[EM] Re: minmax is not a good public election method

[Russ Paielli]

Hi Kevin,

Kevin Venzke stepjak-at-yahoo.fr |EMlist| wrote:
> Russ,
> 
> --- Russ Paielli <6049awj02 at sneakemail.com> a écrit :
> 
>>Imagine the following scenario. MMPO has been adopted for a major 
>>election. The results come in, and a Condorcet winner exists but does 
>>not win. With MMPO, that would not be unlikely at all. Maybe the 
>>Condorcet Loser even wins. Now imagine what the supporters of the CW are 
>>going to say:
> 
> 
> Hmm, I'm not sure if the CL can win when there is a CW. I should think about
> that.

That would be the ultimate embarrassment, eh?

>>"Hey, wait a minute. Are you telling me that my guy beat every other 
>>candidate but didn't win the election? What kind of brain-dead system is 
>>this? Who were the idiots who designed this scheme? Why do we bother 
>>with pairwise tallies if we are just going to ignore them?"

I hope you understand that I was not trying to label you or anyone else 
an "idiot" for advocating MMPO. I was just highlighting the fact that 
people can get very heated when they feel they've been "cheated" in some 
way.

> 
> Well, it seems to me that when the CW loses in MMPO, it is never very ugly.
> The MMPO winner would not have any majority-strength loss, for instance.

That depends on how you define "majority." If you define it in terms of 
the number of voters who voted a preference in that particular pairwise 
race, then *every* pairwise loss is a "majority-strength" loss. We've 
been over this before, of course. You could also define a "majority" in 
terms of all registered or all eligible voters, but why? What is the 
fundamental difference between someone abstaining from the entire 
election or from one particilar pairwise race? And why should one be 
counted and not the other in your definition of "majority"? (That's 
intended as a rhetorical question, by the way.)

> Electing the CL is more alarming, but the pairwise tallies haven't been
> "ignored" exactly.

The result of who actually wins or loses any particular pairwise tally 
is completely ignored by MMPO.

Not to go to far off on a tangent, but this reminds me of one of my 
little pet peeves with regard to betting on (American) football. I don't 
bet on football (or anything else), mind you, but the way it is done 
bothers me.

For each game, a "point spread" is established by some central gambling 
authority. Say the Eagles are favored by 10 points over the Lions. That 
means that if you bet on the Lions, you get 10 additional points. That 
implies that you are no longer betting on who actually wins or loses the 
frickin' game! You are betting on who wins or loses some contrived 
"game" that the teams themselves hardly care about.

As the end of the game nears, do you think a team will take risks to win 
the "contrived" game that they would take, if necessary, to win the real 
game? Of course not. And suppose the game is a cliff-hanger right down 
to the last play. The betters couldn't care less who actually wins, at 
least not as far as their bet is concerned. Something is fundamentally 
wrong there. A much better system would place odds on who wins the real 
game, similar to what is done in horse racing.

>>Imagine the field day Leno and Letterman would have. Then someone would 
>>come along and try to explain that MMPO satisfies FBC and LNH. Yeah, 
>>right. Imagine how well that will go over with the general public!
> 
> 
> I'm not convinced. The public might not understand FBC but I think they
> would understand LNHarm enough to be unsure if they want to shed it in
> favor of obsessively finding the pairwise winners.
> 
> I think the public would inevitably complain about MMPO's failing of
> Plurality. Personally, I'm more distressed about failing Minimal Defense.
> 
> 
>>As most of you realize, we have a dilemma here. You can design an 
>>election method that counts sincere votes in a reasonable way, or you 
>>can design one that provides little or no incentive to vote insincerely, 
>>but you can't do both at once. You want FBC and/or LNH? Then you can't 
>>satisfy Condorcet.
> 
> 
> Well, I've suggested a method or two now which satisfy FBC and come so close
> to satisfying Condorcet as to make no difference.

You've proposed some very innovative ideas, and you are obviously a very 
skilled analyst. However, I don't agree that you can "come so close to 
satisfying Condorcet as to make no difference."

>>So which concern should prevail? Should sincere votes be counted the 
>>best way we know, or should guarantees be given that sincere voting 
>>won't backfire? It seems obvious to me that any election method must 
>>first be able to deal well with sincere votes before any other concern 
>>is addressed. A good tally method for sincere votes is a *prerequisite*. 
>>*Encouraging* sincere voting is also important but is ultimately 
>>secondary to actually counting sincere votes properly.
> 
> 
> This isn't obvious to me. What if, hypothetically, satisfying Condorcet meant
> that no more than two candidates felt they could enter the race? In that case,
> I think we would be paying too much for Condorcet. It would be better to
> improve the incentives, and perhaps not satisfy Condorcet all the time.

In that case, I'd personally just give up on trying to find a good 
ranked-ballot method.

By the way, try googling Kenneth Arrow sometime. He's still alive, and 
he has had a highly successful career. If I recall correctly, he's been 
the president of a couple of professional societies, and he's recieved 
several major awards in addition to the Nobel Prize he won way back, but 
I don't think any of them have much to do with voting systems. So why 
did Arrow "drop out" of voting theory so long ago after winning a Nobel 
Prize? Maybe he understands something we don't. It's just a thought.

>>By the way, I see that Kevin just added an Approval cutoff to MMPO. I 
>>consider that an improvement but it still falls short.
> 
> 
> Eh? I only added it because Mike's version looks problematic. I don't advocate
> it, since it breaks LNHarm.

Sorry for the misunderstanding

>>I still say that Smith/Approval or DMC/RAV are probably the best we can 
>>do. The rules are reasonably simple, and the Approval cutoff gives the 
>>voter a critical additional mode of expression without violating CC. And 
>>allowing equal ranking makes it at least as good as Approval for those 
>>who choose to use it that way.
> 
> 
> I prefer Condorcet//Approval with the special tie rule. Maybe I should give
> it an actual name: "Improved Condorcet Approval." It satisfies FBC, and only
> fails Condorcet by letting people vote to create pairwise ties. When the CW
> loses an "ICA" election, he doesn't have a good claim over the ICA winner.

I don't know how ICA works, but it sounds interesting. Do you mind 
explaining it or pointing me to its definition? Thanks.

--Russ