Re: [EM] [CES #3845] condorcet & range voting -- which one yields more condorcet winners?

2011-10-13 Thread Kevin Venzke 
Hi Jameson,

--- En date de : Mer 12.10.11, Jameson Quinn  a écrit :




 

Maybe you should do sims first, emit flames second.


That's a fair criticism, and one I continue to violate in this message. I 
wonder if Kevin Venzke has any sims which speak to this question.
 
 
Thanks for remembering me. The question is sincere Condorcet efficiency between 
Range and something like MCA? I have three scenarios on-hand (two 1D, one
spectrumless, all three-candidate) and MCA is a bit better than (four-slot) 
Range in
all three. But it is rare that either method has the efficiency of a Condorcet 
method.
 
My sims use 100% strategic voters and polling by the way.
 
A few comparisons.
Spectrumless (blocs have random preferences):
IRV 92.2% (of trials with a CW)
WV 91.7%
MCA 90.6%
Range 90.3%
Approval 88.4% (note that Range doesn't quite become Approval due to the voters
being divided into a fairly small number of strategizing blocs)
FPP 84.8%
 
1D with random candidate positions:
WV 99.1%
IRV 98.4%
MCA 97.9%
Range 97.6%
Approval 96.5%
FPP 84.0%
 
1D with random candidate positions but distance from center halved 
(center-heavy):

WV 97.8%
IRV 97.3%
MCA 96.9%
Range 95.7%
Approval 93.2%
FPP 76.6%
 
Note that the voters have the ability to get what they want a high percentage 
of the
time no matter the method. But they may have to compromise or use other 
strategies
in order to do it. For instance...
 
Here are the percentage of elections in which at least a quarter of the voters 
ended up
compromising (favorite betrayal). The three figures follow the order of the 
scenarios 
above.
 
WV 0.4%, .02%, 0% of elections
IRV 9.3%, 4.6%, 5.7%
MCA 0%, 0%, 0%
Range 0%, 0%, 0%
Approval 0%, 0%, 0%
FPP 17.4%, 17.3%, 18.7%
 
I could have produced other figures as well, such as the rather alarming burial 
rate
under WV. But the point is just that the Condorcet efficiency with strategic 
voters,
this single figure, doesn't tell a complete story.
 
Another point is that I'm not using one cookie-cutter strategy for all methods 
here.
The voters' strategy is deduced by AI, not by me.
 
I do realize I need to get around to making more of my work available.
 
Kevin
Election-Methods mailing list - see http://electorama.com/em for list info

Re: [EM] [CES #3845] condorcet & range voting -- which one yields more condorcet winners?

2011-10-12 Thread Jameson Quinn 
2011/10/11 Warren Smith 

> On Tue, Oct 11, 2011 at 8:19 AM, Jameson Quinn 
> wrote:
> > Note that the "more Condorcet-like than Condorcet" is only true for Range
> if
> > voters are strategic and somewhat knowledgeable about the polls. It is
> true
> > for Majority Judgment under the same conditions, but also when any
> fraction
> > of voters are honest and ideology is one dimensional; I believe that it
> > holds for N dimensions but I have not proven it. It is true for SODA if
> most
> > voters agree with their candidate's rankings. I believe that these
> > conditions for MJ and SODA are broader than the conditions for Range.
> >
> > Jameson
>
> --But wait -- the simulations in
>
> http://rangevoting.org/StratHonMix.html
>
> found that TopMedianRating returned fewer Condorcet winners than
> average-based range voting.
>

In that sim, Range elected 13279/2 CWs, and Median elected 12472/2.
This is a significant difference, but not a huge one. On the other hand are
two effects:
1. Range's greater strategy incentive
2. The tendency for voters to polarize, giving exactly one of the two
frontrunners an *honest* rating near zero. This is a *separate* effect from
strategic exaggeration. If true, this tendency increases the probability
that an honest median vote is strategically strongest, but does not do as
much for Range.


> I believe these sims were conducted with random tie-breaking though
> (not Balinski-Laraki
> nonrandom tiebreak method).
>

I suspect you used enough rating categories (100?) that the difference is
immaterial there.


>
> Maybe you should do sims first, emit flames second.
>

That's a fair criticism, and one I continue to violate in this message. I
wonder if Kevin Venzke has any sims which speak to this question.

Jameson

Election-Methods mailing list - see http://electorama.com/em for list info