[Election-Methods] RE : Re: RE : STV in the context of modeling voters

[Jonathan Lundell]

On Dec 30, 2007, at 1:28 PM, Kevin Venzke wrote:

> Jonathan,
>
> --- Jonathan Lundell <jlundell at pobox.com> a écrit :
>>> I didn't quite understand your post... In the first part of your
>>> post you
>>> seemed to be discussing the strategy of selecting the candidate you
>>> want to
>>> win. It didn't seem you were discussing the voting strategy once you
>>> know
>>> what you want. So how could this become easier when you're allowed
>>> to rank
>>> the candidates (whether with a LNHarm guarantee or without)? Surely
>>> it's
>>> harder to form a ranking than to pick a single favorite?
>>
>> I'm arguing otherwise, or at least that ranking N candidates on a
>> contingency basis is of difficulty O(N), while trying to rank on the
>> basis of quantifying and consolidating multidimensional preferences  
>> is
>> O(unquantifiably harder).
>
> I must not be understanding you. I thought surely you would face the  
> latter
> problem no matter what election method you were using. Are you saying
> plurality voting only requires the second, and IRV only requires the  
> first?

No. I'm saying that quantified multidimensional models of voters  
probably don't correspond to the way most voters arrive at their  
choices.

>
>
>> I'm also, at least implicitly, suggesting that the use of voter  
>> models
>> to evaluate election methods is interesting, but that the difficulty,
>> to the point of impossibility, of building and verifying such models
>> that reflect real-life voter behavior makes it problematical to draw
>> much in the way of conclusions.
>>
>> For example, someone recently pointed out that IRV elects the middle
>> of three candidates only 1/3 of the time, with a voter model based on
>> random distribution along a single dimension. I conclude that if I
>> ever have occasion to hold an election in such an environment, I  
>> might
>> or might not use IRV, depending on whether I had decided a priori  
>> that
>> the middle candidate was preferable (in which case I could suggest a
>> trivially simple algorithm for picking the winner).
>
> If you have a trivially simple algorithm for electing the middle of  
> three
> candidates, and which doesn't come with strategic incentives that  
> stop it
> from working properly, I would love to hear what it is.

I'm suggesting that if a) what we want is the middle candidate, and b)  
some disinterested observer is able to reliably discern who that is,  
then we should simply make that observer a dictator. In this context:

> For me, even a totally strategy-free method would not be acceptable
> if I could not stand to the side as an observer and agree that the  
> method
> gave proper results.


I don't buy the premises, of course.

>
>
> The reason why, in a one-dimensional setting, we might want to elect  
> the
> centermost candidate, is because this candidate should defeat each  
> of the
> other candidates head-to-head.

Only if you're prepared to make a priori assumptions about voter  
distribution.

Consider abortion as an example of an issue that has influenced voter  
decisions in the past and no doubt will in the future. My impression  
is that, among voters for whom abortion is an important influence on  
their voting decisions, distribution is highly bimodal.

I don't care to spend much energy on the question though; I'm arguing  
that the behavior of a given voting system in such a setting is not of  
much practical interest, because the setting is so far removed for  
typical voter behavior.

>
>
> Alternatively, if utility of a candidate to a voter is posited to be
> proportional to the proximity of the candidate to the voter, then the
> centermost candidate should provide the greatest utility.
>
>> And I'm suggesting that we know from experience that voters can and  
>> do
>> select a single winner, and that IRV builds on that by asking the
>> voter merely to do it iteratively.
>>
>>> --- Jonathan Lundell <jlundell at pobox.com> a écrit :
>>>> And its implicit
>>>> decision model conforms to something that I, as a voter, can make
>>>> sense of. And in making my contingent choices, STV's later-no-harm
>>>> property becomes important to me. Of course I'd like it to be
>>>> monotonic, etc, but on *my* value scale, its advantages, in  
>>>> practice,
>>>> outweigh its shortcomings--in practice.
>>>
>>> For me, even a totally strategy-free method would not be acceptable
>>> if I
>>> could not stand to the side as an observer and agree that the method
>>> gave
>>> proper results.
>>
>> I question the ability of any observer, in general and in principle,
>> to do any such thing.
>
> That's very interesting. The MMPO method satisfies LNHarm like IRV,  
> and
> additionally has no favorite betrayal incentive. The only problem is  
> that
> it has scenarios like this:
>
> 1,000,000: A
> 2: A=C (ranked equal)
> 2: B=C
> 1,000,000: B
>
> MMPO elects C. Can you find any basis on which to criticize this  
> result?

Sure, if for example we have 2,000,000 voters who find C unacceptable.  
A is pro-life; B is pro-choice; C wants universal forced sterilization  
and Nobel Peace Prizes for both Bill Clinton and George W Bush.

>
>
> It's hard for me to imagine that an analyst might *only* value the
> strategic incentives (or lack of) of a method, and not be concerned  
> with
> any qualities of the outcome. Personally I want to reduce strategic
> incentives *in order to* improve the outcome.

Don't we all? But I don't want to presuppose that we can in principle  
determine the "quality of the outcome".