[EM] Rank Codes

[Kristofer Munsterhjelm]

On 14/02/2021 10.20, Juho Laatu wrote:
> Just a thought.
> 
> One simple technique that allows voters to use almost infinite number
> of different ranks would be to use one long box next to the name of
> each candidate. Voters would draw a mark in the box (or leave it
> unmarked). Marks towards the left side of the long box would be
> considered "good", and marks towards the right side of the box "less
> good". The ballots would be read by computers that would seek for a
> mark in each box, and assign a numeric preference value depending on
> the position of the mark in the box. There would be some additional
> perpendicular light lines across the boxes to help making the order
> of the candidates accurate.
I've been thinking about such an interface for an Expensive Pencil too;
it would make it very easy to gather rated information. It would be more
noisy than a deliberate ratings system because it's hard to get a
pixel-perfect line going, but if there are enough voters, then perhaps
there's some rated analog of the Condorcet Jury theorem that would imply
that the effect of that noise goes to zero as the number of voters
approaches infinity.

However, there's one problem: the voters can't easily equal-rank under
such a scheme. Therefore, no method passes FBC unless it also passes
Strong FBC, which is an extremely hard criterion to pass. Pragmatists
may not care (e.g. Schulze has very low favorite-betrayal rates under
the impartial culture distribution anyway), but it's annoying if you'd
like your method to unambiguously pass.

-km