[EM] An hypothetical voting system based on Score-Voting and Majority-Judgement which I do not advocate.

[Kristofer Munsterhjelm]

On 12/14/2012 08:26 AM, ⸘Ŭalabio‽ wrote:
> 	2012-12-13T06:53:10, Kristofer Munsterhjelm:

>> - If the voters know that +99 and -99 will be discarded, that
>> effectively turns +99 and -99 into 0. Thus they'd not use those values,
>> instead knowing their "real maxima" to be +98 and -98.
>
> Yes, but they have to give at least 1 candidate has to get a
> negative
> -99 and another has to get a positive +99. The could hive those score to
> other candidates and give negative -98 to their most hated candidate and
> positive +98 to their favorite, but when removing the top and bottom 3rd
> in the next step will likely remove both the negative -98s and the
> positive +98s too. It is hard to find a way to strategically vote in
> this system.

In one sense, it's true, but in another, the problem remains. To show 
the problem, let's call +99 "Q" and -99 "R" just to show that the 
argument doesn't depend on the values of the ratings you'll be throwing 
away.

Then, to a game-theoretically rational voter, Q and R are effectively 
"no opinions", because they're not going to be included in the count. So 
his strategy should be to give Q and R to parties or candidates of which 
he has no opinion (and thus don't want to say anything about). Whether Q 
is called "+99" or "+50" doesn't matter.

The requirement that the voter spend at least one Q and R makes strategy 
a little harder. However, it's not all that much harder. If the voter 
knows there are two no-hopes that have no chance of winning, he could 
spend his Q and R on them. More precisely, if there are candidates that 
are either sure to win or lose, the voter can allocate Q and R to them. 
If enough voters feel this way, some of them may nominate a "throwaway 
candidate" for that purpose.

If I were to make a method with the logic of throwing out extreme 
ratings, I would use a truncated mean or something similar (e.g. 
Winsorising) that is general instead of specific. That is, if you have a 
10% truncated mean, it's going to throw away the upper and lower 5% no 
matter whether those ratings are +99 and -99 or +98 and -98. In 
contrast, your method may give a different result if every voter halved 
the range he was going to vote over (if he could), which seems pretty 
strange.

>> - For any given society, there's probably some optimum such level
>> if  you want to use utilitarian voting.
>
> 	Perhaps.  That is a great open question.

If utilitarian voting is Range-like, then you effectively have a 
trade-off between resistance to strategy (higher breakdown point is 
better) and responsiveness to honest voter input (higher statistical 
efficiency is better).

I think the trade-off between strategy and performance is also present 
in ranked voting methods. Say you have an election (ballot set) called 
E_a. In E_a, the best choice if all the ballots were honest is to choose 
winner A. However, say another ballot set E_b can be transformed into 
E_a by people that prefer A to B executing some strategy (i.e. lying on 
their ballots), and that similarly, under honesty, the best winner for 
E_b is B.

Then if your creating a method, and your method values honesty, it 
should pick A when given E_a. But if your method is supposed to resist 
the A-voters' strategy in the case of E_b, it should pick B when given 
E_a. You can either have it elect A (in which case you get better honest 
performance) or B (in which case you get better strategy resistance). 
The method itself can't read the voters' minds so it can't know if E_a 
is the real deal or a strategized E_b, so it has to *assume* one way or 
the other.

>> - If everybody strategizes no matter what out of the reasoning that
>> "voting strategically can't harm me so I'll do it even if it probably
>> won't help me either, because there's a probability epsilon> 0 that
>> everybody else will think so too, and then I better get mine", then it
>> really doesn't matter what you'll pick - it'll all go to Approval anyway.
>
> 	Degrading to Approval Voting is not bad.  Approval voting still beats Plurality.

Beating Plurality is a pretty low bar :-)

>> - If more than a majority strategizes, you have no chance of
>> respecting the honest votes, since it's impossible to determine which
>> votes *are* honest.
>
> No system is perfect. We need to try to find a system as resistant
> to  strategy as possible. That is all we can do. This system definitely
> resists strategic burial and strategic exaggeration.

Oh, I wasn't saying your method doesn't resist strategy. I was just 
outlining how far it's possible to go, in general.

>> - If there are lots of strategizer-hedgers (who'll strategize even
>> when it's pretty certain it won't help), but not a majority, MJ deals
>> better with that than does Range.
>
> 	Perhaps.  In mine humble opinion, Majority-Judgement needs more testing.

You might be interested in 
http://hal.archives-ouvertes.fr/docs/00/24/30/76/PDF/2007-12-18-1691.pdf 
, "Election by Majority Judgement: Experimental Evidence", which is 
about just that. It details MJ exit polls in France, as well as what the 
data from these exit polls suggest.