[EM] To Condorcetists:

[Juho Laatu]

On 14.5.2012, at 22.03, Michael Ossipoff wrote:

> You said:
> 
> I note that that would lead to an interesting political system that has
> probably not been tested anywhere in the world yet.
> 
> [endquote]
> 
> Single winner elections have actually been tested!  And widely used, Juho! I
> kid you not!

Yes, I know. I was thinking that good single-winner methods have been designed to elect single winners. They are not designed to elect representative bodies from single-winner districts. In theory use of a Condorcet method in the Senate and House of Representatives elections could lead to electing all representatives from a small centrist party (zero from Democrats, zero from Republicans). Probably that is not the intent.

I see multi-winner methods as a separate set of methods where the requitements are quite different from the single-winner method requirements. Representative bodies have multiple members, so by default they should use multi-winner methods. Pluraity based two-party systems are a special case that uses Plurality (a s-w method) to achieve the two-party effect in a m-w election. But in general, a good s-w method is not necessarily a good m-w method, and a good m-w method is not necessarily good s-w method.

I do understand that jumping e.g. to PR and multi-winner districts in the U.S. may not be possible in the short run. But I have not heard anyone naming use of single-winner methods in single-winner districts to elect multi-member representative bodies as their ideal target. Maybe that could be one intermediate step (first easy step) on a path towards something else.

When I say that Condorcet methods are good methods I mean that they are good methods for typical single-winner elections (or to be more exact, for _compromise_seeking_ single-winner elections) (and I don't mean single-winner districts in multi-winner elections).

> What would be somewhat new would be single-winner elections without
> Plurality's forced falsification requirement. And yes, the lack of testing
> and prior experience would be a problem for most reform proposals. But not
> for Approval, because, as I said, that minimal change from Plurality is so
> simple that it would be obvious that it would be an improvement, and nothing
> other than an improvement.

Both Approval and Condorcet methods are compromise seeking single-winner methods in the sense that they tend to elect centrist compromise candiates with no requirement of proportionality. In that sense changing the method from Plurality to Approval may lead to major changes in the distribution of the seats in the long run (although the technical change is small). If one starts from a two-party set-up, in the first elections Approval may just allow some approvals to be given also to third parties, but it may still elect practically all representatives from the two old parties. Old party supporters might generally bullet vote. But in the long run things may change. Approval has the tendency to elect centrists, not from the two major parties of the two wings. One may consider also that property to be an improvement. But from another viewpoint, maybe people don't want the system to change in that way. If one wants to allow also third party candidates to win, maybe the long term target could be proportional representation of all the parties. The problems of Condorcet (when used as part of a a multi-winner method in single-winner districts) are quite similar.

It is possible that use of Approval would not lead to as clear tendency of electing centrist candidates as described above. But that does not mean that the method would behave in some other sensible way that would be easy to predict and easy to justify. Large parties might continue to bullet vote, and small parties might not grow strong enough to widely challenge the old strong ones. Voters might stick to the stong ones since they are considered to be the strongest players in Washington anyway.

> If you're merely saying that you make no recommendation regarding voting
> systems here, then that's fine. If you're saying that Approval is less
> qualified for
> recommendation, than that claim would call for justification.

I think the U.S. citizens should decide, and I try to avoid taking position on what they should do. However, the reason why I see potential problems in the use of Approval or Condorcet in electing representative bodies is that they are not planned to be used that way. There is no nice theory behind that would support that kind of a political system. But maybe also such untested political system could be one useful (intermediate?) step in the reform process. U.S. citizens to decide.

For presidential elections Approval and Condorcet would be fine (or as good as they are as single-winner methods) except that the presidential elections of the U.S.A. are not pure single-winner elections in the sense that the whole presidential system is based on the assumption of having two major parties alternating in that posiion. I don't know what would happen if some compromise candidate from a minor centrist party would be elected as the next president, and he would have to build the presidential machinery (government etc.) in Washington from scratch. I also wonder how the minor party president and his single-party government would cooperate with the Senate and House of Representatives. And would people be happy to give all the power of the president (including military) to a representative of a small minority for four years? Maybe all that would work fine, or maybe not well enough. U.S. citizens to decide.


> I don't see any denial of Gibbard-Satterthwaite or other problems. My
> understanding is that many people like Condorcet methods because they think
> that their co-operation/defection problems are relatively small (although
> they exist at least in theory).
> 
> [endquote]
> 
> Nonsense. Can you justify that claim? I've showed a whole range of numerical
> examples, from the 27,24,49 example to the 33,32,34 example. I've told how
> the problem would come about, in Condorcet, just as well as in Approval.
> Condorcet is not strategy-free, or anything close to it.

Yes. I don't see anyone denying the mathematical properties of the methods. But towards the end of my mail I said that I want to see if the methods are vulnerable also in practice. It is not sufficient to demonstrate that one person with full knowlege of the content of the actual votes and ability to change any of them could change the result. One good approach to demonstrate the practical vulnerability of a method would be to write some generic rules (or rules for some specific common situation) that voters or parties could use in real life elections to change the result. I have not seen any such good guidance that could be used to cheat Condorcet methods.


> I believe that ICT
> might actually come close to delivering on that promise. Condorcet does not.
> I've abundantly told why. If you specifically disagree with something I've
> said about that, then tell which statement you disagree with, and why, and
> be specific.

I hope I already answered that. Convincing practical guidance on how voters should vote in real life elections could do the job. I have seen some recommendations like "bury as much as you can", but I don't consider that very credible since often it doesn't make sense to bury, and this strategy may also make the results worse. A good startegy would be one that is likely to improve the results and can be implemented in real elections. I also assume that voters slightly prefer parties that want to play fair game.


> I can see a potential problem of numerous voters voting D > all_others > R
> or R > all_others > D in the first Condorcet elections just to make sure
> that their worst competitors will not win. My hope is that they would soon
> learn that there is no need and no sense to do so.
> 
> [endquote]
> 
> You see, that hope isn't good enough, unless it has some evidentiary
> supporet.

I'd take the risk. Anyway, I expect that Democrats and Republicans would not adopt Condorcet if their plan is to recommend all their supporters to bury the candidate(s) of the other major party. That would be too foolish. (They might however be happy with having some insignificant number of irrational burial srategists since those people in some sense help them making their worst competitors look bad.) So, the major parties would probably not recommend rampant burial. There's good hope that most voters would give up the idea of automatically burying the strongest competitors if the main parties agree that sincere voting gives better results (although they'll of course continue demonizing their strongest competitors verbally). I hope that there would be many experts and media telling people that Condorcet works just fine with sincere ranking.

I have not heard of any strategy problems in the Condorcet elecions that have been held. Unfortunately we don't have yet any good examples from strongly competitive political Condorcet elections. I have also not heard of any examples of significant or systematic strategic voting in real life IRV elections (excluding Australia again).


> Yes, Condorcet's strategy is far too
> complicated to be known. But people will know that the Democrat can only be
> maximally helped to beat the Republican by favorite-burial.

That strategy ignors the fact that it may do more harm than good. Did you assume rational or irrational voters here? It would not be a rational strategy to recommend all Democrats to bury all Republicans and vice versa.


My summary is that in order to prove that Condorcet methods will not work well enough in real life elections I'd like to see also demonstrations on how Condorcet methods can be cheated in practice. That could be done by giving rules that can be followed. For example: "If voter sees situation x, then he should modify his vote so that ..." or "If party sees situation y, then it should publicly recommend its supporters to modify their vote so that ...". When planning the strategies one must take into account that the polls are not exact and the opinions may change before the election day. Also the srategic moves migh change the opinions of the voters (e.g. the target party supporters may rank the strategists sincerely lower). It is also not possible to control the voting behaviour of all the supporters. An alternative approach would be to claim that many enough voters will use irrational strategies, and therefore Condorcet (and maybe also other) methods are likely to lead to unsatisfactory results.

Juho