very strong, and
I'll add another approval or more approvals. (Another reason for an
additional approval might be to try to get a second better candidate
into a runoff, thus taking less risk of a poor outcome.)
My guess is that using Bucklin would eliminate more than half of
runoff elections, and with some good runoff algorithms, based on real
election history, it could eliminate even more. (For example, suppose
that Candidate A gets 49.9%, and no other candidate gets more than
25%. It is so unlikely that a runoff would elect other than A that
it's not worth the effort. Sometimes a cutoff is set at 40%. But that
is too simplistic, what matters more than the absolute percentage is
the lead that must be overcome to shift the result. 40% A, 39% B, 21%
C is not at all decisive between A and B.)
Using Bucklin for party convention elections, where holding a runoff
isn't impossible, would be more efficient. Bucklin with top-two
runoff if a majority is not found (or some similar criterion in
intention), and with write-in votes allowed, doesn't suffer from the
spoiler effect, and would quite likely settle on a majority winner
very quickly. If the convention-goers want to be sticklers and only
vote for their favorites, they only punish themselves....
Range voting (Score voting) is even better, but should include
approval indication, or it can, like other single-ballot methods,
come up with a candidate who won't be supported by a majority. Range
voting with runoffs when needed is actually an ideal voting system.
Bucklin, however, has a long history, was very popular when used, and
is a bit simpler to vote, probably.
It was called American Preferential Voting when used, the political
scientists of the time (early 20th century) were very aware of single
transferable vote, and considered it inferior when used for
single-winner elections.
(Single Transferable Vote is considerably better when used for
multiwinner elections, though there are better methods still, for
sure. True multiwinner STV has been rejected after use in the U.S.,
but not for good reasons. It was rejected because it resulted in fair
representation for minority groups. I urge election activists opposed
to IRV not to jump for the temptation of praising those rejections as
wise. They weren't. They were racist and prejudiced in other ways
against the fair choices of the voters. In Ann Arbor, MI, IRV was
rejected on arguments similar, apparently, to some of those being
advanced in Burlington now: it deprived the Republican of his
"rightful" victory over the Democrat, which had been previously
happening because of vote splitting in a college town between the
Democratic candidates and the Human Rights Party candidates. However,
the situation in Burlington is pretty different: the problem there is
that there are three major parties there, and IRV does very poorly in
that context. It worked in Ann Arbor, and, for that reason, a
referendum on it was scheduled for when the students were on break,
mostly out of town!)
----
Election-Methods mailing list - see http://electorama.com/em for list info
