[EM] IRV failure modes, vote splitting

[Greg Dennis]

I think we may need a better theory to explain why IRV elects the Condorcet
candidate with near-perfect frequency in practice. We're now somewhere
around 215 IRV elections in the US since San Francisco started in 2004, and
Burlington 2009 is still the only case. That includes many highly
competitive, crowded fields with more than 2 strong candidates, including
the Minneapolis mayoral race last year and the San Francisco Mayoral race
and Maine Democratic gubernatorial primary this year. Probably a decent
research project there for the taking.

On Thu, Dec 20, 2018, 2:33 PM Richard Lung <voting at ukscientists.com wrote:

>
> Following is repetitive but not everyone will have heard it. The classic
> judgment is by Simon Laplace (backed by JFS Ross), who decided in favor of
> Borda method over Condorcet method, because ranks are not of equal
> importance. Condorcet has since been weighted (and not arbitrarily like
> Borda). But Condorcet appears to be generally a cross-referencing of some
> given voting method, for internal consistency, not a method in itself.
> IRV/RCV/Alternative Vote suffers from premature exclusion of candidates.
> Perhaps because USA is two-party polarised, the Burlington result has not
> repeated yet.
> An IRV drawback is also potentially true of STV, especially in small
> multi-member constituencies. In Ireland in 2016, two brothers got
> supporters to allocate extra first preferences to younger brother, to
> ensure not excluded on first elimination count. They were elected the two
> most popular.
>
> My invention does not eliminate candidates, tho it uses exclusion counts,
> as well as election counts: "FAB STV: Four Averages Binomial Single
> Transferable Vote" maximises ranked choice information by counting all
> ranked choices, thru keep values (weighting) of preference and unpreference
> (reverse preference) and, in both cases, abstention (possibly making a
> quota for leaving a seat empty).
> Official elections are generally "uninomial" (preference-only)
> The difference of FAB STV from the voters viewpoint is that they have
> Bidirectional preference (two-way choice).
> from
> Richard Lung.
>
>
> On 20/12/2018 05:15, robert bristow-johnson wrote:
>
>
>
> ---------------------------- Original Message ----------------------------
> Subject: [EM] IRV failure modes, vote splitting
> From: "John" <john.r.moser at gmail.com> <john.r.moser at gmail.com>
> Date: Wed, December 19, 2018 10:34 pm
> To: election-methods at electorama.com
> --------------------------------------------------------------------------
>
> > [I'm not subscribed, so please CC me on responses]
> >
> > After looking at Burlington, VT 2009, I've started to ask some questions
> > about IRV.
>
> what were you looking at?  this, from Warren Smith?:
>
>    https://rangevoting.org/Burlington.html
>
> It's a good analysis.  don't let this excellent analysis of the Burlington
> 2009 get conflated with all of Warren's claims about Score Voting or
> Approval Voting.
>
> also (other people on this list know this), i am a Burlington resident (i
> happen to live in Bernie's ward), politically active, and was there in
> 2009.  at the time Andy Montroll was a friend of mine.
>
> > I strongly prefer Tideman's Alternative Smith.
>
> personally, i think that Tideman Ranked Pairs (using margins) is just
> better.  and it's simpler.
>
>
> > IRV eventually reduces to a three-way race, and I see some interesting
> > things occur there.
> >
> > In Plurality, you can add a candidate on the same ideological end as your
> > opponent—if your opponent is more-liberal than you, then add another
> > more-liberal opponent—and you split the vote. This lets you win a close
> > race when you can't get majority.
> >
> > In IRV, you also need to add a candidate, and the same rules kind of
> work.
> > By adding a conservative to a two-liberal race, you siphon votes from
> your
> > moderate (winning) opponent.
>
> well, yes, but most of the moderate liberal voters had marked the more
> liberal as their second choice and also the reciprocal.
>
>
> > That happened in 2009: Bob Kiss loses; add Kurt Wright and Kurt Wright
> > wins plurality, Montroll has the fewest votes and is eliminated, Kiss
> beats
> > Wright and wins.
>
> Andy had the fewest 1st-choice votes of the three, but he was the
> 2nd-choice of most of Kurt's voters and nearly all of Bob's voters.
>
>
> >
> > Interesting thoughts.
> >
> > In the three-way race, it looks like Plurality without Majority elects
> the
> > Condorcet loser. Kurt Wright lost to both Kiss and Montroll,
>
> in a pair-wise comparison, yes.  loses to Andy by 929 votes and lost to
> Bob by 252 votes (out of nearly 9000).
>
> > and didn't
> > have a majority. Lacking a majority, pulling the center candidate would
> > tend to favor your opponent. This is because Conservatives will roughly
> > split between two Conservatives; Liberals will roughly split between two
> > Liberals; and so the three-way race necessarily has an imbalance whereby
> > there are more Conservatives or more Liberals, and so a split vote
> > resulting in no majority winner implies that there are more of the group
> > splitting the vote.
>
> i would put it more simply:  if there *is* a single-dimensional political
> spectrum from left to right, the center candidate is far more likely to be
> preferred as a 2nd-choice to the voters on the extremes than the candidate
> at the opposite extreme.  both GOP voters and Prog voters selected the Dem
> candidate as their second choice.
>
>
> > The plurality winner is smaller than this group, and
> > so is the Condorcet loser.
> >
> > That's a logical outcome, not a mathematical one. We can construct
> > situations where the liberals abandon the more-liberal candidate for the
> > conservative; that's not likely in practice.
> >
> > As per vote splitting, above, this three-way situation would tend to take
> > votes from the more-center candidate. The more-center candidate would be
> a
> > second choice for both ends, and so typically has overwhelming support
> > against each candidate WHEN THERE IS NO MAJORITY WINNER.
>
> well, we need to define exactly what we mean by a "majority winner".  if
> there are three or more candidates it's possible no single candidate gets
> majority support.  but between any pair of candidates, there is a majority
> winner unless they tie.
>
> that's all Condocet does.  just pairs the candidates with all possible
> pair combinations and says consistently, "If more voters mark their ballots
> preferring Candidate A to Candidate B than voters marking the contrary,
> then Candidate B is not elected."  that simple rule pretty much defines
> Condorcet.  and the problem in Burlington 2009 was 587 more voters marked
> their ballots that they preferred Andy over Bob than the number of voters
> preferring Bob over Andy, yet Bob was elected.
>
>
> > That suggests IRV provides no real advantage in the three-way race:
>
>
> IRV doesn't do too bad in a 3-way race if one of those three has far less
> support than the other two.  but in Burlington 2009, all three had
> approximately equal support.  the Prog was the IRV winner, the GOP was the
> FPTP winner, and the centrist Democrat was the Condorcet winner.  any of
> those three were a plausible winner.
>
> i don't know of another governmental race using Ranked-Choice Voting (what
> they now call it) where the method failed to elect the Condorcet winner.
> but i think it will happen again.
>
> When the spoiler has virtually no chance in winning (but a good chance of
> spoiling), IRV will choose the same as the Condorcet.  it's only when there
> are 3 or more plausible contenders that this hiccup happens.
>
>
>
> > - If Plurality isn't a Majority, IRV falls to vote-splitting and never
> > elects the Condorcet candidate, but avoids the Condorcet loser.
> >
> > - If Plurality IS a Majority, IRV elects THE SAME CANDIDATE as Plurality,
> > providing no advantage over Plurality in a three-way race. That candidate
> > is the Condorcet candidate.
> >
> > - When encountering a three-way Smith Set, Tideman's Alternative Smith is
> > identical to IRV after eliminating all non-Smith candidates.
> >
> > These are, again, practical outcomes, not mathematical ones.
> >
> > IRV's single advantage over Plurality appears to be that it
> mathematically
> > avoids the Condorcet loser, although when doing so it practically
> > eliminates the Condorcet winner.
>
> no it doesn't.  not most of the time.
>
> again, Burlington 2009 is the only governmental election using RCV i am
> aware of in which the single-transferable vote method failed to elect the
> Condorcet winner.  I think all of the other RCV elections have.
>
> but that's still not a good reason to use IRV over a Condorcet-compliant
> method.
>
> there are at least 4 solid reasons why IRV failed in Burlington in 2009.
> the most important is that it failed to resist a spoiler.  because of that,
> a promise to the voters that IRV made: that they would not need to vote
> tactically, they could serve their political interests well by voting for
> their favorite first, and marking their second favorite second.  but 1500
> Wright voters found out that, simply by marking their favorite as #1, they
> caused the election of Bob Kiss, where, if they had been tactical, they
> could have insincerely bumped Kurt down a notch and let Andy rise to the
> top and they would have prevented their lowest choice from winning.  it
> promised this spoiler thing would not happen.
>
>
> --
>
> r b-j                         rbj at audioimagination.com
>
> "Imagination is more important than knowledge."
>
>
>
>
>
>
>
>
> ----
> Election-Methods mailing list - see http://electorama.com/em for list info
>
>
> ----
> Election-Methods mailing list - see http://electorama.com/em for list info
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20181221/3e5e9fe8/attachment.html>