[EM] IRV in Burlington VT

[Dave Ketchum]

NOTE:  While Condorcet and IRV both use rank ballots, they are  
DIFFERENT methods such as Condorcet, but not IRV, seeing ALL that the  
voter votes.

On Jan 9, 2010, at 9:41 PM, Warren Smith wrote:

>>> Given this experience, I would suggest that Burlington switch to a
>>> method different from plurality and different from IRV.   Some  
>>> obvious
>>> contenders are approval and range voting.
>>
>> Warren, why is Condorcet not included?
>
> --well, let's see.  First of all, there are problems with rank-order
> ballots, some of which Burlington indeed encountered.  Like
>
> * What if I rank A top and B bottom, but nobody in between?
> Burlington treated this as "A top, all rest coequal last."  Probably  
> wrongly.

Condorcet should see this as exactly two candidates being ranked, with  
one higher than the other.  Any unranked candidates should default to  
a rank below all ranked.
>
> * What if I rank A and B coequal?
> Most IRV elections treat this as "your vote is discarded"
> but probably better would be "half a vote for each."

This truly is a problem for IRV; Condorcet should see no trouble and  
should accept it.
>
> Those problems do not arise with range or approval.
> That's one reason life is simpler.

Life is different; simplicity is a topic to debate.
>
> San Francisco got a factor 7 increase in ballot spoilage when they  
> switched to
> IRV, see
>   http://www.rangevoting.org/SPRates.html

And Range would have no spoilage problem?  Need better educating?   
simpler rules?
>
> Range and approval also of course are simpler than Condorcet to  
> count, ignoring
> issues of ballot-casting-errors.

Approval has less capability, so its simplicity fits in.

Condorcet is complex to count.  However, voting seems more the current  
topic - how can Range claim to be simpler for this?
>
> There are other issues with Condorcet too, like there are many
> variants, but it'd be a lot to go into.   You can see some here
>   http://www.rangevoting.org/RangeYieldsCondSumm.html

Condorcet cycles have inspired MANY proposals - but hardly  
justification for giving up without more debate.
>
>
>> the problem with Approval voting is that it requires too little
>> information from the voters.
>> and the problem with Range voting is that it requires too much.
>
> --well, actually, not really.   First of all, the "more" info in range
> voting appears to
> be actually simpler for voters to provide, than the "less" info in a
> rank-order or approval vote. See
>  http://www.rangevoting.org/ManyChoices.html#hot
> so you might as well go for more, since range voting is simpler for
> both voters and for counters than Condorcet & IRV, and fewer voter
> errors.

The claim that you are almost responding to is that approval omits  
indications of relative liking among those approved, and that Range  
adds in stating magnitude of differences in liking.

Mentioning IRV here is a distraction.
>
> However, approval is simpler than range voting for counting and in
> terms of voter error-reduction.
>
> In terms of simplicity Approval>Range>Condorcet & IRV.

Debatable.
>
>
> In terms of winner-quality  Range>Condorcet&IRV&approval>plurality,
> in my opinion.

Debatable.
>
>
>> if you support a particular candidate (let's say the Dem) but you
>> approve of two (say both the Dem and the Prog), you have a problem
>> with voting strategy with Approval.  maybe, depending on how other
>> folks vote, you can help your favorite against the candidate you
>> tepidly support, so you punch "Approve" for your fav only.  or, maybe
>> your fav is actually not in really in the running and the election
>> becomes a contest between your fallback candidate and the guy you
>> think is a piece of crap.  then "what to do, what to do?? oh me oh
>> my!" (the savvy voter must thing strategically.)  most people will
>> just approve the single candidate they support and, if nearly all do
>> that, there is no difference between Approval and Plurality.
>
> --well, you are right this can be an issue with approval voting,
> although it it's more dubious you are right that most voters will just
> vote plurality-style.   I think approval
> voting will look like plurality voting in SOME elections, but in most
> it will not.  See e.g.
>    http://www.rangevoting.org/FrenchStudy.html
> where evidently plurality was not being simulated at all.   One of the
> virtues of
> range voting compared with approval is that it suffers less from that
> particular issue.
>
>> again "what to do, what to do?? oh me oh my!"
>
> --well, strategic voting is not easy in Condorcet or IRV either.
> In fact, it is probably harder.  But a lot of voters just use the  
> naive strategy
> of ranking A top & B bottom where A & B are the two "appear most  
> likely to win"
> candidates (e.g. Dem & Repub).   With that voting strategy,  
> Condorcet and IRV
> will always elect the same winner as plain plurality, and hence  
> should still
> lead to 2-party domination developing. This has historically been the
> case with IRV.
> For Condorcet it is more dubious.

With Condorcet, and with Range, you can show liking of more than one  
candidate, and liking one candidate better than another.  With either,  
voters need to be educated as to what they are dealing with.
>
> For the fact IRV rarely switches the winner away from the plain-plur
> winner, while the
> (sounds similar!) non-instant "top-2-runoff" system switches a lot
> more often, see
>    http://www.rangevoting.org/TTRvsIRVrevdata.html

Huh?  Plurality voters could not completely state their desires - and  
runoffs seem to have been added to improve results for such.  Assuming  
that Condorcet and Range claim better voting, they should expect less  
surprises from runoffs - and a related lesser desire to invest the  
time and labor of doing runoffs.
>
>> the Ranked-Order Ballot extracts the correct amount of information
>> from the voter.
>
> --wouldn't agree with that one.   It discards strength information.
> That matters a lot in Hitler vs Carter vs Ford.  Well, you maybe want
> to indicate you
> like Hitler a LOT less than C & F, but you cannot.  (Or Stalin vs
> Hitler vs Carter.)
> So then those who like Hitler a little more, win the election, and
> those who like him
> a lot less, i.e. Jews, lose the election.  And their lives.   Because
> strength info
> was ignored even though all voters were perfectly willing to honestly
> provide it if only they'd been asked.

Debatable.  Assuming Hitler being less desirable, this should show up  
in the vote counts, as well as in the strengths you emphasize.  While  
there can be extremes, translating a felt strength into a voted  
strength can be difficult.
>
>> even though there is the *possibility* of a cycle (and thus a
>> *possible* strategy of some to try to throw a Condorcet election into
>> a cycle) the likelyhood is soooo low, because of political alignment
>> along the major axis of the political spectrum.
>
> --I'm unsure how likely it is.  Everybody else is unsure too,  
> because enough
> data simply has never been collected to be sure.
> A cycle just happened in the Romania president election, see
>   http://www.rangevoting.org/Romania2009.html
>
> Maybe it happened in a lot of other elections too. Maybe not and  
> Romania
> was a fluke.  No way to know for sure.   My bias tends to be the same
> as yours, i.e.
> I'm guessing top-cycles are 1% or 2%, but it is only a guess.

Part of it is that the alignments cycles respond to should be rare,  
BUT when they happen proper response is important.
>
>> some of us are morons, that is true.
>
> --especially the ones who vote against you :)
>
>
> -- 
> Warren D. Smith
> http://RangeVoting.org  <-- add your endorsement (by clicking
> "endorse" as 1st step)
> and
> math.temple.edu/~wds/homepage/works.html

Dave Ketchum