[EM] Dave: Condorcet

[Dave Ketchum]

On Mar 28, 2012, at 10:42 AM, MIKE OSSIPOFF wrote:
> Dave:
>
> You wrote:
>
> Which leaves me promoting Condorcet.  It allows ranking but, unlike
> ABucklin or IRV, all that a voter ranks gets counted.  Further, any
> voter able to match their desires to Plurality or Approval for a
> particular election, can vote by those rules and have them counted
> with the same power by Condorcet rules.
Why bother?  I see Fine and Soso as best so would consider approving  
both.  Trouble is that this would imply equal liking, perhaps getting  
Soso elected while I like Fine much better.

With Condorcet I can rank Fine above Soso, so that my ranking can  
improve the chance of Fine getting elected, while Soso has a chance if  
Fine fails.
> [endquote]
>
> Several problems with Condorcet:
>
> 1. Any method at all more elaborate than Approval, and especially a  
> method
> as different and elaborate as Condorcet is going to be much more  
> suspect
> than Approval is, to the public. Opponents will be able to say, "We  
> don't
> know enough about this voting system. It needs much study." Approval  
> is
> unique among voting system reform proposals, in that it's so simple,  
> and such
> a small modification of Plurality, that it's easy to show that it's  
> an improvement
> on Plurality, and only an improvement. That's much more difficult  
> for more
> complicated methods such as Condorcet.
Agreed that in many elections many voters would be satisfied with  
Approval ability.  Condorcet lets voters use that same ability - but  
use a bit more when they feel need for that much more.

Conceded that counting is more complex than for Approval, but the  
voting is only a bit more complex and the complexity only matters when  
the voters wish to use the capability.
>
> 2. For a handcount, Condorcet has much more count labor than does  
> Approval.
> If there are N candidates, then there are N(N-1)/2 pairs of  
> candidates. For each
> voter, a vote is counted among each pair of candidates. Now, in  
> Approval, let's
> estimate that, on the average, a voter approves half of the  
> candidates. N/2
> approvals to be counted per voter. That means that Condorcet has  
> (N-1) times more
> votes to be counted, as compared to Approval. Sure, the Condorcet  
> handcount needn't
> take longer, if you hire more counters. But, for one thing, the  
> count is equally
> more expensive, whether you hire more counters to work the same  
> hours, or the
> same number of counters to work longer. Besides, the more count-work  
> there is,
> the more opportunity and risk for count-fraud there is.
I question "handcount" when the labor gets heavy, such as for a  
senator or governor.  Anyway:

Approval estimates are a bit heavy - even with many candidates, voters  
cannot afford to vote for much more than 2 or 3 - voting for more  
risks electing those they like less.  Condorcet voting can not expect  
voters voting for many for the same reasons, though some more due to  
the ranking.

However the necessary complexity is overstated above.  Do need to  
count how many votes for each candidate as in Approval.  Then what of  
the pairs?  00 - nothing if neither voted for.  01 or 10 - just count  
the one voted for in the later add.  11, but equal ranked - counting  
them both leaves a tie.  11, but unequal rank - adjust the loser while  
counting the ballot so that when the count is adjusted for the 11 the  
result will be correct.  After counting all ballots, add each  
candidates vote count to each of that candidate's pairs.

Looking for winner.  Look at any pair - the loser cannot be the CW so,  
with luck, N-1 comparisons will delete all the non-CWs.  Cycles, etc.  
can complicate this a bit, but not enough to expect hundreds of  
compares for 20 candidates.
>
> 3. Condorcet fails FBC. Though it's unlikely, there are situations  
> where you
> can only prevent a worse result by favorite-burial. No matter how  
> rare those
> situations are, some, probably many, people will favorite-bury  
> therefore, because
> , for many people, helping their "lesser-evil" is everything. With  
> Approval, ABucklin,
> MCA, MTA, and their conditional versions, and also with ITC, it's  
> possible to
> _guarantee_ that there can never be any reason to vote anyone over  
> your favorite.
Assuming FBC is doable, what are the chances of getting the message  
out to enough doers without the word getting out to enough anti-doers  
and thus them seeing to flunking?
>
> Even if you only value reason #1, above, that alone is reason enough  
> to only propose
> what is proposable, enactable _now_.  Approval.
Agreed that Approval is cheap enough to do for its small value.  That  
does not stop something of more value being worth considering.
>
> One other thing: I used to claim that burial strategy is well- 
> deterred in Condorcet.
> But now I admit that I was only considering 3-candidate elections.  
> With more candidates
> burial isn't well-deterred. If I can find a candidate who will be  
> sufficiently well-beaten,
> then I and my faction can safely make hir soundly pair-beat the  
> sincere CW, without risk
> of electing that candidate whom we're ranking over the sincere CW.
Huh? If you do not risk your candidate getting elected you have no  
chance of more than annoying the CW.
>
> ICT has some good protection against burial, because burial can only  
> work for a
> candidate who is ranked #1 by more people than anyone else in the  
> cycle.
ICT or ITC?  Your zillion titles are beyond understanding.
>
>
> ICT would be a better proposal than Condorcet, since it also meets  
> FBC and CD (it's
> defection-resistant, unlike Condorcet). But ICT share's Condorcet's  
> problems #1 snd
> #2, above.
>
> Mike Ossipoff
Dave Ketchum
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20120401/93efc370/attachment-0002.htm>