[EM] Majority-Judgement using adjectives versus alphabetical scales versus numerical ranges.

[Kristofer Munsterhjelm]

On 12/09/2012 07:25 PM, Richard Fobes wrote:
> On 12/9/2012 9:12 AM, Jameson Quinn wrote:
>  > ...
>  > 2012/12/8 ⸘Ŭalabio‽ <Walabio at macosx.com <mailto:Walabio at macosx.com>
>  > ...
>  > ¡That is so last week! I wish to find a way to merge
>  > Score-Voting and Majority-Judgement into something even better.
>  >
>  > In order to find something better, we would at least need to know what's
>  > wrong with what we have. There seems to be a lot of disagreement about
>  > that point; smart people can't agree on whether Score or MJ is better.
>  > So I think research is in order before we tear off and design 15 new
>  > systems.
>
> As I see it:
>
> * Advantage of Majority-Judgement (MJ): Makes it (relatively compared to
> score and approval methods) more difficult to vote strategically.
>
> * Advantage of score ballots: Collect the greatest amount of information
> from the voter.
>
> Combining those two advantages could yield a better method. I encourage
> Ŭalabio to explore that possibility.
>
> Yet I would recommend adding yet another advantage, namely the ability
> to fully rank all the choices. As I've said before, credibility for the
> correctness of the most popular choice is undermined if the method
> cannot also identify the second-most popular choice, and so on down to
> the least-popular choice.

You can fully grade every choice in MJ. It is true that this grading 
does not provide a direct relative measure, but Balinski and Laraki 
consider that to be an advantage. They say that if you base your method 
upon comparisons between candidates (like ranking), then the method will 
invariably fail IIA and be subject to Arrow's theorem, but if the method 
is based on evaluation, then it may go beyond the limitations given by 
Arrow's theorem.

It's a subtle distinction, but important to MJ. In MJ, you're not 
supposed to say "do I like this candidate better than that one", but 
rather "how good is this candidate, how good is that candidate, on the 
scale provided to me with a common meaning to everybody voting".

In the experimental evidence paper whose link I gave to Ualabio, B&L 
claim that the data from the exit polls show that the concept of 
"favorite" is largely illusory and disappears under MJ. In particular, 
the voters grade many candidates at the same highest grade on their 
ballot, and the voters don't all make use of the greatest absolute grade 
("Excellent").

If the voters feel evaluating in the manner above is a good approach, 
then I think MJ is a method that works well. If, on the other hand, 
making relative comparisons comes much more naturally to them, then some 
sort of ranked method is the best approach (again in my opinion). In the 
latter case, (if B&L are right,) we have to contend with IIA one way or 
the other, so we might try to find a Pareto-optimal ranked (or 
relatively rated) method.