[EM] possible improved IRV method

[Dave Ketchum]

On Wed, 28 Jun 2006 22:00:15 -0400 Allen Pulsifer wrote:

> Hello Dave,
> 
> There are a few things I like about IRV as opposed to Condorcet.
> 
> First, I think IRV is a reasonably straight-forward extension of current
> runoff methods.  I think it will be more readily understood and accepted.
> In contrast, Condorcet could be described as "wonkish" 

Agreed runoffs help recover from Plurality's problems, as best that may be 
done, but that does not make them something to brag about when looking for 
something good.

Hearing of them reminds me of recent French presidential elections.  A clear 
majority would have accepted whichever of the majority-thinking candidates 
won - but MANY such candidates split up the majority vote.  Leaving a couple 
minority candidates to get to the runoffs - and voters to threaten riots.

Better to talk of tournaments, and deciding between A and D based on their 
comparative backing, without getting distracted by candidates with only 
minority backing, such as B and C below.  Perhaps time for an example:

  31 A
  26 B>D
  23 C>D
  20 D

Here IRV will discard D, C, D, and then declare that A wins over B.
Condorcet will discard B and C as minorities, and recognize that 69 D 
properly beats 31 A.

Note that 21 C>D plus 22 D would have caused IRV to award D.
> 
> Second, IRV satisfies the "Majority Rules" criteria.  In the Majority Rules
> criteria, at the end of the election, you can point to the group of voters
> who form a majority and were responsible for the election of the winner.
> This serves three important purposes.  First, it lends legitimacy to the
> outcome.  Second, it tells everyone who the governing coalition is.  Third,
> it provides feedback to the candidates that is critical for making them
> responsive to the voters.

"lends legitimacy" does not excite me, though "owns legitimacy" might.

Neither IRV nor Condorcet require an actual majority - each does the best it 
can, within its rules, with whatever the voters offer.

As to telling everyone, I claim that the Condorcet array gives a better 
picture than IRV offers - especially to candidates.
> 
> The final effect is not to be underestimated.  The candidates are not fixed
> constellations in the sky that we choose between.  Instead, they constantly
> fine-tune their positions, attempting to maintain the backing of a majority.
> This fine-tuning is a good thing -- it is what ensures our elected officials
> represent us.
> 
> Our two party system however does have issues.  It has evolved into a
> polarized equilibrium, where the Republicans and Democrats have drawn a line
> that divides the electorate down the middle.  Worse, with the primary
> system, the nominees themselves are at the middle of their respective
> parties, which is about the 25% percentile point relative to the electorate
> as a whole.  The only centrists tend to be "Blue" candidates from "Red"
> states, and "Red" candidates from "Blue" states.  One of the goals of
> alternative voting system is to provide a viable way for voters to pick from
> the middle rather than the extremes.

All of which argues for moving to ranked choice, where IRV is an improvement 
and usually agrees with Condorcet, though subject to the problem I illustrate above.
> 
> Getting back to Condorcet, there is a majority in each pair-wise comparison,
> but for each pair, it is a different set of voters.  There is no way, at the
> end of the election, to go back and say "This is the majority that elected
> the winner".

As you said above about candidates learning, so can parties - and all can learn 
from Condorcet arrays even if the learning causes all to do better at 
responding to voter desires - even if they improve enough that third parties 
almost never win elections.
> 
> In certain cases, Condorcet can also result in a strange outcomes.  Take for
> example, the following (admittedly contrived) situation:
> 
> 166:A>B>D>C
> 166:A>C>D>B
>  83:A>D>B>C
>  83:A>D>C>B
>  83:B>C>D>A
>  83:B>D>A>C
>  83:B>D>C>A
>  83:C>B>D>A
>  83:C>D>A>B
>  83:C>D>B>A
>   2:D>A>C>B
>   1:A>B>C>D
> 
> The total votes is 999.  Candidate A, with 499 top rankings, is only 1 vote
> shy of a majority.  His two second rankings would bring him across the
> threshold.  Nonetheless, Candidate D, with 2 first preferences and 498
> second preferences is the Condorcet winner.  Doesn't that seem backwards?

Each of us sees strangeness subjectively.  When I look:
     B/C/D backers agree on voting against A.  B and C disagree on some 
issue internal to that group, neither being strong enough to win an election.
     So A vs D is the contest of interest, is a near tie, and even accepting 
a marginal win beats doing a runoff, since all the voters have had an 
opportunity to indicate preference - as have the candidates, who should have 
based their campaigning on being near a tie.
> 
> Note I believe I may have been incorrect about my previous proposal meeting
> the Condorcet criteria.  While under that proposal a candidate can only be
> eliminated if he loses a pair-wise comparison, the comparisons happen after
> candidates have been eliminated and votes redistributed.  This would seem to
> have some sort of effect, but I'm still hashing it over.
> 
> In addition, the ideal voting system to be monotonic.  Classic IRV is
> recognized as not monotonic.  One of the goals of my proposal was to at
> least find a monotonic method.  I don't know if it accomplishes this of not.
> This may be a problem inherent in IRV: whenever you eliminate a candidate
> and then retabulate votes, it introduces the potential for non-linearities.
> 
> With respect to "strategic voting", I believe the ideal voting system would
> be essentially immune from strategic voting.  Each voter would state their
> sincere preferences, and the tabulation would be responsible for determining
> the winner as if each voter had executed their optimal strategy.

Agreed that strategic voting is a problem.
     Seems like some acceptable strategies get falsely accused.
     Seems like more do not have the ability.  If clan leader A needs to know 
all that will happen without his strategy, he is in trouble UNLESS he can be 
sure that there is no possibility of clan leader B existing and changing that 
environment.
     Further, clan A voters need to know what their strategy is to vote it, 
without clan B finding out and responding.
> 
> Finally, your comment regarding tabulation and precinct voting is a point
> well-taken.  I have another idea for a possibly improved IRV that I will
> post separately.
> 
> Best Regards,
> 
> Allen Pulsifer

-- 
  davek at clarityconnect.com    people.clarityconnect.com/webpages3/davek
  Dave Ketchum   108 Halstead Ave, Owego, NY  13827-1708   607-687-5026
            Do to no one what you would not want done to you.
                  If you want peace, work for justice.