[EM] Re: Issues, Condorcet, and IRV (was: IRV vs. plurality)

[Eric Gorr]

At 6:16 PM -0400 8/9/03, Dave Ketchum wrote:
>On Sat, 9 Aug 2003 08:14:46 -0400 Eric Gorr wrote:
>
>>At 5:26 PM -0400 8/8/03, Dave Ketchum wrote:
>>
>>>On Fri, 8 Aug 2003 11:08:30 -0400 Eric Gorr wrote:
>>>
>>>>At 3:03 AM -0400 8/8/03, Dave Ketchum wrote:
>>>>
>>>>>      Except for cases involving IRV spoilers, winners are 
>>>>>IDENTICAL - repeat IDENTICAL!!!
>>>>>  Condorcet backers emphasize the IRV spoilers because those are 
>>>>>the major reason for choosing Condorcet over IRV (Condorcet 
>>>>>claims simpler counting of votes as an additional advantage).
>>>>
>>>>
>>>>
>>>>What do you mean by IRV Spoilers?
>>>
>>>
>>>
>>>Looked to me as the same thing as happens under Plurality:
>>>      In 2000, Gore backers griped that Nader got in their way in Florida.
>>>      Here, C backers got in the way of B getting an easy win in IRV.
>>
>>
>>Just for reference, here is the example again:
>>
>>40:A>B>C      40:A
>>35:C>B>A  or  35:C>B
>>30:B>A>C      30:B
>
>
>First you say "here is the example" - and then you present TWO 
>UNEQUAL examples:
>       You give B a solid win.
>       Referenced example has B win via a cycle.

You were talking about IRV & Spoilers here and we were discussing 
these equivalent examples which, under IRV, would give the win to A 
even though B should win.

I gave the one on the left because you declared the one on the right 
to be unrealistic because it was unlikely the A & B backers would not 
have second and/or third place choices. I was pointing the same flaw 
under IRV can be shown even if the A & B backers provided second and 
third place choices.

>>You have presented one interpretation, but it is not the only one, 
>>which considering it in a more generic form.
>>
>>C could simply be a third, unique option, which was least liked by 
>>the broadest number of people.
>
>
>The referenced example is consistent with C being near to B, rather 
>than being unique (if unique, likely some As would take an interest 
>in C).

In the example above?

C is not near to B. If this were true, the B backers would have voted 
C as their second place choice...they didn't. So, something else must 
be going on. But, the various spins one can put on this are not 
things I find particularity relevant.

It was only the C backers which seemed to like B better then A.

>>>>I think it has already been shown that the winners between IRV 
>>>>and Condorcet are not identical in more then just these cases.
>>>
>>>
>>>
>>>I concede below that it can be other than odd cases.  As to odd cases:
>>>      Markus recently hit us with a demonstration that Condorcet 
>>>can be tortured into odd results with paper voters - and doing 
>>>that particular torture with real voters seems unbelievable:
>>>      Prospect was that A was winning, three more ballots turned up 
>>>with A in first rank, and these ballots caused D to win.
>>>      Method was to start with a particular method of resolving 
>>>cycles (nothing wrong with the method - just provides knowledge 
>>>for constructing the torture), provide a 6-member major cycle 
>>>containing two 3-member minor cycles, and define vote counts for 
>>>two pairs connecting the minor cycles such that lower ranks in the 
>>>three odd ballots will change which of these pairs is discarded 
>>>first, and thus which minor cycle contains the winner.
>>>
>>>     Turns out the three voters could have succeeded via bullet voting,
>>>BUT, for all they knew without the vote counts in front of them, could
>>>have happened that their lower rank votes were needed to let the A cycle
>>>contain the winner.
>>
>>
>>Here is that example again (in a form so it can be easily fed into my site):
>>
>>4:A>B>C>D>E>F
>>2:A>B>F>D>E>C
>>4:A>E>B>F>C>D
>>2:A>E>F>B>C>D
>>2:B>F>A>C>D>E
>>2:C>D>B>E>F>A
>>4:C>D>B>F>E>A
>>12:D>E>C>A>B>F
>>8:E>C>D>B>F>A
>>10:F>A>B>C>D>E
>>6:F>A>B>D>E>C
>>4:F>E>D>B>C>A
>>3:A>E>F>C>B>D     - the extra three votes
>>
>>According to my site, there is a tie between A & D with or without 
>>those extra three votes added. The reason is that when I consider 
>>the victories that form the minor cycles, those victories get 
>>rejected because they form cycles.
>
>
>The example was presented by Markus Schulze, designed to give the 
>results I described, when counting via BEAT PATH method.  Clearly 
>you must have used a different method to get your different results.

Yes, as I clearly stated, I was using my site and obviously not using 
either Beatpath or Basic Condorcet - that left only one option.

>I do not like declaring a tie, for we prefer declaring a winner 
>without tossing a coin when at all possible.

Of course, but this example clearly shows the flaw in the Beatpath 
Winner method.
However, in a better RP method, the flaw disappears and a tie is declared.

However, ties can and do happen. How they are resolved is something I 
would like to keep out of the scope of the actual vote count method.

>>This clearly seems to be the right thing to do in this case as the 
>>voters have provided no clear information on how they feel between 
>>A & D.
>>
>>This, unfortunately, does not appear to be an odd case.
>>
>
>Are you saying that six candidates tangled in cycles within a cycle 
>is not odd - could be common in the books that discuss 
>possibilities, but this many candidates contending this evenly in 
>real voting looks odd to me.

No, I am saying that it is not odd because there is an unambiguous 
result using Ranked Pairs from my site...a tie.

It is an odd case under Beatpath because clearly finding those three 
extra votes obviously should not cause D to win, which they do.


-- 
== Eric Gorr ========= http://www.ericgorr.net ========= ICQ:9293199 ===
"Therefore the considerations of the intelligent always include both
benefit and harm." - Sun Tzu
== Insults, like violence, are the last refuge of the incompetent... ===